What the empirical research says about trademark examination and registration outcomes

Trademark doctrine is written as if the same mark, the same goods and the same evidence produce the same result. The empirical research tests that assumption against the record of millions of applications and hundreds of court opinions. What follows is a reading of the studies practitioners should know, the public data behind them, and what they cannot tell us. Summary and opinion are labeled.

Who examines matters

The largest study of examination is Michael Frakes and Melissa Wasserman, Are There as Many Trademark Offices as Trademark Examiners?, 69 Duke L.J. 1807 (2020). It covers more than 7.8 million applications from 1982 onward, examined by 1,308 examining attorneys who reviewed more than 6,000 applications each on average (at 1824–25). Applications are assigned to examining attorneys at random. If examining attorneys applied the law alike, their outcome rates should cluster closely. They do not.

• Publication rates average 78 percent, with a standard deviation of 11.5 points. At the 2nd percentile an examining attorney publishes about 56 percent of applications. At the 98th percentile, all of them (at 1831–33).
• First substantive decisions vary more: from 11 percent publication at the 2nd percentile to nearly 84 percent at the 98th (at 1840).
• Registration rates average about 58 percent, with a standard deviation of 13.6 points (at 1842).
• Sustained oppositions average about 7 per 1,000 applications, but about 22 per 1,000 for examining attorneys at the 98th percentile (at 1845). That suggests some examining attorneys approve marks that third parties later defeat at a higher rate.
• Time to a publication decision averages about 315 days, with a standard deviation of about 135. Adjusting for application characteristics cuts that spread roughly in half (at 1848).

The spread survives the authors' controls for application characteristics, and it did not narrow over time. Variation in publication rates was "just as high in the 2010s as it was in the 1980s" (at 1852). The study does not identify the cause.

Limits. The controls are only as good as the variables in the dataset, and a mark's distinctiveness is hard to code. Random assignment within a law office does not rule out differences between law offices. The data end before the 2025 fee rules and the 2026 AI tools.

The Docket · WeeklyAnalysis like this, plus every TTAB decision summarized. Fridays.

Email addressSubscribe

The register is crowded

Barton Beebe and Jeanne Fromer studied 6.7 million applications filed from 1985 through 2016 and 2.1 million Office actions issued from 2003 through 2016, compared against the most common words and surnames in American English (Are We Running Out of Trademarks?, 131 Harv. L. Rev. 945 (2018)). They found that "about three-fourths of the time" we use a word, it is a word someone has claimed as a trademark. They also found that §2(d) refusal rates have risen. And applicants have shifted toward longer, more complex marks, which the authors describe as "second-best" choices.

Consequence. For an owner choosing a mark, the obvious short word is the one most likely to be taken. That is a reason to search early and to widen the list of candidates, not to file and hope.

The register is not fully accurate

The USPTO's own data show how much of the register claims more than its owners use. Since 2017 the Office has audited a sample of registrations at the time of their Section 8 or 71 declarations. It asks for proof of use for additional goods or services. The Office reports that audits have led it to "cancel or remove goods or services from registrations in more than fifty percent of audited registrations" (Post Registration Audit Program). Separately, the Office reported on May 12, 2026 that it had issued eleven administrative orders in six months, targeting about 10,500 applications and registrations. It also reported removing more than 52,000 invalid filings in the prior fiscal year.

Consequence. A registration found in a clearance search may cover goods its owner does not sell. Under the law, such a registration still counts until it is amended or cancelled. Expungement and reexamination exist for this reason.

In court, marketplace strength dominates

Barton Beebe's study of all 331 reported district court opinions from 2000 to 2004 that applied a multifactor likelihood-of-confusion test found that a few factors decide most cases (An Empirical Study of the Multifactor Tests for Trademark Infringement, 94 Cal. L. Rev. 1581 (2006)). The decisive ones are similarity of the marks, the defendant's intent, proximity of the goods, and the mark's strength. Of the 102 opinions finding confusion likely, 90 percent found that strength favored that result (at 1610). On strength itself, the study found that the inherent-distinctiveness half of the analysis "has broken down" (at 1582). Where a court's view of inherent strength conflicted with its view of acquired, marketplace strength, the marketplace view won almost every time. That happened in 22 of 23 opinions finding a mark inherently weak but commercially strong, and in 24 of 27 opinions in the reverse situation (at 1636).

The question now before the Supreme Court

That finding bears on RiseandShine Corp. v. PepsiCo, Inc., No. 24-1016, in which the Court granted certiorari on June 29, 2026. The question presented is "[w]hether trademark strength is a question of fact in a likelihood-of-confusion analysis under 15 U.S.C. § 1114." The petition states that twelve circuits treat strength as a question of fact and that the Second Circuit alone treats it as a question of law. The dispute is between the owner of RISE for canned coffee and PepsiCo, which launched MTN DEW RISE ENERGY. In a summary order of December 19, 2024, the Second Circuit affirmed summary judgment for PepsiCo. It held that the district court properly analyzed strength as a question of law. An earlier appeal in the case at the preliminary-injunction stage is reported at 41 F.4th 112 (2d Cir. 2022).

As of October 20, 2026, the docket shows the petitioner's merits brief (September 8). It shows amicus briefs from the United States supporting vacatur and from the International Trademark Association, the Intellectual Property Owners Association, trademark law scholars and others, several supporting neither party. The respondent's brief is due November 4, 2026. No argument date is listed. We do not predict the outcome.

Opinion. Whatever the Court decides, the empirical record suggests the practical stakes lie in marketplace evidence. If acquired strength already decides the strength factor in nearly every case that turns on it, then calling strength a question of fact sends that evidence to the jury. Calling it a question of law leaves it with the judge. Either way the evidence is what matters.

The data are public

The research above rests on public data that practitioners can use too:

• The Office of the Chief Economist's Trademark Case Files Dataset: 12.7 million applications and registrations from 1870 to March 2024, with prosecution events, classification, ownership and oppositions. It is documented in Graham et al., 22 J. Econ. & Mgmt. Strategy 669 (2013).
• The Trademark Assignment Dataset: recorded transfers since 1952.
• The TTAB Reading Room: Board decisions with proceeding type, issues, outcome, panel and precedential status. Our own count of precedential decisions comes from it.

Opinion. Three questions could be studied with these sources and have not been, as far as we know. How often refusals of each type are affirmed on appeal. Whether examiner spread has changed since 2020. How Madrid designations fare compared with direct filings. We intend to report on all three.

This is general information, not legal advice. For the owner's first step, see the Before You File guide on searching before you file.