Secondary Meaning Surveys: How to Prove a Descriptive Name Points to One Source (2026)

Short answer: a secondary meaning survey measures whether the relevant public hears a descriptive term as the name of one company rather than as a description of a product. The legal target, set by the Supreme Court in Kellogg Co. v. National Biscuit Co., is whether the primary significance of the term in the minds of the consuming public is the producer rather than the product. Three design choices decide whether your number means anything: a screened universe of real category buyers, a neutral open question before any prompted one, and a control term. Most commercial brand studies skip the control, which is exactly the piece that converts a percentage into evidence.

Every company that names a product after what the product does eventually runs into this problem. Fish-Fri, Pretzel Crisps, Booking.com, Zero, American Airlines: descriptive names are easy to market and hard to own. The law offers a bargain. A descriptive term is not protectable on day one, but it becomes protectable if you can show the public has come to treat it as a source identifier. Proving that is an empirical question about real people, and the standard tool for answering it is a survey.

The distinctiveness scale, and why only one rung needs a survey

Courts sort word marks onto an increasing scale of distinctiveness. The Supreme Court restated it in Two Pesos and the Fifth Circuit worked through it in detail in Zatarains, Inc. v. Oak Grove Smokehouse, Inc., 698 F.2d 786 (5th Cir. 1983).

The practical consequence is sharp. As Zatarains put it, a descriptive term may be elevated to trademark status with proof of secondary meaning, while a generic term may never achieve trademark protection at all. So the first question is not how to run the survey. It is whether you are on the rung where a survey can help you.

The legal target you are actually measuring

The test the Fifth Circuit applied, quoting Kellogg, is that the plaintiff must show that the primary significance of the term in the minds of the consuming public is not the product but the producer. Read that carefully, because it is not the same as brand awareness, and it is not the same as liking.

You are not measuring whether people have heard of you. You are measuring whether, when they hear the words, they think the words name a single company rather than a kind of thing. That is a specific cognitive claim, and it maps onto a specific question sequence.

There is also a shortcut worth knowing. Under 15 U.S.C. 1052(f), the Director may accept as prima facie evidence that a mark has become distinctive proof of substantially exclusive and continuous use as a mark for the five years before the claim of distinctiveness is made. Five years of exclusive use is not a guarantee, and it does nothing for you in a contested infringement case against a well-resourced opponent, but it is the cheapest route when nobody is fighting you.

What the accepted numbers actually look like

Commercial brand teams tend to assume a legal threshold exists. It does not. What exists is a set of decided cases.

In Zatarains, the plaintiff ran two surveys through its expert. The first was a telephone study of 100 women in the New Orleans area who fry fish or other seafood three or more times per month; 23 percent named Zatarains Fish-Fri as a product they would buy at the grocery to use as a coating. A parallel mall-intercept study of 100 respondents produced 28 for the same questions. The district court found secondary meaning in the New Orleans area, and the Fifth Circuit affirmed while saying plainly that considering the question afresh it might have reached a different conclusion, because the issue is close.

Two things follow. First, 23 percent was enough, in combination with sales and advertising evidence, in a defined geography. Zatarains had sold 916,385 cases of Fish-Fri between 1964 and 1979, and the court treated that circumstantial evidence as weighing heavily. Second, secondary meaning can be geographically limited. The finding was for the New Orleans area, not the United States.

The same opinion supplies the cautionary half. The identical survey instrument produced only 11 of 100 respondents naming Chick-Fri, chicken fry, or Zatarains Chick-Fri as a coating for frying chicken, and the registration was cancelled. The reason was not the number. It was the screener, and we come back to that below.

The control term is what converts a percentage into evidence

The single most useful design element in a secondary meaning survey is a control: a term that nobody could plausibly own, run through the identical question sequence with a matched sample.

Royal Crown Co. v. Coca-Cola Co., 892 F.3d 1358 (Fed. Cir. 2018), shows why. Coca-Cola submitted a survey asking whether respondents associated the term ZERO with one or more particular companies. Sixty-one percent associated ZERO with one company. On its own that number is unreadable, because a chunk of any sample will associate almost any word with one company if you ask them to. The survey also ran the control term DIET, and only 6 percent associated DIET with one company. The gap, not the raw 61, is the finding.

Without a control you cannot answer the question an opponent will ask first: what would this number have been for a word you concede is generic? In USPTO v. Booking.com B.V., 591 U.S. 549 (2020), the survey showed 74.8 percent of participants thought Booking.com was a brand name. The same survey found 33 percent thought Washingmachine.com, which corresponds to no company at all, was also a brand. Justice Breyer built his dissent on exactly that comparison. Whatever you think of the outcome, the lesson for a research team is free: the control item is where the argument happens.

Surveys expire

Royal Crown also delivered a warning that applies far outside trademark law. Coca-Cola conducted its survey more than five years before the close of testimony, and the Federal Circuit held that a survey is only probative if it deals with conditions at the appropriate time, questioning whether stale data could support the finding even as corroboration.

If a percentage is load-bearing for a decision, it has a shelf life. A brand-perception number from 2021 is not evidence about 2026, and no amount of methodological care fixes that. Cheap research is the only real defence, because it lets you re-run rather than re-cite.

How to design one

1. Define the universe from the claim, not from your CRM. Prospective and actual purchasers of the category, in the geography where you claim rights. See survey universe for the full protocol.
2. Screen, and record the screener. Category purchase in a defined recency window, and an exclusion for anyone working in the industry, in advertising, or in market research.
3. Ask unaided before aided. An open question first, so that respondents supply the frame instead of receiving it.
4. Ask the source question in source language. Do you associate this term with one company, more than one company, or no company in particular. Then: what company, and what makes you say that.
5. Run a control term through the identical sequence. Matched sample, same interviewer script.
6. Report the gap, the base sizes, the screener, and the refusals. A number without its denominator is not a finding.

The reason-why probe in step 4 is what most instruments truncate, and it is the difference between a percentage and an explanation. Traditional survey tools force a compromise here: you either accept a one-word answer or you pay a moderator to sit through 40 interviews.

How Koji fits

Koji is an AI-native research platform built on conversational interviews rather than static forms, which changes the economics of this specific study in three ways.

First, Koji supports all six structured question types - open_ended, scale, single_choice, multiple_choice, ranking, and yes_no - inside a conversation rather than in a form. The classification question can be single_choice and the reason-why probe can be open_ended, in the same session, with the AI following up on whatever the respondent actually said. That combination is the whole design of a defensible distinctiveness instrument, and the structured questions guide walks through how each type is rendered and analysed.

Second, the AI interviewer chases the reason-why on every respondent, not on a subsample. Where a traditional survey collects one word, Koji collects one word plus the explanation behind it, and the automatic analysis clusters those explanations into themes without anyone hand-coding transcripts.

Third, intake forms and screener logic mean the universe definition is enforced at the door rather than repaired in the crosstabs, and voice or text is the respondent choice rather than yours. A study that used to take six weeks and a five-figure fielding budget runs in days.

One honest limitation. If you are heading into litigation or a contested proceeding before the Trademark Trial and Appeal Board, retain a qualified survey expert and let them design and interpret the instrument. In Elliott v. Google, Inc., 860 F.3d 1151 (9th Cir. 2017), two of the three surveys offered were excluded because they had been designed by counsel rather than by someone qualified to design or interpret surveys. What platforms like Koji make cheap is everything that happens before that point: piloting the wording, testing the screener, and building the ordinary-course research record that makes any later expert study credible.

Common mistakes

• Measuring awareness and calling it secondary meaning. Recognition is not source identification. Booking.com and Washingmachine.com are the proof.
• No control term. The most common defect, and the one that is impossible to repair after fielding.
• A screener written for a different product. See the Chick-Fri result.
• Leading the witness. In American Thermos, the plaintiff asked respondents to name any trademark or brand names for vacuum containers. The court noted the question left little opportunity for the revelation of a generic use, and discounted it.
• Claiming nationwide rights on a regional sample. Zatarains won New Orleans, not America.
• Letting the number age. Five years is too old. One year is usually too old for a fast-moving category.

Frequently asked questions

What percentage proves secondary meaning?

There is no statutory threshold. Courts have credited figures in the low twenties in combination with strong circumstantial evidence, as in Zatarains where 23 and 28 percent were affirmed as a close call. Practitioners commonly treat 50 percent as comfortable and anything under 20 percent as weak, but the gap between your term and your control matters more than the absolute number.

What is the difference between secondary meaning and acquired distinctiveness?

They are the same concept under different names. Courts usually say secondary meaning; the Lanham Act and the USPTO usually say acquired distinctiveness, and 15 U.S.C. 1052(f) is the registration route.

Do I need a control group for a brand study that is not about litigation?

Yes, for the same reason. Without a control you cannot tell whether your number reflects your brand or reflects the way people answer that style of question. The control is cheap and it is the first thing a sceptical executive will ask about.

How many respondents do I need?

Reported trademark surveys commonly run 200 to 400 completes per cell in the relevant universe, and the classic cases ran larger. Zatarains used 100 per city and the Fifth Circuit cautioned that samples of that size may not be adequate to prove much of anything. Base size per cell matters more than total N.

Can AI-moderated interviews substitute for an expert survey in a legal proceeding?

Not as a substitute for a retained expert. They are extremely useful before and around one: piloting question wording, testing whether the screener reaches the right people, and building a record of routine research that predates any dispute.

How does this connect to my ordinary brand tracker?

Directly. A tracker measures awareness and perception over time; a distinctiveness study measures whether the name functions as a name. Running the classification question and a control inside your existing tracker costs almost nothing and gives you a dated series instead of a one-off number.

Related Resources

• Structured Questions Guide - the six question types and when to use each
• Survey Universe: Defining Who Counts - the screener and sampling protocol behind every number above
• Likelihood of Confusion Surveys - the Eveready and Squirt formats
• Genericness Surveys - what happens when a name goes the other way
• Name Testing Research - validating a name before you commit to it
• Brand Awareness Survey Guide - the tracking study this sits inside
• Survey Evidence in Court - admissibility, control groups, and the five-element record