Trademark Confusion as a Matter of Law, Andrew Michaels

Is infringement a proposition about the world, or about the
law? Fact: an empirical predictive question about the world. Law: a normative
judgment about whether there should be legal responsibility—is the confusion
likely enough that we should find infringement? 2d/Minority view: Q of law
based on underlying facts. 9th and majority view: Q of fact reviewed for clear
error. 9th said that LOC decisions have “limited precedential value” b/c they
stand on their own facts, reducing the need for de novo review. But that might be
a reason to treat it as a Q of law to get more consistency and predictability.
Issue of law would make it easier to decide on SJ/without trial, compared to
claims that SJ should usually be avoided.

Judges and juries are thought to be good at different
things. Jury: community; judge: compare with other cases/predictability
allowing businesses to order their affairs more easily. Easier to decide on SJ:
might help prevent bullying of parodists, other users.

Appellate v. trial court: underlying factors of intent,
actual confusion might be better assessed by trial court; balancing/weighing of
factors and legal comparison might be more suited for appellate court. Some
factors may be more factual: evidence of actual confusion; intent (witness testimony/credibility).
Similarity of marks should be legal because the jury has to ask “compared to
what?” whereas the court can look at other cases. Same with products. Strength
of mark conceptually should be Q of law; commercial strength is a matter of
fact for witness testimony.

Lisa Ramsey: Matal v. Tam—constitutional issue exists, and
LOC is a speech protective doctrine, implicating Bose. JDI even says that MTD can
be ok because of contextual considerations. If it can be resolved on a MTD, it
can be a Q of law.

RT: Bose v. Consumers Union on de novo review when the facts
have constitutional significance. On the “compared to what” for similarity of
marks/similarity of products? the theory is “similar enough that consumers are
likely to confuse them.”

James Dabney: time was that likely confusion would be enough
for an injunction, not damages or disgorgement; now things are different.

Google v. Oracle—is this legal or factual? Similar issue of
mixed question of law & fact.

Q: right to jury trial?

A: could ask them questions about the factors; could ask for
an advisory jury verdict, which they do a lot for patent obviousness/did with
GvO. Multifactor=often an issue of law.

McKenna: LOC factors were made up; makes it feel more fact
bound b/c courts think they have to walk through the factors even when they are
ill-fitting. The legal standard is supposed to be: substantial number of
reasonable consumers. Look at negligence where courts are more willing to grant
SJ because they are more willing to consider what reasonableness is.

Factors and Fictions: The Empirical Collapse of the
Likelihood-of-Confusion Test Across the Federal Courts, Thomas Reichert

Every circuit makes the same 4 commitments: (1) the test is
flexible; bright lines misfire; (2) no factor is dispositive; (3) the set is open
so you can bring in other considerations/add factors; (4) provides structure and
allows appellate review. But: How often is this true?

Used an LLM to read every confusion opinion 1970-2025,
temperature set low to inject less randomness, and ask whether the court
considered a factor and how strongly it favored/disfavored confusion. The model
is not trying to judge factor weight itself, just trying to identify what the
court said about how the factor weighed. Around 11,000 opinions analyzed.
Courts analyze 6.35 factors/case; only 40% consider them all.

Hand audited 1002 codings; 97% agreement on weight and 100%
on direction.

The key factors: similarity of marks and proximity of
goods/services. If both favor confusion, predict confusion; if both don’t,
predict lack of confusion. Can predict 93.5% of every federal TM case. Consistent
across circuits and time, though less in 8th circuit where the case count is
small, and there’s a dip in the 90s (his hypothesis is domain names). The other
factors operate as “structured overrides.” Defeaters are where both factors
favor confusion but the court finds none. A lot of cases: no actual confusion,
high buyer sophistication, good faith adoption, and weak mark strength. Substitutes:
a predicate factor was weak/divided, but confusion found anyway: strong/famous
mark; bad faith intent.

The test is already hierarchical. Courts should say so.
Appellate practice wrongly rewards factor by factor mark. Could right size
discovery/do less initial discovery. Tell juries the machinery: model jury
instructions in 9th Circuit already tell juries what weighs more and we could
do more.

Could apply the same questions to © fair use; sentencing.

Betsy Rosenblatt: has been done for © fair use—you may not
want to reinvent the wheel. But one interesting thing about © fair use that
might or might not match w/TM infringement is that how one comes out on
transformativeness tends to predict how the case comes out. It doesn’t mean
that transformativeness is the whole game; but it influences how the other
factors work rather than rendering them unimportant. In general we may want
those other factors to be doing more work than they’re doing, not less. You may
have identified a problem rather than a solution. Should juries perhaps pay
more attention to sophistication? Right now they don’t have a good definition.

Should parody be a special case b/c the factors work differently?

Q: Fed Cir has criticized TTAB for relying too much on
similarity, so that result is pretty funny (the Fed Cir was most likely to rely
on the 2).

McKenna: it’s not that the other factors just come out—the question
of how much similarity there is b/t marks and goods are not found in nature. It’s
not a © comparison. All the other information is just influencing the judgment
about similarity. It’s context for which you understand levels of similarity,
informed by all the other information. So sequencing discovery would be
difficult b/c you’d be ruling out the contextual information you need to make
judgments about similarity of marks and similarity of products.

A: we’re measuring the opinions, not the reasoning process.

McKenna: sure, but your prescriptions make assumptions about
how the reasoning works.

A: Crowding in the market can definitely change similarity assessments.

Q: if courts were honest and said it’s a 2 element test,
with a determination made through a bunch of subfactors, would that work
better?

A: that’s the next paper. You can create a flowchart of how
to do the analysis with substitutes/defeaters. You could do a burden shift! [Burden
of production I assume, not burden of proof.]

Ramsey: dilution doesn’t consider relatedness of goods—does that
matter?

A: didn’t look for any correlation w/dilution.

Ramsey: some courts say strength increases likely confusion,
but academics and parody cases say that strength can decrease likely confusion
b/c people know what the real thing looks like.

A: strength moves w/the verdict generally, but can
substitute for proximity if the mark is very strong.

Ramsey: should separate out commercial & conceptual
strength & see what happens.

Q: the other factors were originally not relevant to competing
goods situations; practitioners got into the habit of applying Polaroid/etc. in
all circumstances when it wasn’t needed in the direct competition cases. That
would support the empirical observation that competition and similarity, the
two pre-Polaroid metrics, were actually always the most important.  

21st Century Trademark Surveys, Rebecca Tushnet (with Chris
Sprigman & Stephan Tontrup)

A statutory interpretation component: what do the terms in
the statute like affiliation and connection actually mean? Weird that we don’t
have much of an answer after 80 years, isn’t it? So we believe the definitions
we are using are grounded in the proper legal meaning of affiliation et cetera.

The empirical part: we currently don’t tell survey
respondents (or jurors) what “affiliation” etc. means and we also don’t have any
good reason to think that they know what it means for legal purposes, which doesn’t
include references—if you think of Sprite when you see Poppi Lemon-Lime, there
can’t be deception about affiliation or connection because you really did think
that, but a layperson could say “yes, there’s a connection”—the survey may not
even be revealing mistakes of law, as Sotomayor et al have discussed with
parodies, but mistakes about the meaning of the words used in the survey! So
let’s try to fix that with a training module as in genericness surveys and see
what happens. Including allowing a response “this is about the
trademark/trademark owner.”

And implications for jury instructions: survey respondents
and jurors are in the same position.

Larger questions: there are lots of areas where we want to
know how some audience perceives communication: 1A compelled speech/will you be
associated w/the statement; labor law: how employees perceive employer
speech—we don’t ask the workers! True threats—hypothetical reasonable
person.  But only in the Lanham Act do we
actually use surveys! [Probabalistic—less than half can still be a large number
of people for economically, socially, or politically significant messages. Why
is probabilistic thinking persuasive in TM & not other areas? Plaintiffs’
bar? Courts willing to credit that “substantial numbers” matter even if not
majority b/c they can imagine the harm to the substantial consumer mass/the
consumer mass may not be imagined to share any other minoritarian identities
(or may be imagined to need special protection—cite Ann Bartow on gender)?
Possible lesser importance of public interest lets courts defer to surveys in
TM and rely on policy preferences in 1A—though that lack of interest in reality
on the ground is not necessarily good for 1A jurisprudence.

Rosenblatt: affiliation and approval are easy to get wrong;
even experts get them wrong. Pattern jury instruction?

Q: we’re in the post literate era: disconnect b/t regular
people and lawyers. If literacy rates are going down, we need to define terms
for them.

Ramsey: courts focusing on text of statute: approval
language concerns me. We don’t want people to be confused about permission. Don’t
ask compound questions. “goes along with” is a bad definition too.

RT: approval and permission aren’t the same thing but this
is where mistake of law comes in. There is an issue with repeating questions
too—that’s more likely to get a “yes” somewhere in there.

Rothman: Working on project w/Joel Steckel—one of the things
we worked w/was mini survey about meaning of these terms and people were using
lay definitions. How should they be defined?

An Axe to Grind? The Legal History and Trademark Challenges
of Guitars, Mark Blankenship

When does a guitar shape identify a type of guitar v.
manufacturer? Sears Roebuck catalog is the precursor to Amazon and Temu, making
gear affordable to players who didn’t live near a music store. Different claims
over time—Japanese “knockoffs” that eventually resolved into new body shapes as
well as some generic ones. Other issues: German court allowed © claim in guitar
body shapes; separability would be an issue in the US unless the guitar also
includes features like shark fins.

from Blogger https://tushnet.blogspot.com/2026/08/ipsc-breakout-session-3-trademark.html