Plaintext
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
UNIVERSITY OF
COLORADO LAW REVIEW
Volume 90, Issue 2 2019
LISTENERS’ CHOICES
JAMES GRIMMELMANN*
Speech is a matching problem. Speakers choose listeners,
and listeners choose speakers. When their choices conflict,
law often decides who speaks to whom. The pattern is clear:
First Amendment doctrine consistently honors listeners’
choices for speech. When willing and unwilling listeners’
choices conflict, willing listeners win. And when competing
speakers’ choices conflict, listeners’ choices break the tie. This
Essay provides a theoretical framework for analyzing speech
problems in terms of speakers’ and listeners’ choices, an ar-
gument for the centrality of listener choice to any coherent
theory of free speech, and supporting examples from First
Amendment caselaw.
INTRODUCTION
The elevator pitch version of the First Amendment is that
it protects speakers’ choices about what to say.1 The govern-
* Professor of Law, Cornell Tech and Cornell Law School. This is a revised version
of a draft presented at the 26th Ira C. Rothgerber, Jr. Constitutional Law Confer-
ence at the University of Colorado Law School on April 13, 2018. I presented even
earlier versions of this essay at the 2013 Intellectual Property Scholars Confer-
ence, the 2014 Freedom of Expression Scholars Conference, and the Cornell Infor-
mation Science Colloquium in fall 2017. My thanks to the participants there for
their comments, and to Aislinn Black, Jack Balkin, Bryan Choi, Don Gifford,
Mark Graber, Kate Klonick, Lawrence Solum, Alexander Tsesis, Rebecca
Tushnet, and Felix Wu. This essay may be freely reused under the terms of the
Creative Commons Attribution 4.0 International license,
https://creativecommons.org/licenses/by/4.0.
1. E.g., Police Dep’t of Chi. v. Mosley, 408 U.S. 92, 95 (1972) (“To permit the
continued building of our politics and culture, and to assure self-fulfillment for
each individual, our people are guaranteed the right to express any thought, free
from government censorship.”).
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
366 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
ment must be neutral among viewpoints and topics, which
means letting speakers say whatever they want. 2 The central
task of First Amendment theory and doctrine is to explain
when and why deviations from this neutrality principle are jus-
tified. 3 A secondary set of doctrines deal with peripheral ques-
tions about how speakers speak.4
In this Essay, I propose to flip the script. What would First
Amendment theory and doctrine look like if we started from
the how of speech, rather than the what? I will reinterpret
what Thomas Emerson called “the system of freedom of expres-
sion” 5 as a mechanism for matching speakers with listeners. I
will consider how speakers decide to whom to speak, how lis-
teners decide to whom to listen, and how law creates and re-
solves conflicts among speakers and listeners. In so doing, I
will mostly abstract away from the content of speech. This
Essay is concerned with how speakers and listeners find each
other, not what they say to each other once they do. 6
From a matching perspective, listeners’ choices matter
more than speakers’. The paradigm case of free speech involves
a matched pair of a willing speaker and a willing listener. A
consistent commitment to protecting these willing speaker-
listener pairs results in a system of First Amendment law that
regularly defers to listeners’ choices. In cases involving
speaker-speaker conflicts, listeners break the tie. And when
speakers prevail in speaker-listener conflicts, it is often be-
cause other listeners tip the balance.
Part I of this essay briefly describes the concept of match-
ing problems, shows how free speech can be recognized as one,
and highlights a few recurring constraints on possible speaker-
listener matchings. Part II looks closely at different configura-
2. Id. (“But, above all else, the First Amendment means that government
has no power to restrict expression because of its message, its ideas, its subject
matter, or its content.”).
3. See, e.g., EUGENE VOLOKH, THE FIRST AMENDMENT AND RELATED
STATUTES: PROBLEMS, CASES AND POLICY ARGUMENTS (6th ed. 2016) (devoting
two chapters and 336 pages to content-based restrictions on speech).
4. See, e.g., id. (devoting one chapter and 49 pages to content-neutral re-
strictions on speech).
5. THOMAS I. EMERSON, THE SYSTEM OF FREEDOM OF EXPRESSION 1 (1970).
6. Of course, the content of speech gives speakers and listeners their various
preferences among each other; we cannot pretend that all speech is the same. My
point is that important patterns emerge if we look, as much as possible, only at
the structure of the resulting preferences and avoid peeking at the content-based
reasons motivating them.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
2019] LISTENERS’ CHOICES 367
tions of willing and unwilling speakers and listeners to tease
out the matching issues they raise, with particular focus on lis-
teners’ choices among speakers. Part III discusses a few ca-
nonical Supreme Court cases as examples of listener-driven
matching. The Conclusion adds a few thoughts on listeners’
rights and fake news.
I. MATCHING SPEAKERS AND LISTENERS
A. Matching Problems
In mathematics, a “matching” is a way of pairing up the
members of two sets. A “matching problem” is the challenge of
finding a matching that satisfies various constraints. One
familiar example is the “marriage problem”: pair up a set of
male-identifying persons with an equal-size set of female-
identifying persons in a way that maximizes collective happi-
ness. 7 Other familiar matching problems include shoes and
feet, airplane seats and passengers, organ donors and trans-
plant recipients, articles and law reviews, clerks and judges,
and medical students and residency programs. 8 The unifying
feature of these problems is that they involve pairwise interac-
tions between members of two asymmetric groups, where the
individual preferences about with whom to interact are com-
plex, heterogenous, and frequently incompatible.
Matching problems typically have numerous possible solu-
tions, and different algorithms for choosing matchings yield dif-
ferent ones. Mathematicians who study abstract matching
problems and operations researchers who study real-world
matching problems frequently characterize those solutions as
better or worse along various dimensions, and seek algorithms
that yield better ones. Some of their results are striking. For
example, consider the 1962 paper by David Gale and Lloyd
Shapley that essentially kicked off the study of matching
problems. 9 In it they consider an algorithm for the marriage
7. D. Gale & L.S. Shapley, College Admissions and the Stability of Marriage,
69 AM. MATHEMATICAL MONTHLY 9 (1962). This version of the marriage problem
is unrepresentative in that a matching marries off everyone; there are other
matching problems in which not all participants are required to be matched.
8. For a readable general-audience overview of real-life matching problems
and their solutions, see ALVIN E. ROTH, WHO GETS WHAT—AND WHY: THE NEW
ECONOMICS OF MATCHMAKING AND MARKET DESIGN (2015).
9. Gale & Shapley, supra note 7.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
368 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
problem in which each unengaged man proposes to the unen-
gaged woman he most wants to marry, and each woman
accepts the proposal of her favorite current suitor. 10 The men
whose proposals were rejected repeat the process by proposing
to the next woman on their lists; women break their engage-
ments if a better suitor comes along. 11 A little surprisingly, this
frenzied and hard-hearted expression of patriarchy in action is
guaranteed to result in a matching in which everyone gets
married. 12 Even more surprisingly, the resulting matching is
“stable”: there is no pair of disappointed lovers who prefer each
other to their actual spouses. 13 And still more surprisingly (or
perhaps not at all), this algorithm makes men happier than
women: out of all possible stable matchings, it selects the one
that best satisfies men’s preferences. 14 If it is the women who
propose to the men, exactly the reverse is true: the resulting
stable matching is the one that best satisfies women’s prefer-
ences. 15 Thus, the choice of matching algorithm has both
10. Id. at 12.
11. Id.
12. Id. at 12–13. If it didn’t, there would be a woman Weronika who never
accepted a proposal and a man Mateusz who never had a proposal accepted. But
Mateusz must have proposed to Weronika at some point, and she wouldn’t have
turned him down unless she had a proposal in hand from someone else. Since men
never revoke proposals in the Gale-Shapley algorithm, this contradicts the
assumption that Weronika ends up unmarried.
13. Id. at 13. Suppose that Manfred and Wilhelmina would rather be paired
with each other. Either Manfred proposed to Wilhelmina at some point or he
didn’t. If he did, the reason they’re not together is that Wilhelmina must have
rejected his proposal; that is, she ended up paired with someone she preferred to
Manfred. But if Manfred never proposed to her, it means he never worked his way
down his list as far as Wilhelmina; that is, he had a proposal accepted by someone
he preferred to Wilhelmina. Either way, there is a contradiction with the assump-
tion that Manfred and Wilhelmina would rather be paired with each other. There
is no such pair of unmatched lovebirds.
14. DAN GUSFIELD & ROBERT W. IRVING, THE STABLE MARRIAGE PROBLEM:
STRUCTURE AND ALGORITHMS 11 (1989). Intuitively, the reason is that the Gale-
Shapley algorithm goes only just as far down the men’s preference lists as neces-
sary to find a stable matching. When Wilfreda rejects Mortimer’s proposal in favor
of Marston’s, it shows that any matching pairing Mortimer and Wilfreda is unsta-
ble. Wilfreda’s rejection shows that she would rather be paired with Marston than
Mortimer, and Marston’s proposal shows that he prefers Wilfreda to any other
woman who has not already rejected him. No matter whomever else Marston
might be paired with in a matching pairing Mortimer and Wilfreda, Marston and
Wilfreda would rather be paired with each other. Since any attempt to rejigger
the pairings in Mortimer’s favor makes the resulting matching unstable, it means
that the actual matching is the best Mortimer can do out of all possible stable
matchings.
15. Id.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
2019] LISTENERS’ CHOICES 369
allocational and distributional consequences.
The stable marriage problem is a toy mathematical exam-
ple. But real-life matching problems have high stakes. Finding
a good matching algorithm is immensely important. The resi-
dency match is a good example of the practical politics of
matching.16 Medical students “propose” to residencies, so that
the system is designed to favor students’ preferences among
programs over programs’ preferences among students, while
retaining stability. 17 In the 1990s, the algorithm was tweaked
so that student couples could enter the match as a pair and re-
ceive geographically compatible matches if possible, but with-
out favoring one member of the couple over the other. 18 Organ
donations are another example where designing good matching
mechanisms has immense payoffs. 19 Maximizing the number of
compatible donor-donee pairs is literally a matter of life and
death.20
We do not need to go further into the mathematical details
of matching problems here. 21 My points are few and simple.
First, matching problems are everywhere. Second, matching
problems have better and worse solutions. A world with algo-
rithms that find kidney donation chains is better than a world
without those algorithms; fewer people die waiting for a kidney
to become available. Third, “better” and “worse” are often both
relative and political. A world where men propose to women is
better for men; a world where women propose to men is better
for women. Fourth, all real-life matching problems are “solved”
in the sense that there is a matching. At any given time, cer-
tain people are married to certain other people, and others are
16. See generally Alvin E. Roth & Elliott Peranson, The Redesign of the
Matching Market for American Physicians: Some Engineering Aspects of Economic
Design, 89 AM. ECON. REV. 748 (1999).
17. See ROTH, supra note 8, at 133–44.
18. Id. at 144–49. See generally Roth & Peranson, supra note 16.
19. See, e.g., Alvin E. Roth et al., Kidney Exchange, 119 Q.J. ECON. 457 (2004).
20. See Itai Ashlagi et al., Nonsimultaneous Chains and Dominos in Kidney-
Paired Donation—Revisited, 11 AM. J. TRANSPLANTATION 984 (2011).
21. For those who are interested, two elegant and illuminating treatments of
matching algorithms can be found in JON KLEINBERG & ÉVA TARDOS, ALGORITHM
DESIGN (2006) and DONALD E. KNUTH, STABLE MARRIAGE AND ITS RELATION TO
OTHER COMBINATORIAL PROBLEMS: AN INTRODUCTION TO THE MATHEMATICAL
ANALYSIS OF ALGORITHMS (Martin Goldstein trans., 1989). An exhaustive treat-
ment of the stable marriage problem and its many extensions is GUSFIELD &
IRVING, supra note 14.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
370 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
unmarried. That is a matching.22 Fifth, there are many differ-
ent matching mechanisms. Some, like school assignments in
many districts, are explicit: they are the output of purpose-
built algorithms. Some, like the placement of feet in shoes, are
implicit: they rely on uncoordinated social processes. And sixth,
mechanism choice bears on both the quality and the fairness of
the resulting matching. As the persistence of divorce shows, the
real-life marriage matching is neither optimal nor stable.
Matching matters.
B. Speech as a Matching Problem
Speech is a matching problem. Speakers speak; listeners
listen. In each case, the question is to whom? I might at any
given moment be listening to a politician’s speech, an adver-
tisement for dish soap, a friend’s endless nattering about base-
ball, or a subway rapper’s freestyling exploits. The politician
might, at any given moment, be speaking to a national audi-
ence on C-SPAN, a potential donor on the telephone, a col-
league, a sibling, a reporter, or a therapist, to name just a few
more.
Some numbers may illustrate the scale of the problem.
Suppose that, at any given time, one billion of Earth’s seven
billion people are speaking in some form, and another billion
are listening. That gives each listener a billion different choices
of where to direct her attention. 23 But that is just one listener.
The number of ways to find a speaker for every listener is
vastly larger. Each such matching is a list a billion lines long
(the number of listeners); each line in such a list has one of a
billion possible values (the number of speakers). 24 The number
22. It is not a matching that satisfies the constraints of the stable marriage
problem (which requires that everyone be paired off). It is a matching that satis-
fies the constraints of the more general problem in which people’s rank-ordered
preferences can include “unmarried” as well as potential partners.
23. See generally JAMES GLEICK, THE INFORMATION: A HISTORY, A THEORY, A
FLOOD (2011). Information overload is hardly a new problem. See generally ANN
M. BLAIR, TOO MUCH TO KNOW: MANAGING SCHOLARLY INFORMATION BEFORE
THE MODERN AGE (2010); ALEX WRIGHT, GLUT: MASTERING INFORMATION
THROUGH THE AGES (2007).
24. Here is a simplified illustration of the problem. Suppose there are two
listeners, Alice and Bob, and three speakers, Xu, Yaz, and Zelda. Then there are
nine possible matchings:
• Alice listens to Xu, and Bob listens to Xu.
• Alice listens to Xu, and Bob listens to Yaz.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
2019] LISTENERS’ CHOICES 371
of possible matchings is thus one billion to the billionth power,
or 109,000,000,000. That is a one followed by nine billion zeroes, a
number that defies human comprehension. And yet, every
minute of every day, speakers and listeners settle on one of
these different matchings.
Different matchings serve the goals of information policy
differently. Consider a matching in which the Qin Emperor
speaks and everyone on earth listens, a matching in which
people talk only to their closest neighbor, and a matching in
which each speaker has a randomly chosen pen pal. Each of
these matchings is badly deficient. The first is dictatorial, the
second fragmented, the third chaotic. We should hope to do bet-
ter. Not matching is not an option; one way or another, the
choice will be made. The alternative to matching speakers with
listeners is no speech.
A few factors are particularly important in determining
which of the 109,000,000,000 or so matchings we end up with. Most
obviously, speakers and listeners constantly make choices.
When Spike comes up to Liz at a party and starts shouting in
her ear about the Federal Reserve, he is a speaker selecting her
as a listener. When Liz flees to the den and turns on the
Orioles game instead, she is a listener selecting different
speakers: the game’s announcers. Sometimes, third parties—
such as the state or media intermediaries—make these choices
for listeners. When a flash-flood warning starts scrolling across
the bottom of the screen, the government and the TV station
intervene to redirect Liz’s attention away from the announcers’
speech and to the Emergency Warning System’s speech in-
stead.
Although every matching is a product of billions of human
choices, these choices are not unbounded. They take place
within an extensive framework of geographic, linguistic, eco-
nomic, cultural, and technological constraints. Spike lives in
• Alice listens to Xu, and Bob listens to Zelda.
• Alice listens to Yaz, and Bob listens to Xu.
• Alice listens to Yaz, and Bob listens to Yaz.
• Alice listens to Yaz, and Bob listens to Zelda.
• Alice listens to Zelda, and Bob listens to Xu.
• Alice listens to Zelda, and Bob listens to Yaz.
• Alice listens to Zelda, and Bob listens to Zelda.
If there are n listeners and m speakers, then there are mn distinct matchings that
assign each listener a speaker. Here, n=2 because there are two listeners, and
m=3 because there are three speakers, for a total of 32=9 possible assignments.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
372 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
Missouri, not Malaysia. He doesn’t speak Malay. He doesn’t
have the money for a plane ticket. And, to be honest, not that
many people, here or in Kuala Lumpur, are interested in his
conspiracy theories about the Federal Reserve. Past choices by
speakers and listeners, by governments, and by third parties,
play out in present constraints. Spike’s decision to major in
Spanish affects whom he can effectively speak to or listen to. So
do the signing of the Anglo-Dutch Treaty of 1824—and the
countless other past events that we collectively refer to as “his-
tory.” Speakers and listeners make choices, but they make
them against a backdrop of choices already made, choices that
make certain matchings infeasible.
Although there are good reasons to try to honor the
matching choices made by speakers and listeners, it would be
inadvisable to honor every such choice. Indeed, it would be
impossible. Three intrinsic limits are important.
First, people’s choices about speech, like people’s choices
about everything, almost always fall short of the theoretical
ideal of fully informed and fully rational decision-making.
Some things we do end up being unintentional choices about
speech, as when George W. Bush referred to a New York Times
reporter as a “major league asshole” in front of a microphone he
didn’t realize was on, or when a tenant rents an apartment
near a nightclub that’s far louder than she expects. Other times
we make conscious choices to speak or to listen that we later
regret, as when I saw the Brian De Palma/Nicolas Cage turkey
Snake Eyes on opening weekend. I call these internal limits on
choices about speech: people make mistakes when they choose
whom to speak to or listen to. Put another way, “choice” itself is
a constructed category. Sometimes we regard people’s actions
as sincere expressions of their preferences about speech when
they are nothing of the sort, and vice versa.
Second, some combinations of choices are physically or
logically impossible. These are structural limits on how many
choices it is possible to honor at once. If I want every person on
Earth to stop what they’re doing and hear me simultaneously
as I issue ransom demands from my evil lair deep within the
sun, it is simply not going to happen. If two aging arena-
rockers both want to play to a sold-out crowd at Madison
Square Garden on the same evening, there is no way to make
both of them happy. If I want to tell you about trademark law
and you would rather watch Game of Thrones, one of us is
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
2019] LISTENERS’ CHOICES 373
going to be disappointed. Physical, geographical, and infra-
structural factors create structural limits on choice. So do
disagreements among speakers and listeners.
Third, choice is not the only important value. As a society
we put external limits on choice by trading it off against other
important values. These values are familiar: they are the
harms to society and to third parties that weigh against speak-
ers’ rights to say whatever they want. A bomb-bearing listener
and a speaker explaining where inside a train station to deto-
nate a bomb to cause maximum carnage might choose to com-
municate with each other. But there is a strong social interest
in minimizing carnage; we have good reasons to interfere with
their choices here.
C. Scarcities
Internal, structural, and external limits are general fea-
tures of speaker-listener matching. They play out differently in
different contexts. Three specific constraints on speaker and
listener choices are common and important enough to require
separate discussion. They are speakers’ limited capacity to get
their speech to listeners (bandwidth), listeners’ limited capac-
ity to listen to multiple speakers at once (attention), and listen-
ers’ limited information about what speakers will say in the
future (ignorance). Bandwidth and attention are structural
limits; ignorance is an internal limit.
1. Bandwidth
Basic characteristics of the media that people use to
speak—capacity, range, audience, and cost—shape who speaks
to whom.25 Compare a fiber-optic network with a network of tin
cans connected by string. A major fiber-optic cable can transmit
many thousands of high-resolution cat videos per hour; a pair
of tin cans can handle one low-fidelity voice conversation.
Fiber-optic networks can reach billions of people while the tin
cans let you reach one at a time. Fiber-optic networks span the
globe; the tin cans will barely get you from here to next door.
25. See generally MARSHALL T. POE, A HISTORY OF COMMUNICATIONS: MEDIA
AND SOCIETY FROM THE EVOLUTION OF SPEECH TO THE INTERNET (2011) (discuss-
ing media in terms of accessibility, privacy, fidelity, volume, velocity, range,
persistence, and searchability, and from both speaker and listeners’ perspective).
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
374 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
But while almost anyone can build and deploy a local-area
point-to-point personal tin-can network, transcontinental cable
laying has always been reserved for the few, the regulated, the
well-capitalized.
Pre-modern media sharply limited the set of speaker-
listener matchings. They had low capacities, variable but typi-
cally limited ranges, small audiences, and moderate to low
costs. An unamplified public speaker can be heard by a few
hundred people who must be gathered in the same place as the
speaker; a handwritten letter can travel the globe but can only
be read by a handful of people at once. Some kinds of match-
ings are simply incompatible with these limits; a listener circa
220 A.D. in Rome would not have been physically capable of
hearing a speaker in Xi’an.
The pre-Internet mass media that dominated the twen-
tieth century—particularly, television, radio, and mass-
produced artifacts like newspapers, books, and records—were
different. They had large capacity, large audiences, national
range, and high costs.26 This made bandwidth a crucial bottle-
neck between speakers and listeners, and one with a charac-
teristic structure. A limited set of speakers could make use of
these media, but those with access could then reach very large
audiences. These media allowed a few to speakers reach large
numbers of listeners. They did not let large numbers of speak-
ers each reach smaller but scattered audiences.
Contrast the characteristic twenty-first century medium:
the Internet, an astoundingly high-capacity network that spans
the globe, has billions of users, and enables famously “cheap
speech.” 27 There are disparities among speakers, to be sure, of
wealth, class, power, nationality, gender, language, education,
and many others. Not everyone can push out a two-hour, high-
definition video to a million viewers precisely at midnight—but
almost anyone can get a book’s worth of text to a few hundred
or few thousand readers within minutes, or 280 characters to
millions of readers within seconds. Generally speaking, anyone
can speak to a worldwide audience of any size for the cost of a
basic smartphone: in some places as low as $20.28
26. Id. at 101–51 (discussing print), 152–201 (discussing audiovisual media).
27. See Eugene Volokh, Cheap Speech and What It Will Do, 104 YALE L.J.
1805 (1995).
28. See Jamie Carter, The Land of the $20 Smartphone, TECHRADAR (Mar. 11,
2017), https://www.techradar.com/news/the-land-of-the-20-smartphone [perma.cc/
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
2019] LISTENERS’ CHOICES 375
These shifts in media also shift who holds the power to
choose among possible speaker-listener matchings. Wherever
there is a speech-transmission bottleneck, there is also a
speech-selection bottleneck. Almost all the speech-selection
power in mass pre-Internet media rests with those who choose
which speech will be “pushed” through those media. Speakers
either have access to the medium or do not; listeners’ choices
are limited to ordering from the speech menu placed before
them, and the menu is not long. In such an environment, the
capacity constraints both force an upstream choice among
speakers and make that choice inherently controversial.
Speakers, government, conduit operators, and various interest
groups will all have claims about which speakers should be
given access. Listeners’ choices may provide rhetorical support
for certain ways of deciding among speakers, and some of those
ways will attempt to ascertain what listeners in general would
prefer, but individual listeners will not be able to exercise
choice as listeners over which speakers receive access.
Bandwidth is different on the Internet because transmis-
sion capacity no longer creates a chokepoint that forces an
early upstream choice of a sharply restricted set of speakers.
Speaker-listener matchings are no longer determined in the
first instance by access to the mass media; listeners no longer
experience the sharply restricted set of choices among speech
characteristic of the mass media. Provided they know it exists
and where it is to be found, listeners can “pull” the speech they
want from its source. This can shift some control over speaker-
listener matchings downstream, closer to listeners. 29 (This
shift, we will see, tends to advance free speech goals. 30)
4BYP-2V9Y].
29. Interestingly, it is generally older work, from the first flourishing of legal
scholarship about the Internet, that best articulates the close connection between
an abundance of speech and listener choice. See Eugene Volokh, Freedom of
Speech in Cyberspace from the Listener’s Perspective: Private Speech Restrictions,
Libel, State Action, Harassment, and Sex, 1996 U. CHI. LEGAL F. 377; Volokh,
supra note 27; Jerry Berman & Daniel J. Weitzner, Abundance and User Control:
Renewing the Democratic Heart of the First Amendment in the Age of Interactive
Media, 104 YALE L.J. 1619 (1995).
30. See infra Part II.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
376 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
2. Attention
There is an important asymmetry between speakers and
listeners. Speech, being information, is infinitely replicable.
Nothing in the nature of speech itself prevents a speaker’s
message from reaching the entire world, given sufficient time,
effort, and expense. A speaker might lower her voice for rea-
sons of privacy or discretion, but she does not have to. Speak-
ing to one listener does little to detract from her ability to reach
a second.
The vis vitae of listening, however, is not speech but atten-
tion, and human attention is always and everywhere limited.
Listeners consume speech nonrivalrously, but speech consumes
listeners’ attention rivalrously. 31 If I am reading a book, I will
have a much harder time also paying attention to the news on
the radio. Unlike speakers, listeners must choose. Listeners
have an overwhelming array of choices in the modern media
environment. But they still only have two eyes, two ears, and
one brain.
Limited attention particularly matters now because there
are billions of speakers in the world, and millions of them are
trying to reach a public audience. And since media bandwidth
no longer stands in the way, almost all of their speech is actual-
ly available to interested listeners. In this sense, listeners are
inundated with choices. There are millions of speakers who are
ready, willing, and able to address them. But it also means that
these speakers are competing fiercely with each other for audi-
ences. A previously buried constraint on speech—listeners’
limited attention—is now closer to the surface. 32
Attention functions differently than mass-media scarcities.
For one thing, it is distributed: each listener controls only a
tiny fraction of the world’s supply. For another, it can be taken
31. A good is rivalrous if one consumer’s use prevents others from using it. It
is nonrivalrous if any number of consumers can use it at no additional cost.
32. Herbert Simon, Designing Organizations for an Information-Rich World,
in COMPUTERS, COMMUNICATIONS, AND THE PUBLIC INTEREST 37, 40–41 (M.
Greenberger ed., 1971) (“Hence a wealth of information creates a poverty of atten-
tion and a need to allocate that attention efficiently among the overabundance of
information sources that might consume it.”); Tim Wu, Is the First Amendment
Obsolete?, in KNIGHT FIRST AMENDMENT INSTITUTE, EMERGING THREATS 7 (David
Pozen ed., 2017), https://knightcolumbia.org/content/tim-wu-first-amendment-obsolete
[perma.cc/8AB3-FVAK] (“[I]t is no longer speech or information that is scarce, but
the attention of listeners.”).
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
2019] LISTENERS’ CHOICES 377
involuntarily: just try to read on the subway when there is a
crying baby or an apocalyptic preacher standing next to you.
For a third, it can be attacked indirectly. As Zeynep Tufekci
powerfully argues, censorship today is best understood as
denial of attention: a government can thwart an opposition
movement either by keeping the media from reporting on pro-
tests (traditional “censorship”) or by swamping the media with
distractions, disinformation, and noise, so that few people
watch the news reports or believe them.33 Keeping people from
knowing about speech can be just as effective as keeping the
speech itself from them, and so can keeping them too distracted
to pay attention to it.34
3. Ignorance
There is another important asymmetry between speakers
and listeners. Because speakers produce information and
listeners consume it, listeners always choose from a position of
partial ignorance in a way that speakers do not. Speech, being
information, is subject to Arrow’s information paradox.35 A lis-
tener who has not yet heard speech is not capable of making a
fully informed evaluation of it. 36 Once she has heard the
speech, she is in a better position to assess whether it is worth
listening to—but by then it is too late. The speech cannot be
unheard.
To repeat, this is a matter of asymmetry between speakers
and listeners. Speakers have excellent knowledge of the speech
they are about to engage in. Listeners have no such advantage:
if someone catches your eye on the sidewalk, you don’t know
33. ZEYNEP TUFEKCI, TWITTER AND TEAR GAS: THE POWER AND FRAGILITY OF
NETWORKED PROTEST 239 (2017).
34. See MARGARET E. ROBERTS, CENSORED: DISTRACTION AND DIVERSION
INSIDE CHINA’S GREAT FIREWALL 41–92 (2018) (adding “friction” and “flooding” to
“fear” as mechanisms of censorship).
35. Kenneth J. Arrow, Economic Welfare and the Allocation of Resources for
Invention, in THE RATE AND DIRECTION OF ECONOMIC ACTIVITY: ECONOMIC AND
SOCIAL FACTORS 609, 615 (1962) (“[T]here is a fundamental paradox in the deter-
mination of demand for information; its value for the purchaser is not known until
he has the information, but then he has in effect acquired it without cost.”).
36. This is also a version of the learner’s paradox from Plato’s Meno. “How
will you look for it, Socrates, when you do not know at all what it is? How will you
aim to search for something you do not know at all?” PLATO, PLATO: FIVE
DIALOGUES: EUTHYPHRO, APOLOGY, CRITO, MENO, PHAEDO 70 (John M. Cooper
ed., G.M.A. Grube trans., 2d ed. 2002).
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
378 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
what they want. Maybe they want to rant at you about the gold
standard, or tell you about an art gallery opening, or ask for
directions to the train station, or get your signature on a politi-
cal petition. You can guess, but you can’t know. Only they can.
One common manifestation of listeners’ limited
knowledge—or at least one commonly asserted in the free
speech literature—is a preference for familiar speech over un-
familiar. Some listeners are like six-year-olds demanding to
watch The Little Mermaid for the eighty-ninth time: they know
what they like and see no need to venture beyond it. Others are
like Sam I Am turning up his nose at green eggs and ham; they
are unwilling to give speech a proper hearing, even if in hind-
sight they would have been glad to have heard it.
Another manifestation of listeners’ ignorance is that
speakers are sometimes in a position to deceive listeners about
their speech. Clickbait is a simple example. This Headline
Promises Shocking Revelations. The Article Will Underwhelm
You. False speech also plays on listeners’ ignorance. A fully in-
formed listener would not typically choose speech that is
fraudulent or defamatory; more often, listeners to actionably
false speech are listening to it only because the speaker has
successfully hidden its falsity from them.
And a third way in which listeners’ ignorance plays out is
that simply finding the speech they want can be a massive
challenge in an age of informational abundance. 37 Easing the
transmission bottleneck makes the underlying selection prob-
lem salient in a new way. It was easy to see what was on TV
when there were only three channels. Now, not so much. The
characteristic media companies of the Internet age are selec-
tion intermediaries: search engines, portals, advertising net-
works, social networks, news aggregators, recommendation en-
gines, marketing analytics providers, and the many others who
specialize in pairing up speakers and listeners. 38 These inter-
mediaries have a significant informational advantage over
speakers and listeners, and thus a significant power to shape
37. Speakers face a version of this problem: they don’t know which potential
listeners in the world would be most receptive to their speech. But while speakers
are ignorant about listeners, listeners are ignorant about the speech itself.
38. Some of these intermediaries are also transmission intermediaries: they
make speech available or deliver it to listeners, in addition to identifying speech of
interest to listeners or listeners of interest to speakers. The roles are conceptually
distinct, even when a single entity performs both.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
2019] LISTENERS’ CHOICES 379
speaker-listener matchings by making different suggestions.
They vary not just in how much of this power they retain for
themselves, but also in whether they tilt more toward listeners
or speakers in their matchmaking. Search engines are highly
listener directed: different queries let users seek out different
speech. 39 Advertising engines that let marketers slice and dice
their target eyeballs along thousands of demographic and be-
havioral axes are highly speaker-directed. 40 Facebook’s
algorithmic News Feed is somewhere in between: it nudges us-
ers into reading friends’ posts in a default, Facebook-selected
order while giving users some controls to tweak that order and
the ability to speak directly to or listen directly to specific other
users they choose. 41
II. THE CENTRALITY OF LISTENER CHOICE
So much for what we can do about speaker and listener
choices. Now for what we should do. Or perhaps, I should say,
what we must do. This Part makes a simple argument: any
theory of free speech that does not take listener choice seri-
ously fails as a theory of free speech. It is not necessary to
delve deeply into the normative justifications for free speech.
Whatever those justifications are, they must protect the entire
communicative pathway from willing speaker to willing lis-
tener. And once they do, they cannot effectively protect speak-
ers who have willing listeners unless they also frequently
protect unwilling listeners from unwanted speakers. This con-
clusion follows from the structural constraints on speaker and
listener choice imposed by other speakers and other listeners.
It is inherent in the nature of speech in a world with many
people in it.
39. See James Grimmelmann, Speech Engines, 98 MINN. L. REV. 868, 894,
898–900 (2014) (arguing that good listener-directed selection is normatively desir-
able).
40. See generally JOSEPH TUROW, BREAKING UP AMERICA: ADVERTISERS AND
THE NEW MEDIA WORLD (1997) (giving history of targeted advertising); JOSEPH
TUROW, THE DAILY YOU: HOW THE NEW ADVERTISING INDUSTRY IS DEFINING
YOUR IDENTITY AND YOUR WORTH (2011) (continuing history into Internet era).
41. See generally TARLETON GILLESPIE, CUSTODIANS OF THE INTERNET:
PLATFORMS, CONTENT MODERATION, AND THE HIDDEN DECISIONS THAT SHAPE
SOCIAL MEDIA (2018) (describing in detail how Facebook and other social media
platforms comprehensively filter, sort, and structure the content that flows be-
tween users).
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
380 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
This is a content-neutral analysis. It genuinely does not
care what the speech at issue is. It does not matter what good
it does for speakers, listeners, or society. It does not matter
what harms it is capable of. I will use speeches about the
mayor as an example, but absolutely nothing will turn on the
specifics. The only things we need to know about the speech is
which listeners the speakers are trying to reach and which
speakers the listeners would like to hear. We do not need to
look inside their choices or inquire after their reasons; it mat-
ters only that the speakers and listeners have reasons they
consider sufficient.
A. Willing Listeners
The standard justifications of free speech take as their
paradigm case a willing speaker facing a willing listener. 42 We
can depict this case graphically, and with a simple hypothet-
ical:
Core Free Speech: S gives a speech criticizing the
mayor. L is in the audience.
This is the paradigm case for free speech. S is a willing
speaker, L a willing listener. They form a matched pair: S and
L both want L to hear what S has to say. In such a setting, it is
possible to be imprecise about whose interests are at stake be-
cause it does not really matter. Speakers and listeners pull the
same oar. Treating the relevant interests as belonging purely
to speakers, purely to listeners, or jointly to speakers and lis-
teners makes no significant difference.
To emphasize the point, consider the different ways that
42. When I refer to a “willing” speaker, I mean one who affirmatively wants to
be heard by a specific listener or listeners; when I refer to a “willing” listener, I
mean one who affirmatively wants to hear a specific speaker. “Willing” and
“unwilling” are relational terms. There are cases of pure self-expression in which
a speaker doesn’t care who listens. And there are rare cases of compelled speech
in which a speaker just wants to be silent and the government doesn’t care about
any particular listener. Set these both aside for now.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
2019] LISTENERS’ CHOICES 381
the authorities could interfere with the speaker-listener rela-
tionship between S and L.
Arrested Speaker: S gives a speech criticizing the
mayor. L is in the audience. The police arrest S.
Air Horns: S gives a speech criticizing the mayor. L is
in the audience. The police stand in the room blowing
air horns.
Arrested Listener: S gives a speech criticizing the
mayor. L is in the audience. The police arrest L.
In Arrested Speaker, the police have prevented S from
speaking. This is an obvious and obviously unconstitutional re-
striction on speech. The net effect is that L is unable to hear
what S has to say. This is censorship, plain and simple.
In Air Horns, the police have not literally prevented S from
speaking. But the air horns are just as effective as an arrest in
figuratively silencing S. This too is censorship. The First
Amendment cares about more than just S’s ability to vibrate
her vocal cords or to scribble on a piece of paper. S’s freedom of
speech will be poorly protected indeed if the government can
interdict her speech before it reaches L. 43 There is nothing
43. See Procunier v. Martinez, 416 U.S. 396, 408 (1974) (“Communication by
letter is not accomplished by the act of writing words on paper. Rather, it is effect-
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
382 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
wrong in general with police use of air horns: they are useful,
for example, to warn citizens of serious danger from a building
demolition. Rather, the problem in Air Horns is that the police
officer has used the air horn to thwart an act of communication
involving both a speaker and a listener. The idea of a listener is
inherent in speech and, hence, is also inherent in free speech.
Although Arrested Listener is a little less common, it too is
a restriction on speech. Just as in Air Horns, the police have
not literally prevented S from speaking. But arresting the audi-
ence has the same effect as arresting the speaker or blowing air
horns because, again, L is unable to hear what S has to say. If
Arrested Speaker is a paradigm case of a restraint on speech,
then so are Air Horns and Arrested Listener. What matters is
that the state has broken the communicative pathway from
speaker to listener, not where along the pathway it has acted.
The right to receive speech is the “reciprocal” of the right to
speak. 44 Each is meaningless without the other. A system that
does not protect the freedom to listen will not effectively protect
the freedom to speak, and vice versa.
This may sound like an expansion of the domain of free
speech. In a sense it is. The protected communicative pathway
extends farther, until it reaches a listener. But in another im-
portant sense it is a contraction. We are not concerned with
speech as speech (involving only a speaker), but with speech as
communication (involving both a speaker and a listener). On
the former, core free speech protections attach when someone
speaks; interference with the success of the speaker’s project is
suspect. On the latter, core free speech protections attach when
someone speaks and someone listens; interference with the
success of their joint communicative project is suspect. This is a
narrower principle because it requires both a speaker and a lis-
tener. We can test this principle’s reach with a pair of cases:
ed only when the letter is read by the addressee.”), overruled by Thornburgh v.
Abbott, 490 U.S. 401 (1989).
44. Va. State Bd. of Pharmacy v. Va. Citizens Consumer Council, Inc., 425
U.S. 748, 757 (1976) (“If there is a right to advertise, there is a reciprocal right to
receive the advertising . . . .”).
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
2019] LISTENERS’ CHOICES 383
Lonely Speaker: S gives a speech criticizing the mayor
in a forest with no one else around until the police
arrive to arrest S.
Lonely Listener: L stands in a forest listening to the
wind with no one else around until the police arrive to
arrest L.
We might regard these as free speech cases. Or we might
not. A theory focused on speakers’ and listeners’ personal liber-
ty might regard these arrests as problematic; a theory focused
on democratic discourse might not. It is possible to articulate
theories of free speech that do and do not protect speakers
without listeners and listeners without speakers, a strong indi-
cation that these cases are not of the same importance to free
speech as cases with both. Arrested Speaker, Air Horns, and
Arrested Listener are core cases for free speech. Lonely Speaker
and Lonely Listener are not.
B. Unwilling Listeners
Unwilling-listener cases may not seem central to free
speech, but they are. Their ubiquity makes them invisible.
Most lack the high drama of the Westboro Baptist Church’s
funeral protests. 45 But that is because our system sorts them
out almost automatically. Most unwanted speakers never even
get close to their targets. Screening out vast oceans of unwant-
ed speech is an essential feature of any system that makes free
speech possible. Getting the paradigm willing-listener case
right depends, as a practical matter, on getting immense num-
bers of unwilling-listener cases right as well. And if we do not,
45. See, e.g., Snyder v. Phelps, 562 U.S. 443 (2011).
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
384 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
it is not just listeners but speakers who suffer. 46
In willing-listener cases like Air Horns, the state had an
option—do nothing—that would satisfy both S and L. But if L
is an unwilling listener, the state has no such option. Consider:
Bored Audience: S gives a speech criticizing the mayor.
L, who is within hearing range, would like to get up
and leave.
S and L have a structural conflict because, no matter what
the state does, there is no matching that will make everyone
happy. If the state leaves matters alone (or sends a police
officer to silence S), L will be happy and S will not. But if the
state intervenes by stationing a police officer to keep L in her
seat, S will be happy and L will not.47 We need some principled
46. Unwilling speakers also raise interesting issues, and there are both
significant parallels and significant differences between compelled speech and
compelled listening. See generally Caroline Mala Corbin, The First Amendment
Right Against Compelled Listening, 89 B.U. L. REV. 939 (2009) (comparing com-
pelled speech and compelled listening). But teasing out the differences in relation
to speaker-listener matching is a matter for another occasion.
47. Readers who prefer to break the tie in favor of state passivity should
consider another hypothetical: Loud Speakers:
S sets up an amplifier and speakers on the sidewalk in front of L’s house
and gives a speech at 95 decibels. L would prefer not to listen. Loud
Speakers has the same structure of speaker and listener choices as Bored
Audience: a willing speaker facing an unwilling listener. The difference
is that in Bored Audience the listener has an effective self-help technique
to win the speaker-listener conflict, whereas in Loud Speakers it is the
speaker with the effective self-help. Thus in Loud Speakers, if the state
does nothing, it satisfies S but not L. If it intervenes to make S go away
or unplug the speakers, it satisfies L but not S. It is easily possible to
distinguish Loud Speakers, see generally Kovacs v. Cooper, 336 U.S. 77
(1949) (upholding municipal ordinance prohibiting the use of amplified
sound trucks), but not on the basis of a general rule about state-versus-
private action. Whatever line we draw will have the effect of determining
how people can and cannot speak and when they must or need not listen
to others’ speech, and will have to be normatively justified in view of
those consequences. Appealing to property law (e.g., that S speaks from
a public sidewalk or that S creates a nuisance) merely begs the question,
as the contours of property law are themselves contestable in terms of
the balance they strike between S and L.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
2019] LISTENERS’ CHOICES 385
basis on which to decide whose choices will be respected.
It is common to assert that in such cases the Supreme
Court has chosen to favor speakers over listeners. For example,
in Snyder v. Phelps, Chief Justice Roberts reiterated that “the
Constitution does not permit the government to decide which
types of otherwise protected speech are sufficiently offensive to
require protection for the unwilling listener or viewer.” 48 This
view is not so much wrong as incomplete. There are three pat-
terns of unwilling-listener cases, and they raise different kinds
of issues. The simplest involve one-to-one speech, pitting a
speaker who wants to be heard against a listener who does not
want to hear.49 Other cases involve one-to-many speech, in
which a speaker addresses a large and possibly indeterminate
audience, not all of whom are interested in what she has to say.
Finally, some cases involve many-to-one speech, in which mul-
tiple speakers compete for the attention of a listener. If this
part of First Amendment law seems confused and contradic-
tory, it may stem from a failure to distinguish among these
patterns.
The existence of one-to-many and many-to-one cases com-
plicates the story. Paradigm cases of unwilling listeners involv-
ing one-to-one speech appear to pose a choice between favoring
speakers’ attempts to be heard and favoring unwilling listen-
ers’ attempts not to hear. But in one-to-many cases, there is
also a listener-listener conflict: some would like the speaker to
continue, while others would like her to shut up. Whatever the
state does will frustrate some listeners. And in many-to-one
cases, there is also a speaker-speaker conflict: each speaker
would like to prevail over the others. Whatever the state does
will frustrate some speakers.
There is no general speaker-favoring or listener-favoring
solution to unwilling-listener cases. Some speakers and some
listeners will inevitably be disappointed. One-to-many cases
show that favoring unwilling listeners over speakers can frus-
trate other listeners, while many-to-one cases show that favor-
ing speakers over unwilling listeners can frustrate other
48. Snyder, 562 U.S. at 459 (quoting Erznoznik v. City of Jacksonville, 422
U.S. 205, 210–11 (1975)).
49. The helpful terminology to distinguish “one-to-one” speech from “one-to-
many” speech is taken from Eugene Volokh, One-to-One Speech vs. One-to-Many
Speech, Criminal Harassment Laws, and “Cyberstalking,” 107 NW. U. L. REV. 731
(2013). I have added the term “many-to-one.”
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
386 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
speakers. But paradoxically, these intractable tensions help us
understand what is truly at stake even in one-to-one cases.
1. One-to-Many
Speaker-listener conflicts are rarely just about a speaker
and a listener. Often, other listeners may be interested in the
outcome. Consider:
Controversial Protest: S gives a speech criticizing the
mayor and would like to reach as many listeners as
possible. L1 and L2 are within hearing range. L1 is
not interested in listening to S; L2 is.
Controversial Protest has Bored Audience embedded within
it because L1 is an unwilling listener. S would prefer to speak
to her while L1 would prefer not to be spoken to. Just as in
Bored Audience, it is not possible to make both S and L1 happy.
Silencing S favors L1 over S, but allowing S to continue favors
S over L1. Again, we must decide, as between S and L1, whose
choices to respect.
But we also have another decision to make, because L1
and L2’s choices are in tension as well. If S continues, her
speech will reach both L1 and L2, satisfying L2 but frustrating
L1. If S shuts up, her speech will reach no one, satisfying L1
but frustrating L2. The two listeners’ choices about S cannot
both be honored. One of them will go home unhappy. 50 We
50. Changing the physical structure of the problem changes its conceptual
structure, as well. If L1 can leave or wear headphones, or if S and L2 can
communicate in sign, the conflict may be substantially defused. See infra Section
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
2019] LISTENERS’ CHOICES 387
cannot make the decision between them simply by appealing to
a principle of satisfying all listeners’ choices, or as many as
possible. If we intervene in the name of listeners’ choices to
satisfy L1, we also harm listeners’ choices by frustrating L2,
and vice-versa. Controversial Protest poses both a speaker-
listener and a listener-listener conflict.
These two decisions are linked: S and L2 are united
against L1. If S speaks, both S and L2 are satisfied. If S does
not, both are frustrated. This observation holds the key to the
hypothetical, because there is something distinctive and special
about S and L2 in free speech terms. They are a willing
speaker addressing a willing listener—the easiest case for
protected speech. Put another way, Controversial Protest also
has Core Free Speech embedded within it. Thus, if S is silenced
in Controversial Protest, the state interferes with speech just as
much as it does in Arrested Speaker, 51 just with a different pro-
posed justification. If S is allowed to speak, the state frustrates
L1’s desire not to listen, but this is not the paradigm case of
preventing desired speech from reaching its audience.
Thus, any coherent free speech principle will tend to favor
willing listeners over unwilling ones. If L1’s and L2’s choices as
listeners are inextricably bound up with and opposed to each
other, then the pro-speech outcome comes closer to the core
willing-listener ideal for free speech than the anti-speech out-
come. Listener choices for speech trump listener choices
against speech when the two conflict. Moreover, the difference
between Bored Audience—in which S’s claims as a speaker
against the unwilling L were ambiguous—and Controversial
Protest—in which S’s claims as a speaker against the unwilling
L1 are unambiguous—is precisely the presence of the addi-
tional, willing listener L2. Without L2, this is a hard case; with
L2, it becomes an easy one.
It follows that we should be careful not to mistake one-to-
many for one-to-one cases. Sometimes—as with harassing tele-
phone calls—there really is only a single relevant listener. But
at other times—as with funeral protests—there may be other
listeners. The presence or absence of other listeners can be a
reason to treat these cases differently.
II.C.
51. Or in Air Horns or Arrested Listener.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
388 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
2. Many-to-One
Just as other listeners may have an interest in the out-
come of a speaker-listener conflict, so too may other speakers.
Consider:
Dueling Speeches: S1 is giving a speech criticizing the
mayor. In the room next door, S2 is giving a speech
praising the mayor. L would like to attend S2’s speech.
S1 and S2 are both speakers. They would both like to
speak to L. But those goals are incompatible; L can only attend
one of the two speeches.
S1 and S2, however, do not have equal claims on L’s atten-
tion. As against S2, L is a willing listener. Thus, Dueling
Speeches has Core Free Speech embedded within it. The police
officer who compels L to attend S1’s speech rather than S2’s in
Dueling Speeches has interfered with S2’s ability to communi-
cate with L just as much as the police officer who arrests L in
Arrested Listener. The same considerations that led us to say
the state interferes with S’s freedom of speech in Arrested
Listener 52 should lead us to say that it also interferes with S2’s
freedom of speech by dragging L from S2’s audience into S1’s.
The effect on S2’s speech is the same.
S1 has a harder time making a similar claim. As against
S1, L is an unwilling listener, because L would rather be listen-
ing to S2. (This is Bored Audience again, embedded within
Dueling Speeches.) If the state assists L in escaping from S1’s
auditorium, S1 may be frustrated, but this is the lesser harm of
52. Or in Air Horns or Arrested Speaker.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
2019] LISTENERS’ CHOICES 389
taking sides in the tug-of-war between speaker and listener,
not the greater harm of standing between a willing speaker
and a willing listener. The two conflicts in Dueling Speeches—
S1 versus S2 as speakers competing for the same audience, and
S1 versus L as a speaker trying to reach an unwilling
listener—are bound up with each other. L is simultaneously an
unwilling listener (to S1) and a willing listener (to S2).
Once again, the asymmetry here comes entirely from a
single source: L’s choice as a listener selecting among speakers.
On any theory of free speech that cares about listeners’ choices
in their own right, this is an easy tie to break. L chooses S2,
not S1. But even a theory that purports to care only about
speakers’ choices will still tend to prefer S2 to S1, because S2
and L stand in the paradigm free speech relationship of willing
speaker and willing listener, whereas S1 and L do not. To
choose between S1 and S2 without taking L’s choices into
account, a speaker-regarding theory would need to appeal to
the personal characteristics of S1 or S2 or to the respective
value of pro-mayoral and anti-mayoral speech. Neither is an
attractive starting point for a theory of free speech, because it
is usually thought that any viable theory of free speech ought
to be largely neutral as to speakers and largely neutral as to
viewpoints.
I would say that we should be careful not to mistake many-
to-one cases for one-to-one cases, but that is not quite right, be-
cause every unwilling-listener case has something of a many-to-
one flavor to it. There are so many would-be speakers in the
world that an unwilling listener is probably being deprived of
the chance to listen to someone else. Thus, there is an asym-
metry at work. Zoom out from Bored Audience and you may or
may not find additional listeners, but you will almost always
find additional speakers. You may or may not be looking at
Controversial Protest, but you are almost certainly looking at
Dueling Speeches.
Indeed, in a world with inexpensive, worldwide, high-
bandwidth, digital media, there are always millions of speakers
trying to reach every listener in the world. If they had their
way, everyone would listen to them. Even setting aside these
speech megalomaniacs, every listener with Internet access is
always and everywhere faced with a choice among perhaps a
billion possible speakers:
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
390 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
The only way that a listener can attend to any one pre-
ferred speaker is by tuning the other billion out. Or, to put a
sharper point on it, the only way any speaker can ever be heard
by anyone is by beating out the other billion. It is a minor mir-
acle that speech is possible at all in the modern world. It re-
quires a combination of technologies, norms, and legal support
to ensure that all the other speakers back off enough that a lis-
tener can hear the one speaker as to whom she is a willing
listener. And yet, that is what the twenty-first century system
of freedom of expression accomplishes, billions of times a day.
Every successful communication from a willing speaker to a
willing listener is a triumph of listener choice.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
2019] LISTENERS’ CHOICES 391
3. One-to-One
In light of these one-to-many and many-to-one cases, the
one-to-one cases like Bored Audience should appear in a new
light. What looks like a one-to-one case may not be: it may be a
one-to-many case or a many-to-one case in disguise. Both
Controversial Protest and Dueling Speeches contain Bored
Audience, which means that even a case matching the Bored
Audience pattern of a willing speaker facing an unwilling
listener may not only be about the conflict between them. It
may match a larger pattern as well, and if it does, we should
zoom out until we see the other listeners straining to hear and
the other speakers straining to be heard.
When there are other willing listeners, their choices as lis-
teners help us break the Bored Audience speaker-listener
deadlock in favor of the speaker, who has a broader and willing
audience. And when there are other speakers, this time the lis-
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
392 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
tener’s choice helps break the Bored Audience deadlock in favor
of the listener, who is choosing among speakers, not merely
choosing whether to listen. Narrowly, these considerations cut
in opposite directions: additional listeners give the speaker’s
desire to speak more weight, while additional speakers give the
objecting listener’s objections more weight. But in a broader
sense, both of these considerations are appeals to listener
choice.
It follows, albeit tentatively, that there are plausible free
speech reasons to promote listener choice even in one-to-one
unwilling-listener cases. In Bored Audience and similar cases,
this means taking the side of the listener rather than the
speaker. In every other type of case discussed above—willing
listeners, many-to-one unwilling listeners, and one-to-many
unwilling listeners—the basic commitments of free speech re-
quired us to adopt a principle of favoring listeners’ choices. To
be sure, we are not required by those commitments to extend
the listener-choice principle to one-to-one unwilling-listener
cases, but neither are we prohibited from doing so. Siding
either with listeners or with speakers—or deciding between
them on a case-by-case basis—is consistent with those commit-
ments. But given how uniformly the listener-choice principle
applies in every other type of case, it is plausible and attractive
in one-to-one unwilling listener cases as well.
Thus, a listener-choice principle treats one-to-one and one-
to-many cases quite differently: the unwilling listener in a one-
to-one case can have her choices respected, while the unwilling
listener in a one-to-many case will have to put up with the
unwanted speech. The reason is not that the unwilling listen-
er’s choices themselves are more or less significant in one type
of case versus the other, but rather that in a one-to-one case a
listener’s choices affect no other listeners, while in a one-to-
many case the listeners’ choices are unavoidably intertwined. A
street-corner orator necessarily reaches multiple listeners,
whereas a telephone harasser does not bring other listeners
into the picture. His speech is limited to one highly unwilling
listener. Serving him with a no-contact order will not interdict
any speech to a willing listener (because there are no other
listeners), but it will facilitate the speech of other speakers (be-
cause his victim can go back to reading a novel).
This is not a conclusive argument for listener choice in all
settings. The details depend on contextual factors and norma-
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
2019] LISTENERS’ CHOICES 393
tive arguments. Different foundational theories of free speech
will make very different prescriptions here. My point is nar-
rower. We end up favoring listeners’ choices in many one-to-one
cases without even considering contextual factors and norma-
tive arguments.
C. Targeting, Selection, and Separation Costs
Speakers and listeners don’t just have preferences about
speech, they also act on those preferences. The speaker who
chases a listener down the block and the listener who runs
away are engaged in a struggle over whether they will end up
matched. If one or the other stands still, she concedes to the
other. 53
But if speakers’ and listeners’ actions can create structural
conflicts, they can also resolve those conflicts. Consider
Controversial Protest again:
Controversial Protest (redux): S gives a speech criti-
cizing the mayor and would like to reach as many
listeners as possible. L1 and L2 are within hearing
range. L1 is not interested in listening to S; L2 is.
53. The possibility of these self-help arms races is another reason that it is
hard to state categorical rules about one-to-one cases. Compare wearing earplugs
versus speaking through a megaphone, running away versus chasing after, and
filtering ads versus evading ad filters: in each case the balance of power between
listeners and speakers is different. There may be good reasons for the state to step
in and set limits on how far the arms race can go—effectively picking a winner
just to have it done with—but those reasons depend heavily on the details of who
can do what; that is, on what I will shortly call “separation costs.”
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
394 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
S and L2 form a willing-listener/willing-speaker pair,
whereas S and L1 have irreconcilably opposed preferences. The
point of Controversial Protest is that there is an additional
listener-listener conflict between L1 and L2. The state’s choice
of whether or not to silence S can resolve that conflict only by
disappointing one of the two. Given the core commitment to
protecting speakers addressing willing listeners, it follows that
S should be allowed to speak.
Unlike the conflict between S and L1, which is inherent in
their preferences, the conflict between L1 and L2 is contingent.
It stems from the assumption that either S speaks to both L1
and L2 or S speaks to neither of them. If we relax this assump-
tion, it should be clear that another matching is possible: S
speaks to L2 but not to L1. 54
S speaks to L1 & L2 S speaks to L2 S speaks to no one
S good okay bad
L1 bad good good
L2 good good bad
In this matching, both L1 and L2 go home happy because
their respective preferences (not to listen and to listen) have
been honored. If we are committed to listener choice, this is a
distinct improvement over the previously available options of
speaking to both or to neither. It is still not possible to make
both S and L1 happy. Their preferences conflict no matter
what. But again, if we are committed to listener choice, we will
be less bothered by disappointing S than by disappointing L1.
When L1 and L2 participate in a one-to-many case with S,
their opposed preferences interfere; when they participate in
distinct one-to-one cases with S, their preferences are inde-
pendent. Every listener does as well or better when a one-to-
many case is transformed into an aggregation of independent
one-to-one cases.
54. Of course, there is also the matching in which S speaks to L1 but not to
L2, but that would just be perverse. L1 and L2 would gladly trade places and S
would not object, so this is not a stable matching.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
2019] LISTENERS’ CHOICES 395
That leaves open the question of how to get there from
here. This is where speakers’ and listeners’ actions come in.
Suppose they are in a park, a few dozen feet apart. Any of the
three of them could go a long way toward solving the problem
by walking a short way. S could come closer to L2, L2 could
come closer to S, or L1 could go further away from the others.
No matter who gets up and moves, the end result is the same:
S can speak in a way that is audible to L2 but not to L1. In
addition to these unilateral measures, there may be hybrids
that combine actions from two, or even all three. So S and L2
could each walk halfway toward the other. Or S could post a
flyer announcing his intention to speak against the mayor at
noon by the duck pond. L2, who is interested, will go to the
duck pond at noon. L1, who is not interested, will stay away.
Frequently, speakers and listeners voluntarily collaborate
in this sorting process without any state intervention. Consider
the description of a book on a dust jacket, which lets readers
decide whether they want to read the book. The resulting
sorting into interested readers and disinterested non-readers
suits both speakers (authors and publishers) and listeners
(readers). The ubiquity of these collaborative choice structures
should not blind us to their importance in achieving good
matchings.
However it takes place, sorting is not free. Call the costs
involved “separation costs.” In the park example, there is a lit-
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
396 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
eral separation: someone must move to create a greater dis-
tance between S and L1 than between S and L2. More often,
separation is a metaphor. In the book example, the publisher
pays someone to draft the dust jacket copy and pays to print
the dust jackets. Readers pay with their time when they skim
the dust jacket and reflect. These separation costs are willingly
borne, but they are costs nonetheless.
Sometimes separation costs may be cheap; sometimes they
may be prohibitively expensive. Some separation costs will fall
on speakers to target their speech, some on willing listeners to
opt themselves in, and some on unwilling listeners to opt them-
selves out. Some separation costs may even fall on third parties
or the government (perhaps the parks department maintains
the bulletin board S used to post the flyer). The parties will
quite naturally prefer that someone else do the necessary work
of separating willing from unwilling listeners. 55 To decide
whether we should expect or require someone to bear separa-
tion costs, we will have to consider their magnitude and inci-
dence and compare these variables to the costs of foregone or
unwanted speech.
We should be asking, in a sense, who is the least-cost
avoider of unwanted speech? The answer to that question will
always depend on contextual details and frequently on the con-
tent of the speech. But it is possible to make some general
observations. A few distinctions are significant:
First, there is a duality between making speakers respon-
sible for targeting their speech only to willing listeners and
making listeners responsible for selecting only the speech they
wish to hear. An email marketing list is speaker-targeted; an
inbox spam filter is listener-selected. Asking listeners to “avert
their eyes” or make the “short, though regular, journey from
mail box to trash can” 56 requires listener selection; asking tele-
marketers not to call numbers on the Do Not Call list requires
speaker targeting.57
55. Of course, S may prefer that the problem not be solved at all: he has a bet-
ter case to address the unwilling L1 if that is the only way he can reach the will-
ing L2. Not every speaker wants to reach unwilling listeners; musicians want to
reach fans, not make enemies. But even those speakers who are perfectly willing
not to address unwilling listeners would still prefer not to bear the separation
costs.
56. Lamont v. Comm’r of Motor Vehicles, 269 F. Supp. 880, 883 (S.D.N.Y.
1967).
57. See 16 C.F.R. § 310.4(b)(1)(iii)(B) (2018).
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
2019] LISTENERS’ CHOICES 397
Second, there is a similar duality between making willing
listeners opt in and making unwilling listeners opt out. While
both are species of listener selection, they differ in terms of
whether willing or unwilling listeners are more responsible for
taking action. Do Not Call is opt-out. Those who do not wish to
be called must register with the list. But cable television chan-
nels are opt-in. Those who wish to receive them must affirma-
tively subscribe.
Third, separation costs depend both on the cost of physi-
cally acting and on the cost of acquiring the necessary
knowledge on which to act. It is easy to throw junk mail away
unopened, but harder to tell whether there is junk mail within
an unmarked envelope. Contrariwise, it is easy to know that
there is a DJ stage at the block party outside, but not so easy to
ignore it.
In general, the lower a party’s separation costs, the more
often First Amendment doctrine asks that party to bear them.
The argument that unwilling viewers should avert their eyes
reflects in part a belief that it would be more costly to make
willing viewers wear special glasses. But the captive audience
doctrine deals with situations in which unwilling listeners are
not effectively able to avert their eyes. We do not ask people to
wear earplugs and blindfolds inside their own homes.
The law sometimes asks parties to collaborate in lowering
each other’s separation costs. Amplification limits require
speakers not to use technology that overwhelms listeners’ abil-
ity to self-select. 58 Do Not Call requires listeners to volunteer a
little information that is helpful to accurate speaker targeting.
Caller ID requires speakers to provide a little information that
is helpful to accurate listener selection.
In many real-life settings, the actual system in use is a
complex hybrid of these various separation techniques. Take
spam. Our current hybrid system for sorting email expects
speakers, willing listeners, and unwilling listeners all to play
their part. 59 Unwilling listeners who wish not to receive
commercial promotions from a given sender are expected to
make an opt-out request. Senders, however, are required to
58. See Ward v. Rock Against Racism, 491 U.S. 781, 803 (1989) (holding that
“[t]he city’s sound-amplification guideline . . . is valid under the First Amendment
as a reasonable regulation of the place and manner of expression”).
59. See, e.g., Controlling the Assault of Non-Solicited Pornography and
Marketing Act of 2003, 15 U.S.C. §§ 7701–7713 (2012); 16 C.F.R. pt. 316 (2018).
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
398 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
make those opt-outs simple and convenient and to honor such
requests. 60 To facilitate low-cost selection by unwilling recipi-
ents deciding which emails to read, senders are required not to
use certain kinds of deceptive metadata in their emails.61
Finally, willing recipients who do want a company’s mailings
are frequently expected to affirmatively opt in to promotional
emails. 62 Notice how these devices are all reasonably low-cost,
and how none of them interfere directly with speech to a will-
ing listener. They collectively combine to create a system that
facilitates the separation of willing and unwilling recipients. It
is hardly a perfect system, but from a free speech perspective,
it is far from a disaster either. It does a reasonable job at sepa-
rating willing from unwilling listeners, and thus it facilitates
speaker and listener choice.
III. EXAMPLES
The previous Part made a series of structural claims about
speaker-listener matching. Despite working at an absurdly
high level of abstraction, it showed that any coherent system of
free speech will tend to protect mutually willing speakers and
listeners, 63 favor willing listeners over unwilling listeners, 64 fa-
vor a willing listener’s preferred speaker over competing
speakers, 65 consider protecting unwilling listeners from un-
wanted speakers when other listeners are not implicated, 66 and
try to help willing and unwilling listeners separate them-
selves. 67
There is a tentative listener-choice principle running
through all of these points. Listeners’ choices to hear speech
provide a prima facie reason to permit it. Listeners’ choices not
to hear speech provide a prima facie reason to prevent it, but
one that can be overcome by other listeners’ choices to hear it
when the two conflict. That conflict, in turn, can be defused
when it is possible to disaggregate willing and unwilling listen-
60. See 15 U.S.C. § 7704(a)(3)(A)(i) (2012) (making it unlawful for a person to
send a commercial email that does not contain a functioning opt-out).
61. Id. § 7704(a)(1).
62. Id. § 7704(a)(4)(B).
63. Supra Section II.A.
64. Supra Section II.B.1.
65. Supra Section II.B.2.
66. Supra Section II.B.3.
67. Supra Section II.C.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
2019] LISTENERS’ CHOICES 399
ers’ choices.
To repeat, these principles were derived solely from
abstract theory and with no reference to the contents of speech
or listeners’ reasons for their choices. Yet they still perform
surprisingly well when exposed to actual cases involving actual
speech, speakers, and listeners. This Part examines a few
canonical Supreme Court cases and finds them consistent with
these general principles.
A. Mail
Start with a matched pair of cases involving the United
States mail. In Lamont v. Postmaster General, the Court struck
down a statute interdicting “communist political propa-
ganda.” 68 In Rowan v. Post Office Department, the Court up-
held a statute allowing householders to prohibit delivery of
“erotically arousing or sexually provocative” matter. 69 The rele-
vant difference between the two was listener choice.
Lamont dealt with section 305(a) of the Postal Service and
Federal Employees Salary Act of 1962, which required the
Postal Service to detain “communist political propaganda” from
certain countries mailed into the United States and deliver it
“only on the addressee’s request.” 70 The Postal Service imple-
mented the statute by screening all mail from those countries
and sending a reply card to the addressee of any piece of mail
determined to be statutory agitprop. 71 If the recipient filled out
the card and returned it to the Postal Service, it would then
deliver the mail; but if the card was not returned within twenty
days, the mail would be destroyed. 72 The two cases decided in
Lamont involved the detention of material that today seems
almost absurdly tame, such as the public exchange of stilted
and grandiose open letters between the Chinese and Soviet
Communist Parties as they jostled over the direction of inter-
68. 381 U.S. 301, 307 (1965).
69. 397 U.S. 728, 730 (1970).
70. Lamont, 381 U.S. at 302. A separate statute, the Foreign Agents
Registration Act of 1938, defined the term “political propaganda,” and section
305(a) then defined “communist political propaganda” as political propaganda “is-
sued by or on behalf of” certain specified countries. Id. at 302 n.1 (citation omit-
ted).
71. Id. at 303–04.
72. Id.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
400 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
national communism.73 In both cases, the American recipients
filed suit challenging the constitutionality of section 305 rather
than return the reply card.74
The Supreme Court avoided standing difficulties by invok-
ing listeners’ rights, rather than speakers’ rights. 75 The Court
held that the statute “amounts in our judgment to an unconsti-
tutional abridgment of the addressee’s First Amendment
rights.” 76 Moreover, it justified this conclusion by detailing the
system’s burdens for listeners. The Court noted the “affirma-
tive obligation” it thrust on addressees and the chilling effect of
needing to request delivery of “communist political propa-
ganda.” 77 In a concurrence, Justice Brennan explicitly and
eloquently grounded willing listeners’ rights in the First
Amendment:
It is true that the First Amendment contains no specific
guarantee of access to publications. However, the protection
of the Bill of Rights goes beyond the specific guarantees to
protect from congressional abridgment those equally funda-
mental personal rights necessary to make the express
guarantees fully meaningful. I think the right to receive
publications is such a fundamental right. The dissemination
of ideas can accomplish nothing if otherwise willing address-
ees are not free to receive and consider them. It would be a
barren marketplace of ideas that had only sellers and no
buyers. 78
73. See, e.g., Chinese and Soviet Parties Exchange Letters, PEKING REV., Mar.
22, 1963, at 1, 6, https://www.marxists.org/subject/china/peking-review/1963/PR1963
-12.pdf [https://perma.cc/5HNQ-JK4K].
74. Lamont, 381 U.S. at 304–05.
75. The senders were located abroad and the mail they sent was issued on be-
half of foreign governments. Thus, not only were the senders not before the court,
it was unclear that they had any First Amendment rights to assert. See id. at
307–08 (Brennan, J., concurring). The modern First Amendment status of speech
by foreigners remains unsettled, although “aliens abroad are presumed not to en-
joy First Amendment rights.” Timothy Zick, Territoriality and the First
Amendment: Free Speech at—and Beyond—Our Borders, 85 NOTRE DAME L. REV.
1543, 1549 (2010); see also Bluman v. Fed. Election Comm’n, 800 F. Supp. 2d 281
(D.D.C. 2011), aff’d, 132 S. Ct. 1087 (2012).
76. Lamont, 381 U.S. at 307 (emphasis added).
77. Id.
78. Id. at 308 (Brennan, J., concurring) (citations omitted).
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
2019] LISTENERS’ CHOICES 401
Lamont uncontroversially elevated listeners to the same
plane as speakers. In similar cases where willing speakers and
willing listeners oppose a government seeking to interpose it-
self between them, it does not particularly matter who is the
plaintiff. Allowing either to sue when their interests are “inex-
tricably meshed” avoids difficult and distracting questions
about standing to assert each other’s First Amendment
rights. 79
Rowan took a further and more interesting step. The
Postal Revenue and Federal Salary Act of 1967 established a
procedure for householders to notify the Postmaster General
that they had received “erotically arousing or sexually provoca-
tive” advertising material from a specified sender. 80 Upon
receiving such a notice, the Postmaster General was required
to order the sender “to refrain from further mailings to the
named addressee.”81 Notably, whether the mail was in fact
arousing or provocative was left to the “sole discretion” of the
householder, and as the statute was construed by the Court,
future mailings from that sender to that addressee were
prohibited regardless of their content. 82
In many respects, the statute in Rowan was the more
speech-restrictive of the two. It barred speech outright, rather
than merely imposing an inconvenience (the reply card) on it. It
could be applied to any advertising mail, not just mail advo-
cating on behalf of communist governments. It applied to
United States senders, not just aliens abroad. It imposed the
threat of coercive punishments against speakers, not just the
interception of their speech. And it vested a private party with
unfettered and unreviewable discretion to apply a vague statu-
tory standard. From a speaker-centric perspective, the statute
in Rowan is more offensive, and with Lamont on the books, it
79. See, e.g., Procunier v. Martinez, 416 U.S. 396, 409 (1974), overruled by
Thornburgh v. Abbott, 490 U.S. 401, 413–14 (1989). Procunier involved prisoners’
correspondence, so by focusing on the rights of their pen pals both as senders and
as recipients, the Court could sidestep the question of “the extent to which an
individual’s right to free speech survives incarceration.” Id. at 408; see also
Kleindienst v. Mandel, 408 U.S. 753 (1972) (recognizing First Amendment inter-
ests of would-be audiences for a Belgian socialist denied visa to enter the United
States, while subordinating those interests to the plenary Congressional power to
exclude aliens).
80. Rowan v. Post Office Dep’t, 397 U.S. 728, 729–30 (1970).
81. Id. at 730.
82. Id. at 734–35.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
402 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
seems like an easy case for invalidation.
But when the Supreme Court decided Rowan, it unani-
mously upheld the statute.83 Chief Justice Burger’s opinion
rests on listeners’ rights as listeners. It starts by acknowledg-
ing the tension between speakers’ and listeners’ interests:
But the right of every person “to be let alone” must be
placed in the scales with the right of others to communicate.
. . . To make the householder the exclusive and final
judge of what will cross his threshold undoubtedly has the
effect of impeding the flow of ideas, information, and argu-
ments that, ideally, he should receive and consider. 84
As between the two, it clearly favors listeners:
Nothing in the Constitution compels us to listen to or view
any unwanted communication, whatever its merit; we see
no basis for according the printed word or pictures a differ-
ent or more preferred status because they are sent by
mail. 85
And it finishes with a clear statement of unwilling listen-
ers’ place in the First Amendment’s scheme:
We therefore categorically reject the argument that a ven-
dor has a right under the Constitution or otherwise to send
unwanted material into the home of another. If this prohibi-
tion operates to impede the flow of even valid ideas, the
answer is that no one has a right to press even “good” ideas
on an unwilling recipient. 86
83. Id. at 740.
84. Id. at 736. I leave for another day a detailed analysis of the role that prop-
erty and privacy interests play in unwilling-listener cases where (as in Lamont
and Rowan) the unwanted speech is received in the home. See, e.g., Lamont, 381
U.S. at 304–05; Rowan, 397 U.S. at 729, 738. In brief, real property rights provide
a framework for speech separation through literal, physical separation. “Privacy”
here is the right to be let alone rather than informational privacy—so appeals to
privacy merely restate a listener’s interest in avoiding unwanted speech, rather
than providing an independent ground to recognize it.
85. Rowan, 397 U.S. at 737.
86. Id. at 738.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
2019] LISTENERS’ CHOICES 403
The difference between Lamont and Rowan is the willing-
ness or the unwillingness of the listener. Lamont is a simple
violation of the core willing-speaker, willing-listener free
speech principle. Rowan comes out as it does because mail is a
one-to-one medium. The separation problem was substantially
solved by giving each residence its own postal address. 87 Be-
cause mail is individually targetable, it is reasonable to ask
senders to refrain from mailing unwilling recipients. To be
sure, it might also be reasonable to ask recipients to throw
away unwanted mail. But by the logic of listener choice, in a
one-to-one case, no core violation of free speech is committed if
we ask senders rather than recipients to bear this cost.
Moreover, note that the statute in Rowan required unwill-
ing recipients to opt out by sending a reply card, rather than
requiring willing recipients to opt in. By giving senders one bite
at the apple, the opt-out rule tends to favor willing listeners
over unwilling ones, achieving a more speech-protective result.
It also respects listener choice while imposing a threshold
condition to ensure that the listener makes a more informed
choice. 88 And a burden that is reasonable for unwilling listen-
ers may not be reasonable for willing ones. The same proce-
dure—sending a reply card—was struck down in Lamont as an
unconstitutional burden on speech.
B. Doorbells
The same distinction appears in the Supreme Court’s cases
on door-to-door solicitation. Consider Martin v. City of
Struthers. 89 Struthers, Ohio, had an ordinance prohibiting
door-to-door distribution of “handbills, circulars or other adver-
tisements.” 90 A Jehovah’s Witness argued that the ordinance
87. Substantially, but not completely. More than one person can live at the
same address, and they may not all be equally willing or unwilling. The implicit
assumption that a single head of household speaks for all listeners at the address
is not always true. The statute in Rowan allowed parents to add the names of
children nineteen and under to the removal lists; Justice Brennan’s concurrence
raised the possibility that teenagers might be more willing recipients than their
parents. Rowan, 397 U.S. at 741 (Brennan, J., concurring).
88This is another aspect of the problem in Lamont: postal recipients were re-
quired to make decisions about mail from senders they hadn’t yet received any
mail from; that is, to make uninformed choices about speech.
89. 319 U.S. 141 (1943).
90. Id. at 142.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
404 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
was unconstitutional after being fined $10 for violating it. 91
Justice Black’s opinion striking down the ordinance under
the First Amendment is notable for the contrast it draws be-
tween the government’s decisions and the homeowner’s. In one
notable passage, it makes the same point four times in three
sentences:
We are faced in the instant case with the necessity of
weighing the conflicting interests of the appellant in the
civil rights she claims, as well as [(1)] the right of the indi-
vidual householder to determine whether he is willing to re-
ceive her message, against the interest of the community
which by this ordinance offers to protect the interests of all
of its citizens, [(2)] whether particular citizens want that
protection or not. The ordinance does not control anything
but the distribution of literature, and in that respect it [(3)]
substitutes the judgment of the community for the judgment
of the individual householder. It submits the distributer to
criminal punishment for annoying the person on whom he
calls, [(4)] even though the recipient of the literature distrib-
uted is in fact glad to receive it. 92
Crucially, the opinion explains (albeit in dictum) that laws
punishing trespass by unwanted callers remain constitutional:
This or any similar regulation leaves the decision as to
whether distributers of literature may lawfully call at a
home where it belongs—with the homeowner himself. A city
can punish those who call at a home in defiance of the pre-
viously expressed will of the occupant . . . .93
Subsequent case law involving unwanted newspaper deliv-
eries confirms that homeowners can indeed stop unwanted
speech by giving proper notice. 94
91. Id.
92. Id. at 143–44 (emphasis added).
93. Id. at 148.
94. See, e.g., Tillman v. Distribution Sys. of Am., 224 A.D.2d 79, 88 (N.Y. App.
Div. 1996) (“The State need not, and in our opinion, should not, compel anyone to
read, to buy, or even to touch, pick up, or handle a newspaper of which the indi-
vidual in question wants to have no part.”); cf. Reddy v. Plain Dealer Publ’g Co.,
991 N.E.2d 1158 (Ohio Ct. App. 2013) (denying relief where homeowner had not
provided newspaper with notice of his objection). Different considerations apply,
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
2019] LISTENERS’ CHOICES 405
As before, knocking on doors is a one-to-one medium, so it
is possible both to protect speakers and to ask them to target
their speech only to the willing. And also as before, the compro-
mise on the ground requires the unwilling to opt out rather
than requiring the willing to opt in. The state may enforce a
homeowner’s desire not to be spoken to, but it may not presume
such a desire, even where the presumption is rebuttable. This
rule effectively gives speakers the chance to engage listeners to
see whether they choose to hear more. Some, indeed many, of
those listeners will not, and these cases arise because listeners
object to having been bothered. But these ultimately unwilling
listeners cannot prevail over those who prove willing to enter-
tain the speaker’s message. The protection of unwilling listen-
ers’ choices both depends on, and is limited by, the protection of
willing listeners’ choices.
C. Drive-Ins
Finally, consider Erznoznik v. City of Jacksonville. 95 The
University Drive-In Theatre in Jacksonville screened Class of
‘74, which featured “pictures of uncovered female breasts and
buttocks.”96 A city ordinance prohibited showing such anatomy
“if such motion picture, slide, or other exhibit is visible from
any public street or public place.” 97 The drive-in’s screen was
visible from two public streets and a church parking lot. A
prosecution and a declaratory judgment constitutional chal-
lenge ensued, and the Supreme Court eventually found the
ordinance unconstitutional. 98
Of Jacksonville’s various asserted justifications for the
ordinance, the only one that need detain us here is the theory
that the city could “protect its citizens against unwilling expo-
sure to materials that may be offensive.”99 The opinion itself
presents the issue as a clash between speakers and listeners,
describing the clash as one “pitting the First Amendment
rights of speakers against the privacy rights of those who may
of course, beyond the home, because there the interests of other listeners besides
those with property-based exclusionary rights are implicated.
95. 422 U.S. 205 (1975).
96. Id. at 207 n.1.
97. Id. at 207.
98. Id. at 206, 217–18.
99. Id. at 208. As a measure to protect children, the ordinance was overbroad.
Id. at 212–14. As a traffic control measure, it was underinclusive. Id. at 214–15.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
406 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
be unwilling viewers or auditors” and calling for “delicate bal-
ancing.” 100 But outside the home, Justice Powell drew a distinc-
tion:
Much that we encounter offends our esthetic, if not our
political and moral, sensibilities. Nevertheless, the Consti-
tution does not permit government to decide which types of
otherwise protected speech are sufficiently offensive to re-
quire protection for the unwilling listener or viewer. Rather,
absent the narrow circumstances described above, the bur-
den normally falls upon the viewer to “avoid further bom-
bardment of [his] sensibilities simply by averting [his]
eyes.”101
Since “the offended viewer readily can avert his eyes” by
looking away from the drive-in screen, the Court allowed the
drive-in to continue showing racy B movies.
This much is standard-issue, unwilling-listener rhetoric.
But the real story of Erznoznik—the story of willing listeners—
comes in the footnotes. First, the drive-in was not screening
Class of ‘74 to shock unsuspecting passersby. It was trying to
reach willing viewers (also known as paying customers) rather
than unwilling ones. As footnote 6 observed, “[p]resumably,
where economically feasible, the screen of a drive-in theater
will be shielded from those who do not pay.”102 Thus, the ordi-
nance did not simply burden the drive-in as a speaker, it also
burdened the drive-in’s customers as listeners. Footnote 7
acknowledged that the case involved both “the rights of those
who operate drive-in theaters and the public that attends these
establishments.”103 That makes Erznoznik a one-to-many case.
To prohibit showing Class of ‘74 to protect the choices of un-
willing viewers would interfere with the choices of willing
ones. 104
100. Id. at 208.
101. Id. at 210–11 (quoting Cohen v. California, 403 U.S. 15, 21 (1971)).
102. Id. at 211 n.6.
103. Id. at 212 n.7.
104. Indeed, there is more than a hint in the case that the problem from
Jacksonville’s point of view was not unwilling viewers on public streets but will-
ing ones. Justice Powell’s recitation of the facts states, “[t]here was also testimony
indicating that people had been observed watching films while sitting outside the
theater in parked cars and in the grass.” Id. at 207. One does not take a seat near
a drive-in to avoid the movie; one takes a seat to enjoy it.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
2019] LISTENERS’ CHOICES 407
An exchange between the dissent and the majority makes
clear that Erznoznik was a case about separation costs. Chief
Justice Burger argued in dissent that it was difficult for disin-
terested viewers to look away, saying, “[T]he screen of a drive-
in movie theater is a unique type of eye-catching display that
can be highly intrusive and distracting.” 105 But even crediting
the discredited idea that visual media compel obedience,
screening out the movies would have been far more difficult on
the drive-in’s end. By one estimate, it might have cost $250,000
to erect a sufficient wall 106—a cost so high as to seriously inter-
fere with the drive-in’s willingness to speak (and thus its abil-
ity to reach willing listeners). As footnote 7 of the majority
opinion explained, “The effect of the Jacksonville ordinance is
to increase the cost of showing films containing nudity. In cer-
tain circumstances theaters will avoid showing these movies
rather than incur the additional costs. As a result persons who
want to see such films at drive-ins will be unable to do so.” 107
As between the drive-in and passersby, the latter were the
least-cost avoiders of the speech conflict here.
This point deserves amplification. When we deal with un-
wanted one-to-many speech, we are always asking at least two
questions. The first is a question of separation costs: whether
we are truly dealing with a one-to-many case or simply with
the aggregation of numerous but independent one-to-one cases.
Only once we have an answer to this question about the
dynamics of the situation can we properly consider the speech
itself and the question of its value or harm to speakers and
listeners. Cases like Erznoznik that make sweeping statements
about what listeners must endure may in fact stand only for
much narrower propositions about what they must endure
when targeting is infeasible. Mail and knocking on doors are
targetable, drive-in theaters much less so. These kinds of cases
raise different issues, and it is not possible to lump all unwill-
ing-listener cases together. We must be more careful about the
actual structure of the flows of speech from speakers to listen-
ers and about whose choices influence those flows. Only then
can we properly articulate whose interests are truly at stake,
and what conflicts the law must mediate.
105. Id. at 222 (Burger, C.J., dissenting).
106. Id. at 213 n.8.
107. Id. at 212 n.7.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
408 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
CONCLUSION
Speaker-listener matching is a general, rigorous, and illu-
minating way of analyzing difficult First Amendment prob-
lems. It provides a new and useful way of understanding how
speaking and listening are different but interdependent, it
identifies the relevant relationships in First Amendment cases.
It does not always apply, but when it does, it helps bring out
recurring patterns.
One such pattern, which I have discussed in detail, is lis-
tener choice. This is a golden age for scholarship on listeners’
rights. 108 Recent work includes powerful normative arguments
for putting listeners alongside speakers at the center of free
speech theory and thoughtful listener-oriented analyses of a
diverse array of First Amendment doctrines. Those who are
committed to listeners’ rights may find the matching frame-
work congenial for a few reasons. First, it provides a concrete
way of recognizing listeners’ agency, because it focuses on their
choices rather than on their interests. 109 Second, it shows that
listeners make choices for speakers and not just against them,
which helps align listeners’ and speakers’ rights. Third, it
shows how some conflicts among speakers and listeners can be
accidents of history, so that with appropriate physical, tech-
nical, or legal interventions, the conflicts dissolve. And fourth,
it shows that listener choice really is a recurring pattern:
• In one-to-many cases, willing listeners’ choices prevail over
unwilling listeners’ choices.
• In many-to-one cases, listeners’ choices break the deadlock
among competing speakers.
108. The literature on listeners’ rights is extensive, and a proper survey would
require an article-length literature review. In addition to the other articles in this
issue, significant recent highlights include BURT NEUBORNE , MADISON’S MUSIC:
ON READING THE FIRST AMENDMENT (2015); Leslie Kendrick, Are Speech Rights
for Speakers?, 103 VA. L. REV. 1767 (2007); Toni M. Massaro & Helen Norton,
SIRI-OUSLY 2.0: What Artificial Intelligence Reveals About the First Amendment,
101 MINN. L. REV. 2481 (2017); Seana Valentine Shiffrin, A Thinker-Based
Approach to Freedom of Speech, 27 CONST. COMMENT. 283 (2011).
109. And speakers’ choices, but speakers have not suffered from lack of atten-
tion to their choices.
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
2019] LISTENERS’ CHOICES 409
• In one-to-one cases (once they are properly distinguished
from one-to-many cases) unwilling listeners’ choices are less
of a threat to free speech than they are sometimes said to
be.
• Some one-to-many cases can be disaggregated into one-to-
one cases through appropriate separation, and this dis-
aggregation promotes listener choice.
These observations are pragmatic rather than profound.
They are not an argument for listeners’ rights across the board.
Instead, they suggest that in a world with many competing
speakers and many diverse listeners, the structural limits of
speaker-listener matching will tend to push any system of free-
dom of expression towards respecting listeners’ choices.
But listeners’ rights are not the only direction in which the
theory of speaker-listener matching can go. I would like to
briefly sketch another, which I think is equally important. We
have a tradition of free speech theory and First Amendment
doctrine that obsesses over the problem of bandwidth scar-
city. 110 Much of what we think we know about speakers and lis-
teners—think of spectrum, public fora, and compelled
disclosures, to name just a few—takes it for granted that law
will have to mediate conflicts among speakers and listeners
over limited speech-transmission resources. But in this centu-
ry, the scarcities that matter are attention and information, not
bandwidth. The problem is not that speakers have a hard time
reaching listeners, or that listeners are denied the ability to
hear the speech they want. The problem is that listeners are
flooded with speech, far more than they can possibly listen to,
and that they know extremely little about that speech, so little
that they can’t even make minimally-informed choices about
which speech to listen to.
“Fake news” is a bad way to describe the problem, but the
problem is there, whether we have a good term for it or not.111
110. The canonical text is Jerome A. Barron, Access to the Press—A New First
Amendment Right, 80 HARV. L. REV. 1641 (1967). It has been cited by 596 articles
in the HeinOnline database as of September 23, 2018.
111. Some sources on the issues involved include ROBYN CAPLAN ET AL., DATA
& SOC’Y, DEAD RECKONING: NAVIGATING CONTENT MODERATION AFTER “FAKE
NEWS” (2018); James Grimmelmann, The Platform is the Message, 2 GEO. L.
TECH. REV. 217 (2018); ALICE MARWICK & REBECCA LEWIS, DATA & SOCIETY,
MEDIA MANIPULATION AND DISINFORMATION ONLINE (2018); Mark Verstraete et
al., Identifying and Countering Fake News (Arizona Legal Studies Discussion
GRIMMELMANN_PAGE PROOF_HRS_2.1 (DO NOT DELETE) 2/1/2019 1:41 PM
410 UNIVERSITY OF COLORADO LAW REVIEW [Vol. 90
Something has gone very wrong with the matching process in
the last few years, especially but not exclusively online, some-
thing that is hard even to see from within traditional content-
focused theories of free speech. Fake news and filter bubbles
are problems of attention and ignorance. 112 They have to do
with the content of speech, yes, but much more to do with how
content spreads from speakers to listeners. 113
Free speech theory needs to come to grips with attention
and ignorance as foundational concepts. 114 In an age of virality,
presidential tweets, and recommendation engines run amok,
attention is the resource that powerful speakers fight over, and
listener ignorance is often their weapon of choice. 115 Preserving
a safe space for a democratic culture will require both new
efforts to protect attention from hijacking and also new tools to
channel attention to places where it can be used for good. 116
The truly unifying work on a coherent First Amendment law of
attention and ignorance remains to be written. I hope that it
will not be long. The problem is urgent.
Paper No. 17-15 (2017)), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3007971
[https://perma.cc/X6Q8-YJCD]; Claire Wardle, Fake News. It’s Complicated, FIRST
DRAFT (Feb. 16, 2017), https://firstdraftnews.org/fake-news-complicated/ [https://
perma.cc/Y2ZJ-ZC5H].
112. On filter bubbles, see CASS R. SUNSTEIN, #REPUBLIC: DIVIDED
DEMOCRACY IN THE AGE OF SOCIAL MEDIA (2017); ELI PARISER, THE FILTER
BUBBLE: WHAT THE INTERNET IS HIDING FROM YOU (2011).
113. See Zeynep Tufekci, It’s the (Democracy-Poisoning) Golden Age of Free
Speech, WIRED (Jan. 16, 2018, 6:00 AM), https://www.wired.com/story/free-speech-
issue-tech-turmoil-new-censorship/ [https://perma.cc/EX97-XGGM].
114. Some initial work along these lines includes Jasper L. Tran, The Right to
Attention, 91 IND. L.J. 1023 (2016); Wu, supra note 32.
115. See TIM WU, THE ATTENTION MERCHANTS: THE EPIC SCRAMBLE TO GET
INSIDE OUR HEADS (2016).
116. See generally Jack M. Balkin, Digital Speech and Democratic Culture: A
Theory of Freedom of Expression for the Information Society, 79 N.Y.U. L. REV. 1
(2004).