Designing a Defensible Confusion Survey: A Practical Checklist

Intellectual Property > Trademark Litigation | Published 7 December 2024 - Updated 8 March 2025 | UpLaw - uplaw.us

A confusion survey persuades only after it survives. This checklist tracks the design choices in the sequence that determines both admissibility and weight—which is also, not coincidentally, the sequence in which opposing counsel will take the survey apart.

Keywords: confusion survey, Eveready format, Squirt format, survey universe, control group, net confusion, likelihood of confusion, Daubert, Federal Rule of Evidence 702, double-blind, marketplace realism, secondary meaning survey, genericness survey, Teflon survey, Thermos survey, sample size, screening questions, trademark litigation, survey expert

---

A consumer survey enters a trademark case wearing a lab coat. It arrives with percentages, confidence intervals, and an expert who will testify that the number means what it says. But a survey is not evidence of confusion the way a photograph is evidence of a car crash. It is a constructed measurement—a machine built by human beings to detect something invisible (what strangers believe about the source of a product) and to convert that belief into a figure a court can weigh. Like any instrument, it measures accurately only if it is built correctly, and it produces a clean reading only if it is pointed at the right thing under the right conditions.

That is why the value of a trademark survey lives entirely in its methodology. The universe must capture the right consumers; the format must fit the marks; the questions must not whisper the answer; the administration must be blind; the stimulus must look like the marketplace; and the confusion must be netted against a control before any percentage means anything at all. Get one of those wrong and the number does not become less persuasive—it can become affirmatively misleading, a gift to the other side, and, increasingly, a candidate for exclusion.

This checklist is written for the two people who build the instrument together: the survey expert who designs and executes it, and the trademark litigator who frames the legal question, feeds the expert the record, and will have to defend (or attack) the result at a Daubert hearing and at trial. It is organized by phase, in the order the decisions matter. Each phase carries a WHY note explaining the doctrine and the traps a competent cross-examiner will exploit. Real decisions anchor the analysis; clearly labeled hypotheticals illustrate it. For the underlying doctrine of how surveys are built and read, pair this checklist with the UpLaw guide to expert survey methodology; for the admissibility fight, see Daubert and the Consumer Survey.

How to Use This Checklist

Work the phases in order, top to bottom, the first time through. The decisions are not independent—each one constrains the next. The universe you pick dictates the screeners; the theory of confusion dictates the format; the format dictates how hard the control has to work. Skipping ahead to "what questions should we ask" before settling the universe is how expensive surveys end up measuring the wrong people with beautiful precision.

A word on division of labor. The survey expert owns the science: sampling, instrument design, administration, coding, and statistics. Counsel owns the law and the record: the theory of confusion, the relevant goods and channels, the marks and trade dress actually in use, and the marketplace facts the expert must replicate. Counsel does not (and must not) conduct the survey, but counsel supplies the expert with the raw material and stress-tests every design choice against how it will read in a deposition. The Practical Law survey literature puts the point plainly: attorneys "do not (and should not) conduct the survey," yet they "typically provide the survey expert information and feedback relevant to survey design and construction." That collaboration, documented contemporaneously, is itself part of the defense.

The Order of Attack: Why Sequence Matters

Cross-examination of a survey expert is a demolition performed in a fixed order, and it is the same order in which the survey was built. The examiner starts at the foundation—who did you survey, and why those people?—because a universe error cannot be repaired by anything downstream. If the wrong people were surveyed, it does not matter how elegant the questions were or how large the sample was; the instrument measured the beliefs of the wrong population, and no cross-examination, argument, or re-weighting fixes that. Only after the universe does the examiner move to format, then control, then stimulus, then question wording, then blinding, then sample and coding. Building the survey in that same sequence—and hardening each layer before moving to the next—means that when the attack comes, every question has already been answered in writing.

Phase 1 — The Threshold Decision: Commission a Survey at All?

• [ ] Confirm the disputed issue genuinely turns on consumer perception (source, sponsorship, or affiliation confusion; secondary meaning; fame; genericness) rather than on facts a survey cannot reach.
• [ ] Confirm the budget and calendar support a rigorous design—commonly a five- to six-figure, multi-month undertaking with test and control cells.
• [ ] Confirm a qualified, independent expert believes a defensible survey can actually be built for these marks in this market—and would find what you hope.
• [ ] Weigh the alternatives: strong marks plus closely related goods plus real-world actual-confusion witnesses may carry the day without a survey.
• [ ] Assess adverse-inference exposure: some courts infer that a well-resourced party who declines to survey feared the result.
• [ ] Retain the expert early, ideally before the theory of confusion hardens, and run a pilot under work-product protection before committing to a full study.

WHY / traps. A survey is powerful but never mandatory. Courts and the TTAB "often give properly designed and conducted surveys great weight even though they are not required." A trademark owner survives summary judgment on the multi-factor test all the time without one, and a defendant can prevail on dissimilarity, sophistication, and distance of goods without ever polling a consumer. The decision to survey is a cost-benefit judgment, not a reflex.

Two forces push in opposite directions. On one side sits the adverse inference: a handful of courts have suggested that a party with the resources to run a confusion survey who chooses not to invites the inference that the results would have hurt it. See Playboy Enters., Inc. v. Netscape Commc'ns Corp., 55 F. Supp. 2d 1070, 1084 (C.D. Cal. 1999); Eagle Snacks, Inc. v. Nabisco Brands, Inc., 625 F. Supp. 571 (D.N.J. 1985). The inference is neither universal nor automatic, and many judges are skeptical of it, but you must research the presiding judge's stated views on survey evidence before deciding.

On the other side sits the reality that a mediocre survey is often worse than none. A flawed survey hands the opponent a ready-made Daubert target and, worse, can generate a number that cuts against the party who paid for it. If a competent expert cannot in good conscience build a survey that finds confusion, that is not a scheduling problem—it is diagnostic information about the merits of the case. The professional move is to run a small pilot first. Because a test survey commissioned by counsel for trial preparation is ordinarily protected work product, an unfavorable pilot can be quietly retired. Keep the pilot within the attorney's direction and the protective umbrella; a survey a party commissions on its own, outside counsel's direction, may not be shielded at all, and once an expert is designated for trial the protections around drafts and counsel communications narrow considerably.

Timing deserves its own line. A party racing for a temporary restraining order or preliminary injunction may simply not have the weeks a full study requires; a stripped-down or expedited survey, honestly labeled as such, may be the only option, and its limitations should be disclosed rather than hidden. For where the survey decision sits in the broader enforcement sequence, see the trademark enforcement and litigation lifecycle toolkit.

Phase 2 — Define the Universe (the Load-Bearing Decision)

• [ ] Pin down the theory of confusion: forward, reverse, sponsorship/affiliation, initial-interest, or post-sale.
• [ ] For forward confusion, set the universe to prospective purchasers of the junior (defendant's) goods or services—the people at risk of being drawn to the defendant by the senior mark's pull.
• [ ] For reverse confusion, flip the universe to prospective purchasers of the senior (plaintiff's) goods—the people who may wrongly assume the small senior user is affiliated with the ubiquitous junior user.
• [ ] For dilution/fame, sample the general consuming public of the United States, not a niche buyer pool. 15 U.S.C. § 1125(c).
• [ ] For secondary meaning, sample the relevant purchasers of the goods bearing the claimed mark.
• [ ] Draft screening (filter) questions that actually operationalize the universe: correct product category, purchase timeframe, price band, and role in the buying decision.
• [ ] Guard both edges: avoid an over-inclusive universe (sweeps in non-purchasers, diluting and distorting the rate) and an under-inclusive one (excludes real purchasers, so the result cannot be generalized).

WHY / traps. The universe is the single most important decision in survey design and the single most common fatal flaw. The reason is structural: a survey estimates the beliefs of a defined population by polling a sample of it. Define the population wrong and every downstream number is an accurate measurement of the wrong thing. As the Second Circuit put it in the classic formulation, a survey of the wrong universe is entitled to little or no weight, because "the wrong universe" is not a defect the fact-finder can discount and still use—it is a defect that voids the instrument. See Bristol-Myers Squibb Co. v. McNeil-P.P.C., Inc., 973 F.2d 1033, 1044-45 (2d Cir. 1992) (an over-inclusive universe undermines the survey's probative value). The Practical Law survey literature captures the standard: to be admissible a survey must "properly define the relevant universe," and in a forward-confusion case that universe "typically consists of actual or prospective purchasers of the junior user's products or services." See Manual for Complex Litigation (Fourth) § 11.493.

The forward/reverse distinction is where careful lawyers earn their keep, because the two theories point the survey at opposite populations. In forward confusion, the harm runs downhill from a strong senior mark: consumers see the junior user's product, remember the famous senior brand, and wrongly assume the senior stands behind it. You survey the junior's prospective buyers because they are the ones exposed to the junior's mark. In reverse confusion, the harm runs the other way: a small senior user is swamped by a junior user who saturates the market, until consumers assume the senior is the copycat or an affiliate of the giant. There you survey the senior's prospective buyers, because they are the ones whose perception of the senior user is being distorted. Confuse the two, and you will have measured the wrong shoppers with perfect rigor.

Hypothetical (reverse confusion). Assume a tiny five-year-old apparel maker, Fernweh Trail Co., sells merino running shirts to a devoted ultramarathon niche. A national athleisure retailer launches a "FERNWEH" line with a nine-figure ad budget. Fernweh Trail sues for reverse confusion. The correct universe is prospective purchasers of Fernweh Trail's shirts—the ultramarathon buyers who may now assume their favorite small brand is a line extension of the national retailer. A survey of the retailer's mass-market athleisure customers would answer the wrong question entirely.

The screeners are the universe's plumbing, and they are a favorite back door. A screener that is too loose lets non-purchasers into the sample (the over-inclusion problem walks right back in); a screener too tight excludes real buyers and destroys generalizability. In a case over high-end watches, a screener that asks only "have you bought a watch" is useless; it needs a price floor. The online-survey case law rewards precise screening: in Denimafia Inc. v. New Balance Athletic Shoe, Inc., 2014 WL 814532 (S.D.N.Y. Mar. 3, 2014), the expert first established that the likely buyers of the plaintiff's premium jeans were "online shoppers who purchased denim jeans priced between $100 to $1,000 within the last 12 months," and the court rejected the argument that surveying all adult internet shoppers made the universe too broad—in part because the plaintiff offered no rebuttal expert and the survey found zero confusion regardless. The lesson is not that broad universes are safe; it is that the screener must be tied by evidence to the actual marketplace.

Two cautions. First, the format does not change the universe. A Squirt survey run on the wrong universe is exactly as defective as an Eveready survey run on the wrong universe; choosing a fancier format does not cure a population error. Second, universe definition is where counsel's record and the expert's science meet: the expert cannot know the price band, the channels, or the decision-makers without counsel's factual development. This is the same population-and-market work that drives the multi-factor confusion analysis—see Working Up Likelihood of Confusion: A Practical Checklist and the brand-owner's overview of the confusion tests in Likelihood of Confusion in Trademark Law.

Phase 3 — Choose the Format

• [ ] Choose Eveready when the senior mark is strong or famous: show respondents only the junior mark/product and ask open-ended, memory-based questions—who puts this out, what makes you say that, what else does that company make, would they need permission.
• [ ] Choose Squirt when the senior mark is weaker, or when consumers genuinely encounter both marks together or in close sequence: show both marks/products and ask a neutral same-source/affiliation question.
• [ ] For a Squirt design, embed the test marks in an array/lineup of unrelated marks to avoid a naked two-way face-off that no shopper would experience.
• [ ] Consider the aided Eveready hybrid where the senior mark is not famous enough for pure Eveready but a side-by-side would be too suggestive.
• [ ] Match the format to the facts of the case—do not default to the format your expert used last time.

WHY / traps. The two workhorse formats trace to two appellate decisions and embody two different theories of how confusion happens. The Eveready format, named for Union Carbide Corp. v. Ever-Ready, Inc., 531 F.2d 366, 385-88 (7th Cir. 1976), does not show respondents the senior mark first. It shows only the junior user's product and asks who makes it and why—relying on the respondent's own memory to supply the senior brand if the association exists. That design "minimizes suggestion," which is why it is the gold standard when the senior mark is strong enough that consumers carry it in their heads. Its weakness is the mirror image of its strength: run Eveready on an obscure senior mark and it will under-count confusion, because respondents cannot volunteer a brand they have never heard of.

The Squirt format, from SquirtCo v. Seven-Up Co., 628 F.2d 1086 (8th Cir. 1980), shows respondents both marks (often among others) and asks whether the products come from the same or affiliated companies. It is "generally appropriate where consumers may encounter the marks either together or in sequence in the market." Its danger is suggestion: juxtaposing two marks a shopper would never see side by side can manufacture a perceived relationship that exists only inside the instrument. That is why a well-built Squirt survey buries the test marks in a lineup of unrelated marks—a naked two-way comparison practically invites respondents to hunt for a connection.

There is a recognized middle path, the aided Eveready, which shows both parties' uses (a Squirt trait) but then tests for confusion with open-ended questions about the junior use (an Eveready trait). See Jerre Swann & R. Charles Henn Jr., Likelihood of Confusion Surveys: The Ever-Constant Eveready Format; The Ever-Evolving Squirt Format, 109 Trademark Rep. (TMR) 671, 677 (2019). It can be the right tool when the senior mark is real but not top-of-mind famous.

Hypothetical (format selection). Assume LUMENARA is a nationally advertised premium candle brand. A newcomer sells reed diffusers as "Lumenara Home." Because LUMENARA is strong, the plaintiff's expert should run Eveready: show only the "Lumenara Home" diffuser and ask who makes it. If a healthy share of respondents volunteer the candle company, that is memory-driven confusion with minimal suggestion. Now change the facts: two small regional hot-sauce makers each use a chili-and-flame trade dress and sit on the same grocery shelf. Neither is famous, and shoppers really do see them together. Here Squirt fits the marketplace—but only with the test bottles salted among a lineup of other hot-sauce trade dresses, so the survey does not stage a face-off the aisle never would.

The cross-examiner's move here is to ask why this format. If a plainly famous mark's expert chose Squirt, the honest answer is sometimes that an Eveready pilot showed too little confusion and the party went shopping for a more suggestive design. That story, once told, is fatal. Choose the format the marketplace dictates and be able to explain the choice in one clean sentence.

Phase 4 — Build the Control and Plan the Net

• [ ] Design at least one control cell identical to the test cell in every respect except the allegedly infringing feature—swap in a plainly non-infringing mark, or alter/remove the challenged element.
• [ ] Plan to report net confusion = test rate − control rate, never the raw test figure.
• [ ] Make the control especially robust for Squirt designs, where juxtaposition inflates the raw rate.
• [ ] Consider multiple controls to rule out competing explanations for a "confused" answer (product-category association, belief that the junior itself makes the other product, etc.).
• [ ] Choose a control that is plausible in the relevant market, not a strawman the other side can dismiss.

WHY / traps. A confusion percentage without a control is close to uninterpretable, because every survey generates noise—guessing, pre-existing beliefs, yea-saying, and artifacts of the instrument itself. The control's job is to measure that noise so you can subtract it. The first question a sophisticated reader (or judge, or opposing expert) asks is not "what was the rate?" but "what was the control, and what is the net?" A Squirt survey with no control is among the most exploitable artifacts in all of trademark litigation, because the very juxtaposition that defines the format also inflates the raw number.

The best illustration in the case law is the double control in the Cracker Barrel dispute. In Kraft Foods Group Brands LLC v. Cracker Barrel Old Country Store, Inc., No. 13 C 0780 (N.D. Ill. July 1, 2013), aff'd, 735 F.3d 735 (7th Cir. 2013), Kraft (owner of CRACKER BARREL cheese) challenged the restaurant company's plan to sell CRACKER BARREL ham in supermarkets. Kraft's expert ran an online Eveready survey and, crucially, built two controls. One control showed a whole-ham product from a different brand, to confirm that respondents were not simply free-associating "ham" with "cheese." A second control showed the defendant's own ham but told respondents to assume they were seeing it inside a Cracker Barrel restaurant store, to confirm that the "cheese" answers were not just respondents assuming the restaurant company also makes cheese. Only after both controls ruled out those innocent explanations could the remaining answers be attributed to genuine confusion with Kraft's brand. The court credited a net confusion level of 18.8% and issued the injunction. That is what a survey looks like when the control does real work: it does not just subtract a number, it forecloses the alternative stories the defense would otherwise tell.

Hypothetical (control design). Return to LUMENARA candles versus "Lumenara Home" diffusers. The test cell shows the diffuser and asks who makes it. A defensible control shows an identical diffuser under an invented, plainly non-infringing house name—say "Aster & Ember Home"—to the control group. If 22% of the test group names the candle company but 7% of the control group names it too (perhaps because "candle-adjacent home fragrance" primes some guessing), the reportable figure is a net of 15 points, not 22. The control did not weaken the case; it made the 15 credible.

Two traps. First, an implausible control is nearly as bad as none: the other side will argue you rigged the baseline. In T-Mobile US, Inc. v. Aio Wireless, LLC, 991 F. Supp. 2d 888 (S.D. Tex. 2014), the defendant attacked a brown control color as absurd for the wireless industry; the court rejected the attack because rivals in the space did use brown-family colors, so the control was fair. Pick a control the market can defend. Second, do not let the control leak the test hypothesis—if the only difference between cells is a giant, obvious swap, respondents may sense the game.

Phase 5 — Stimuli and Marketplace Realism

• [ ] Present the marks the way consumers actually meet them: full trade dress, packaging, color, size, and point-of-sale context—not a bare word mark on a white screen.
• [ ] Replicate the relevant channel: shelf display for a grocery product, a product-detail page for an e-commerce item, an app store listing for software.
• [ ] Supply the context a shopper would have: the store where the product sells, its price, and neighboring products, where those cues drive source perception.
• [ ] For an online survey, confirm the stimulus renders faithfully across monitors, browsers, and devices; avoid stimuli that require three-dimensional or tactile inspection.
• [ ] Rotate the order of stimuli and answer options to neutralize order effects.

WHY / traps. A survey measures marketplace confusion, so the stimulus must resemble the marketplace. Two bare word marks floating on a white background, shown side by side, measure something—but not the likelihood that a shopper will be confused in the wild, where trade dress, packaging, price, and shelf position all carry source-identifying information. The realism problem has special force in trade dress cases, where the claimed mark is the total look and feel; stripping it down to a logo tests the wrong thing.

Realism cuts both ways, and the leading case is a caution against over-controlling as much as under-controlling. In Fortune Dynamic, Inc. v. Victoria's Secret Stores Brand Mgmt., 618 F.3d 1025 (9th Cir. 2010), the district court excluded a survey partly because it "improperly compared the products side-by-side" and "failed to replicate real world conditions." The Ninth Circuit reversed, holding those criticisms went to weight, not admissibility, and that the survey raised a genuine fact issue. The takeaway is nuanced: an artificial side-by-side is a real methodological weakness a good cross will exploit—but whether that weakness sinks the survey or merely dents it is a case-specific, often appellate question. Better to avoid the weakness than to litigate it. For how those weight-versus-admissibility calls are reviewed on appeal, see Standards of Review in Second Circuit Trademark Appeals.

Online administration has made realism both easier and trickier. Easier, because a product actually sold online can have its purchase environment reproduced almost exactly. Trickier, because a computer screen may not fairly portray a product meant to be held, and because monitor size, resolution, and browser can distort color and detail. When color is the mark, the point is sharp: in T-Mobile, the defendant argued that an internet color survey was invalid because "color settings vary by computer," but the court held that respondents' ability to associate the magenta with T-Mobile even across shades confirmed the strength of the association. Where the marketplace context supplies source cues—store name, price, adjacent products—the survey should supply them too; the online-survey literature specifically endorses "informing respondents of the store where the product is sold," "telling respondents the price," and "showing other products that are also sold in the same location" when those facts drive perception.

Hypothetical (stimulus realism). In the hot-sauce Squirt survey, do not show two cropped logos. Show photographs of the two bottles as they sit on the shelf—full labels, caps, neck tags, shelf tag with price—embedded in a lineup of other hot-sauce bottles. If confusion survives realistic, cluttered presentation, it is far more probative than confusion conjured by two isolated logos.

Phase 6 — Draft the Questions

• [ ] Use open-ended, non-leading questions that let respondents answer in their own words before any prompting.
• [ ] Always offer a genuine "don't know / no opinion" option so uncertainty is not silently coded as confusion.
• [ ] Probe spontaneous answers neutrally: "What makes you say that?" without hinting at a preferred response.
• [ ] Avoid demand effects—question wording, framing, or sequence that signals the "right" answer.
• [ ] Pre-test the instrument for comprehension and for unintended cues before fielding it.

WHY / traps. Leading questions and demand effects generate confusion that exists only inside the instrument, and they are catnip for cross-examination because a juror can hear the problem read aloud. A question that asks, "Do you think the company that makes these candles also makes this diffuser?" has told the respondent that a candle company is lurking and that a "yes" is available. The Eveready design avoids this by asking the open, memory-based "Who puts out the product shown here? Why do you say that? What other products does that company put out?"—the exact wording an expert used, and the court accepted, in Denimafia. Open-ended questions cost more to code but they are far harder to attack, because they do not put the answer in the respondent's mouth.

The don't-know option is not a courtesy; it is a validity safeguard. Without a genuine "no opinion" escape, respondents who feel pressure to produce an answer will guess, and their guesses inflate the confusion rate. The Reference Manual on Scientific Evidence and the survey literature treat the availability of a "no opinion" response, and neutral probing, as hallmarks of a competent instrument. See Fed. Jud. Ctr., Reference Manual on Scientific Evidence (3d ed. 2011) (Shari Seidman Diamond, Reference Guide on Survey Research); 6 J. Thomas McCarthy, McCarthy on Trademarks and Unfair Competition §§ 32:158 et seq. (5th ed.).

Hypothetical (leading versus neutral). Leading: "These two sauces both use a flame logo—do you think they come from the same company?" That question supplies the very relationship it purports to measure. Neutral: after showing the lineup, "Do you believe any of the products shown are put out by the same company, put out by companies that are affiliated, or put out by companies with no connection—or don't you know?" The neutral version measures perception; the leading version manufactures it.

Phase 7 — Administration: Double-Blind and Validation

• [ ] Administer double-blind: neither the respondent nor the interviewer/administrator knows the survey's sponsor or purpose.
• [ ] For online panels, select respondents to represent the universe's demographics (age, gender, geography) and document the panel source and incentives.
• [ ] Validate online responses—confirm the invited person took the survey, screen out bots and speeders, and use identity/attention checks.
• [ ] Watch response rates; a very low rate can support an argument that a meaningful segment of the universe was missed.
• [ ] Preserve the complete field record: invitations, screener disposition, timestamps, and disqualification counts.

WHY / traps. Single-blind is not enough. An interviewer who knows the "right" answer can telegraph it through tone, pacing, or the emphasis of a probe—usually without any intent to cheat. Double-blind severs that channel by keeping the purpose from the person asking the questions as well as the person answering them. The T-Mobile survey is a clean model: the expert ran it double-blind, so that "the participants were not informed of the survey's purpose or sponsorship" and "the survey firm was not told of the study's sponsor or purpose." That structure is a routine part of the reliability showing.

Online administration removes some old worries and adds new ones. It eliminates interviewer fraud and transcription error, but it raises fresh validation questions—is the respondent the person who was invited, and is the respondent a human being at all? The answers are attention checks, identity confirmations (a date-of-birth question a bot cannot easily fake, panel-membership confirmation), speeder and straight-lining filters, and, where needed, traditional follow-up. Response rate can matter: in University of Kansas v. Sinks, 2008 WL 755065, at *4-5 (D. Kan. Mar. 19, 2008), a court criticized a survey in part because the email response rate was only 2.16%, raising the specter that the responders were unrepresentative. Panels generally fare better than cold email lists precisely because panelists have opted in and their demographics are known. On the science, properly run online surveys now perform as well as—or better than—mall and phone methods on response rate, representativeness, and comprehension, with the one documented tradeoff being somewhat shorter open-ended answers. See Hal Poret, A Comparative Empirical Analysis of Online Versus Mall and Phone Methodologies for Trademark Surveys, 100 TMR 756 (2010).

Phase 8 — Sample, Code, and Report

• [ ] Choose and justify the sampling method: probability sampling is the ideal; a well-executed, properly justified non-probability online panel or mall intercept is widely accepted.
• [ ] Size each cell (test and every control) for a usable margin of error; report cell sizes, margin of error, and confidence level (commonly 95%).
• [ ] Code open-ends objectively: a written coding protocol, coders blind to the hypothesis, and an inter-coder reliability check.
• [ ] Preserve and produce the verbatims, the questionnaire, the data file, and the code frame.
• [ ] Document the design rationale in writing contemporaneously, so the reliability showing exists before the challenge.
• [ ] Interpret the net figure in context: as a rough guide, a net around 15% and up tends to support confusion, below roughly 10% tends to negate it, and 10-15% is a genuine gray zone—always as one factor among the Polaroid / Sleekcraft / DuPont set.

WHY / traps. Sample size is where over-claiming meets arithmetic. Too few respondents per cell produces a margin of error so wide that the net figure could be almost anything, and an expert who reports a bare percentage without a margin of error is inviting the question how sure are you? to which the honest answer is not very. Report cell sizes and the margin of error up front; a net of 15% with a ±4-point margin is a finding, while the same 15% with a ±12-point margin is a coin flip dressed as science.

Coding is where quiet mischief hides. Open-ended answers must be classified—"confused," "not confused," "don't know"—and a single answer like "I'm not really sure, maybe the candle people?" can be scored as confused or as uncertain depending on who holds the pen. The defense is procedural and boring on purpose: a written code frame fixed before coding begins, coders blind to which cell and which hypothesis they are scoring, and an inter-coder reliability statistic demonstrating that independent coders reached the same classifications. Above all, preserve and produce the verbatims. A report that will not surrender its underlying responses is signaling that the responses do not say what the summary claims; opposing counsel will demand them, and a court will draw its own conclusions from resistance.

On interpretation, resist the temptation to treat any threshold as a bright line. Courts have credited nets in the high teens (the 18.8% in Cracker Barrel) and discounted lower figures, but the number is one input to a multi-factor legal test, not a verdict. A strong net confirms a strong case; it does not, by itself, decide likelihood of confusion, which the fact-finder weighs alongside mark strength, similarity, proximity of goods, intent, and the rest. For how those factors interact at the dispositive-motion stage, see Applying the Polaroid Factors at Summary Judgment; for the survey's place in the larger confusion picture, see Trademark Infringement and Related Rights. And because the survey expert's report must clear the disclosure and timing requirements of the Federal Rules, confirm the design is fixed in time to be disclosed—an untimely or incomplete expert disclosure can be excluded regardless of how good the science is. See Rule 37(c)(1) and the Preclusion of Undisclosed Evidence.

Surveys for Other Issues

Survey methodology is one discipline aimed at different questions. The universe, neutrality, control, administration, and coding standards do not relax when the issue changes from confusion to distinctiveness, fame, or genericness—only the design changes.

Secondary Meaning (Acquired Distinctiveness)

• [ ] Test whether a descriptive term or non-inherently-distinctive feature identifies a single source to the relevant purchasers.
• [ ] Control for respondents who would attribute any term or feature to one source (the "yea-sayer" and "single-source assumption" problems).
• [ ] Present the mark without brand-identifying context, so any single-source association is genuine and not spoon-fed.

A secondary-meaning survey asks, in essence, "when you see this, do you think of one company?" The T-Mobile magenta survey is the paradigm: respondents saw the color with no brand cues and were asked whether they associated it with a single wireless company; a net of 55% (after the brown control) and roughly half naming T-Mobile supported acquired distinctiveness. The control is doing essential work—separating true single-source recognition from a general human tendency to guess "yes, one company." For the doctrinal frame of what can function as a mark and how distinctiveness is proven, see What Can Be a Trademark.

Fame for Dilution

• [ ] Sample the general consuming public of the United States, tracking the statutory fame standard. 15 U.S.C. § 1125(c)(2)(A).
• [ ] Do not substitute a niche buyer universe; fame under the Trademark Dilution Revision Act requires broad, household-name recognition.
• [ ] Measure unaided recognition where possible, since aided recognition overstates fame.

Dilution fame is a demanding, near-celebrity standard, and its universe is deliberately broad. A mark famous only among specialty buyers is, by definition, not famous enough for dilution, so a survey confined to that niche measures the wrong population. See 15 U.S.C. § 1125(c); Jada Toys, Inc. v. Mattel, Inc., 518 F.3d 628 (9th Cir. 2007). For the modern application of dilution and fame to new contexts, see Trademark Doctrine for Virtual Goods and NFTs.

Genericness — Teflon Format

• [ ] Tutor respondents on the brand-versus-common-name distinction with clear examples, then verify they understand it.
• [ ] Have respondents classify a salted list—brand names, common names, and the challenged term—as "brand," "common name," or "don't know."
• [ ] Courts generally prefer Teflon for cleaner, more interpretable data.

The Teflon survey teaches the concept, tests comprehension, then asks respondents to sort a list. In the Teflon case itself, 68% classified TEFLON as a brand, defeating the genericness claim. See E.I. DuPont de Nemours & Co. v. Yoshida Int'l, Inc., 393 F. Supp. 502 (E.D.N.Y. 1975). Courts and commentators generally favor Teflon over Thermos for its cleaner output. See Genericness Surveys in Trademark Disputes: Evolution of Species, 99 TMR 1118, 1131 (2009).

Genericness — Thermos Format

• [ ] Describe the product generically without using the challenged term.
• [ ] Ask what respondents would call or ask for the product—capturing spontaneous generic usage.
• [ ] Add follow-ups to catch respondents naming the term only because it is their favorite brand.

The Thermos survey captures spontaneous naming. In the underlying litigation, when respondents were asked how they would ask for the type of container that keeps liquids hot or cold, 75% said "thermos" and only 11% "vacuum bottle," supporting a finding that the public perceived "thermos" as generic. See Am. Thermos Prods. Co. v. Aladdin Indus., Inc., 207 F. Supp. 9 (D. Conn. 1962), aff'd sub nom. King-Seeley Thermos Co. v. Aladdin Indus., Inc., 321 F.2d 577 (2d Cir. 1963). Its known weakness—respondents who name the term because it is their preferred brand rather than because it is generic—must be managed with follow-up questions. Note too that some courts limit genericness surveys to coined or arbitrary terms and disregard them for ordinary-word composites. See Nartron Corp. v. STMicroelectronics, Inc., 305 F.3d 397, 406-07 (6th Cir. 2002); Booking.com B.V. v. USPTO, 915 F.3d 171 (4th Cir. 2019), aff'd, 140 S. Ct. 2298 (2020).

Hypothetical (genericness). Suppose a maker of refillable draft-beer containers claims exclusive rights in "GROWLERO" and a competitor argues the term is generic. A Teflon survey would teach the brand/common-name distinction, then present a list—"KLEENEX," "stapler," "GROWLERO," "smartphone," "XEROX"—and ask respondents to classify each. A Thermos survey would describe "a sealable container you fill with draft beer at a taproom to take home" and ask what respondents would call it. If most spontaneously say "growler" or "growlero" as a category, the term is in genericness trouble; if most classify GROWLERO as a brand, it survives.

Surviving Daubert: Build for Admissibility from Day One

• [ ] Treat every design choice as a future exhibit at a Rule 702 hearing; document the rationale contemporaneously.
• [ ] Remember the proponent bears the burden to show, by a preponderance, that the survey was conducted according to accepted principles and is reliable.
• [ ] Anticipate that fundamental defects—wrong universe, no control, leading questions, unrealistic stimulus—increasingly draw exclusion, not mere discounting.
• [ ] Keep the verbatims, data, and protocol ready to produce; withholding them reads as consciousness of weakness.

WHY / traps. The admissibility of survey expert testimony runs through Federal Rule of Evidence 702 and the Daubert–Joiner–Kumho gatekeeping trilogy. See Daubert v. Merrell Dow Pharms., Inc., 509 U.S. 579 (1993); Gen. Elec. Co. v. Joiner, 522 U.S. 136 (1997); Kumho Tire Co. v. Carmichael, 526 U.S. 137 (1999). The trial judge has a "gatekeeping obligation" to ensure the testimony is both reliable and relevant, and the proponent carries the burden of showing the survey "was conducted according to generally accepted principles of survey research" and "is sufficiently reliable to be admitted." See Leelanau Wine Cellars, Ltd. v. Black & Red, Inc., 452 F. Supp. 2d 772, 777-78 (W.D. Mich. 2006), aff'd on other grounds, 502 F.3d 504 (6th Cir. 2007); NFL Props., Inc. v. N.J. Giants, Inc., 637 F. Supp. 507, 513-14 (D.N.J. 1986). The Manual for Complex Litigation distills the classic requirements: a proper universe, a representative sample, clear and non-leading questions, a qualified administrator using reliable procedures, accurate data reporting and analysis under accepted statistical principles, and objectivity throughout. Manual for Complex Litigation (Fourth) § 11.493.

For a long stretch, the reflex was that survey flaws "go to weight, not admissibility." That remains the general rule—Fortune Dynamic is a leading example of an appellate court sending survey criticisms to the jury as weight questions—but the safe harbor has narrowed. The December 1, 2023 amendments to Rule 702 made two things explicit that had drifted in practice: the proponent must establish each requirement by a preponderance of the evidence (it must be "more likely than not" that the standards are met), and the expert's opinion must "reflect a reliable application of the principles and methods to the facts of the case." Fed. R. Evid. 702(b)–(d). Courts already exclude surveys whose defects are fundamental rather than cosmetic. In Pinterest, Inc. v. Pintrips, Inc., 140 F. Supp. 3d 997, 1015-17 (N.D. Cal. 2015), the court excluded a survey where the value was substantially outweighed by its tendency to mislead—an outcome that also implicates the Rule 403 balance. On that balance, and how courts weigh probative value against the danger of misleading the jury, see Rule 403 and Unfair Prejudice. The practical upshot: a survey with the right universe, a real control, neutral questions, and a realistic stimulus is not merely more persuasive; it is more likely to reach the jury at all.

There is a jurisdictional wrinkle worth flagging. The Daubert framework governs federal court and, functionally, the TTAB; state courts vary, and some retain a Frye general-acceptance standard or their own codifications. If the case is in or may move to state court, confirm the governing admissibility standard before finalizing the design. And whether a survey question ultimately goes to the judge or the jury can turn on the legal-versus-equitable posture of the claim—see Judge or Jury in Trademark Litigation.

The Cross-Examination Preview

Before you finalize anything, script the cross-examination you would deliver against your own survey and make sure each answer is already built in. The examiner will move in the order of this checklist:

• Universe. "You surveyed people who buy the defendant's product, correct? But this is a reverse-confusion case—shouldn't you have surveyed people who buy the plaintiff's product?" If the theory and universe do not match, the survey is over before the second question.
• Screeners. "Your screener let in anyone who bought any watch in the last year—including a $30 watch—didn't it?" Loose screeners reopen the universe attack.
• Format. "The plaintiff's mark is famous, yet you chose the format that shows both marks side by side. Why not the format that relies on memory?" If the answer is that Eveready showed too little confusion, the survey is done.
• Control. "What was your control cell, and what was the net?" No control, or a strawman control, and the raw number evaporates.
• Stimulus. "You showed two logos on a white screen. Where in the real world does a shopper see that?" Unrealistic stimuli invite the Fortune Dynamic attack.
• Questions. "Isn't it true your question told the respondent that a candle company might be involved?" Leading questions read badly aloud.
• Blinding. "Did your interviewers know which answer you were hoping for?" Anything short of double-blind is a signaling risk.
• Sample and coding. "Your margin of error was ±11 points, wasn't it? And you won't produce the verbatims?" Thin samples and withheld data are the closing blows.

A survey built to answer all eight of those lines cleanly is a survey that will survive. For the mechanics of deposing (or defending) the survey expert who will field these questions, see Taking Depositions in Federal Civil Litigation; and where the survey record is being built in a TTAB proceeding, see Discovery Before the TTAB.

Common Mistakes (Quick-Reference)

• The wrong universe — the single most common fatal flaw; cannot be cured downstream.
• No control — the raw figure overstates confusion and is essentially uninterpretable.
• Wrong format for the mark's strength — Eveready on an obscure mark under-counts; Squirt on a famous mark over-suggests.
• A naked side-by-side — juxtaposition no shopper would ever encounter.
• Leading questions and demand effects — confusion manufactured inside the instrument.
• No "don't know" option — forces guessing and inflates the rate.
• Unrealistic stimuli — bare word marks divorced from trade dress and point of sale.
• Single-blind or unblinded administration — interviewer signaling.
• Tiny samples, unreported margins of error — a coin flip dressed as a finding.
• Subjective coding and withheld verbatims — mischief where classification happens; resistance reads as weakness.
• Implausible control — a strawman baseline the other side dismisses.
• Untimely or incomplete expert disclosure — good science excluded on a procedural default.

Primary and Secondary Authority

Rules and gatekeeping. Fed. R. Evid. 702 (as amended Dec. 1, 2023) & 703; Fed. R. Evid. 403; Daubert v. Merrell Dow Pharms., Inc., 509 U.S. 579 (1993); Gen. Elec. Co. v. Joiner, 522 U.S. 136 (1997); Kumho Tire Co. v. Carmichael, 526 U.S. 137 (1999); Manual for Complex Litigation (Fourth) § 11.493.

Confusion survey formats. Union Carbide Corp. v. Ever-Ready, Inc., 531 F.2d 366 (7th Cir. 1976) (Eveready); SquirtCo v. Seven-Up Co., 628 F.2d 1086 (8th Cir. 1980) (Squirt); Jerre Swann & R. Charles Henn Jr., Likelihood of Confusion Surveys: The Ever-Constant Eveready Format; The Ever-Evolving Squirt Format, 109 TMR 671 (2019).

Universe, control, and administration in practice. Bristol-Myers Squibb Co. v. McNeil-P.P.C., Inc., 973 F.2d 1033 (2d Cir. 1992); Kraft Foods Grp. Brands LLC v. Cracker Barrel Old Country Store, Inc., No. 13 C 0780 (N.D. Ill. July 1, 2013), aff'd, 735 F.3d 735 (7th Cir. 2013); Fortune Dynamic, Inc. v. Victoria's Secret Stores Brand Mgmt., 618 F.3d 1025 (9th Cir. 2010); T-Mobile US, Inc. v. Aio Wireless, LLC, 991 F. Supp. 2d 888 (S.D. Tex. 2014); Denimafia Inc. v. New Balance Athletic Shoe, Inc., 2014 WL 814532 (S.D.N.Y. Mar. 3, 2014); J.T. Colby & Co. v. Apple Inc., 2013 WL 1903883 (S.D.N.Y. May 8, 2013), aff'd, 586 F. App'x 8 (2d Cir. 2014); Univ. of Kansas v. Sinks, 2008 WL 755065 (D. Kan. Mar. 19, 2008); Pinterest, Inc. v. Pintrips, Inc., 140 F. Supp. 3d 997 (N.D. Cal. 2015); Leelanau Wine Cellars, Ltd. v. Black & Red, Inc., 452 F. Supp. 2d 772 (W.D. Mich. 2006), aff'd, 502 F.3d 504 (6th Cir. 2007); NFL Props., Inc. v. N.J. Giants, Inc., 637 F. Supp. 507 (D.N.J. 1986); Playboy Enters., Inc. v. Netscape Commc'ns Corp., 55 F. Supp. 2d 1070 (C.D. Cal. 1999); Eagle Snacks, Inc. v. Nabisco Brands, Inc., 625 F. Supp. 571 (D.N.J. 1985); Schering Corp. v. Pfizer Inc., 189 F.3d 218 (2d Cir. 1999).

Secondary meaning, fame, and genericness. E.I. DuPont de Nemours & Co. v. Yoshida Int'l, Inc., 393 F. Supp. 502 (E.D.N.Y. 1975) (Teflon); Am. Thermos Prods. Co. v. Aladdin Indus., Inc., 207 F. Supp. 9 (D. Conn. 1962), aff'd, 321 F.2d 577 (2d Cir. 1963) (Thermos); Nartron Corp. v. STMicroelectronics, Inc., 305 F.3d 397 (6th Cir. 2002); Booking.com B.V. v. USPTO, 915 F.3d 171 (4th Cir. 2019), aff'd, 140 S. Ct. 2298 (2020); Jada Toys, Inc. v. Mattel, Inc., 518 F.3d 628 (9th Cir. 2007); 15 U.S.C. §§ 1114, 1125(a), 1125(c).

Reference works. Fed. Jud. Ctr., Reference Manual on Scientific Evidence (3d ed. 2011) (Shari Seidman Diamond, Reference Guide on Survey Research); 6 J. Thomas McCarthy, McCarthy on Trademarks and Unfair Competition §§ 32:158 et seq. (5th ed.); Hal Poret, A Comparative Empirical Analysis of Online Versus Mall and Phone Methodologies for Trademark Surveys, 100 TMR 756 (2010); Genericness Surveys in Trademark Disputes: Evolution of Species, 99 TMR 1118 (2009); Int'l Trademark Ass'n survey guidelines.

Survey design and admissibility are intensely fact-specific. Retain a qualified survey expert and experienced trademark litigation counsel for any particular matter.

Related UpLaw Resources

Related Guides

• Reading and Building Trademark Confusion Surveys: A Practical Guide to Expert Methodology
• Daubert and the Consumer Survey: Excluding Expert Testimony in Trademark Cases
• Likelihood of Confusion in Trademark Law: A Brand Owner's Guide
• Applying the Polaroid Factors at Summary Judgment in the Second Circuit
• Standards of Review in Second Circuit Trademark Appeals
• Rule 403 and Unfair Prejudice: The Balancing Test
• What an IP Case Is Worth: A Guide to Intellectual Property Damages

Related Checklists

• Working Up Likelihood of Confusion: A Practical Checklist
• Discovery Before the TTAB: A Practical Checklist

Related Toolkits

• The Trademark Enforcement and Litigation Lifecycle: From Watching to Appeal

Related Articles

• Trademark Infringement and Related Rights: The Enforcement Landscape
• Trade Dress Protection: Where Brand Identity Meets the Functionality Bar
• What Can Be a Trademark: The Subject Matter of Trademark Law
• Trademark Doctrine for Virtual Goods and NFTs: The MetaBirkins Landscape
• Judge or Jury in Trademark Litigation: The Legal-Equitable Divide
• Rule 37(c)(1) and the Preclusion of Undisclosed Evidence
• Taking Depositions in Federal Civil Litigation

This checklist is general legal information from UpLaw (uplaw.us), not legal advice, and does not create an attorney-client relationship. Consult qualified trademark litigation counsel and a qualified survey expert about any particular matter.