Claims Substantiation Standards for Comparative Efficacy Statements

What counts as "competent and reliable scientific evidence"

The FTC defines this standard as "tests, analyses, research, studies, or other evidence based on the expertise of professionals in the relevant area, that have been conducted and evaluated in an objective manner by persons qualified to do so, using procedures generally accepted in the profession to yield accurate and reliable results." That's a mouthful, but it flexes depending on what's being claimed. Claims about efficacy, well-being, and risk demand robust substantiation under FTC standards. Basic performance claims may at times rely upon one well-designed consumer survey, but no real shortcut applies to comparative claims, while advertisers often go wrong when treating those that way.

Comparative claims must meet rigorous substantiation requirements. A claim needs tests, studies, actual proof to rest on, not just what some marketing folks feel about it. The claim's wording must stay within what the testing proves. Each comparison must measure today's offering against today's rival, not an outdated version versus the newest competitor offering. It must choose the nearest actual competitor for sale, one that gives a true test instead of one chosen because it's simple to top. They require direct side-by-side tests, not a pair of solo product trials joined into something they were never meant to show.

Hedged language offers no protection, and businesses that rely on it are off base. "May help" or "helps achieve" still misleads if the weight of the evidence points the other way. Softening wording near a verb doesn't remove any unsupported implication, so the FTC doesn't judge by well-meaning effort. When evidence is weak, conflicting, or just not there outright, the gap must still be disclosed. Omitting opposing results is itself a deceptive act, however cautious the other copy seems.

How research is planned matters as much as doing it. The setup must match how customers really handle the item. Methodology needs to match protocols the field already uses, and any study set up to yield a favorable result should be dropped before a copywriter sees it. In effect, then, the result is one substantiation file maintained alongside the copy, matching every claim to its supporting research, claim by claim.

Where NAD results usually show problems with substantiation files

Holding evidence doesn't equal holding evidence that fits this claim. That shortfall explains most of the NAD rulings against marketers lately, even when some proof is on hand.

Cox Communications hit this issue over "unbeatable 5G reliability" in Cox Mobile Services ads. The claim needed more than Speed-test results, since reliability and fast rates feel distinct to any consumer. NAD required Cox to ensure its testing matched real-world consumer experience, rather than relying solely on throughput from a controlled lab setting. The point reaches far beyond telecom: proof must fit the exact consumer experience a claim describes, never whichever nearby metric runs cheapest to check.

Cebria LLC ran into a similar yet distinct problem when promoting Cebria Supplements through its claims. Cebria's supplement beat placebo with statistically clear results, yet NAD turned that claim down because the company offered no proof the gain was real to the audience the ad addressed: older users worried about age-related recall. Data might pass the p-value test yet mean nothing to someone looking at the copy. NAD treats significance and relevance as two different tests, judged apart. When advertisers conflate the two, they wager on a line NAD has dismissed time and again.

NAD's files show the same mistakes over and over: picking a bad rival to measure against, pushing an ad beyond what the research proved, or using one simple figure to represent something with many parts. All of it is easy to repair, though carrying it out is tedious. Match each term in a comparative claim to evidence inside the substantiation file before finalizing copy, not once NAD's notice arrives.

The disclosure obligation that runs alongside substantiation

Substantiation plus disclosure remain two separate requirements under the FTC Act; many businesses can satisfy one and quietly miss the other. Under Disclosure, facts that could shift a consumer's take on any claim must be surfaced if omitting them would mislead people. A claim becomes material whenever it suggests more confidence than the proof can back up. Even with precise wording, research that has problems won't back up any unqualified superiority claim.

A cholesterol supplement case shows the failure mode: the advertiser had two limited studies and a stack of contradicting literature, then leaned on "may be effective in reducing cholesterol" as if the qualifier did the work. The advertiser had two limited studies and a stack of contradicting literature, then leaned on "may be effective in reducing cholesterol" as if the qualifier did the work. The qualifier couldn't do the job. That disclosure had to list the exact problems right outright, and say more work was required before anyone could trust that claim.

Disclosures must pass a style test, not just a message test: conspicuous and plain enough that an ordinary reader understands without decoding any phrasing. Even a well drafted disclosure won't rescue the main claim when an ad's overall impression gets that reader to trust what it quietly contradicts. The full ad governs. A footnote doesn't have the last say.

Most marketers don't realize how far endorsements and influencer posts push this duty. The brand backs adopted claims, not just ones invented from scratch. A bright write-up or influencer chatter can lead to an implied superiority claim a brand must back up just like it had said it directly. The brand must give its paid relationship with the poster a separate disclosure. Transformation pictures matter just as much: they act like performance promises and require the same proof as written text making that point.

AI performance claims face the closest look in today's enforcement.

September 2024 is when the FTC rolled out Operation AI Comply, its coordinated sweep targeting firms that hide deceptive practices behind AI language. Those companies reveal the FTC's current priorities.

DoNotPay called itself "the world's first robot lawyer." The FTC alleged the service gave inaccurate legal advice without the expertise of an actual attorney, and the final order landed on $193,000 in monetary relief plus a ban on claiming parity with a real lawyer absent solid evidence behind it. Other firms in that sweep faced scrutiny for deceptive AI-related claims.

This rule existed before, so nobody ought to act like it just arrived. When an ad says something works better, thinks better, or does more since AI sits behind it, the company must back this sort of AI-driven promise with the same solid, matching, buyer-facing proof the law has long demanded. The tools have evolved. What governs it hasn't.

The difficulty in substantiating AI claims comes baked right into how the models operate. Since Output changes with each query, account, and release, any benchmark pulled during a single trial might not match what regular folks experience, just like the gap NAD flagged linking Cox's numbers to its reliability claim. The brand's responsibility doesn't end there. Advertising agencies share responsibility for deceptive claims alongside their clients. The obligation to prove any AI performance claim falls on the person who drafted that sentence, and on whoever's name appears above it.

A novel substantiation question: when an AI engine makes the comparative claim for you

Generative AI engines pull from across the web and rank brands against each other ("Brand X is better for Y use case") without the brand writing a word of that comparison itself. This creates a dilemma that didn't exist half a decade back.

Citations in AI engines cluster and shift at the same time. Research on 366,000 citations reveals variability in AI engine outputs and citation patterns. In 2025 AirOps reviewed 45,000 citations and reported that a mere 30% of brands appear in repeated AI responses to an identical query, with just 20% holding on through five repeated attempts at that same query. Since these systems rebuild from scratch every run, a favorable citation now guarantees zero for later.

This uncertainty sharpens the core issue: if AI produces a comparison regarding a company, and that company then uses the reference in its own ads, does standard claim substantiation still apply? The FTC has not yet addressed this specific scenario in its guidance. Yes, it still should. The FTC requires AI-related claims to be truthful and substantiated, regardless of whether the comparison originates from the brand or an external source.

Before it quotes or screenshots an ad’s AI-generated list, a brand must face this blunt test: could it back its own comparison with proof? Quoting it anyway exposes a brand just as much as drafting the claim themselves. Citation volatility creates another risk: a comparison true when a brand saved it might not reflect current outputs, the same timeliness concern governing ordinary comparisons with outdated models.

Fitting substantiation oversight across large multi-brand publishing workflows

Agencies remain liable, right alongside the clients, over deceptive claims that they create and share. A client agreement won't quietly shift that exposure. The risk remains with the ad's creator.

Volume doesn't alter the standard. It alters what the danger looks like. If AI generates comparative language for 20 brands during one day, every sentence holds an identical FTC plus NAD substantiation obligation that any copywriter faces when typing manually. Automation increases volume. It doesn't manufacture support for that output, and Failing to substantiate claims properly can lead to NAD challenges for multi-brand agencies.

Multi-brand agencies may face challenges when workflows lack coordination across teams and systems. An unsupported comparative claim slips by unnoticed whenever such fragmentation exists.

Checks before release fix the problem, and they’re simple. Before attorneys touch it, each version gets sorted as either superiority or parity or monadic. A comparative claim publishes only with a substantiation file that names the trial, when and how it ran, plus the exact claim covered. The proof is tested against the latest releases of the client's and competitor's offerings. Each statement is tested for buyer importance as well as measured against the control. The completed ad also goes through a net-impression check, visuals too, in order to confirm it all stays within what that file can back up.

That still holds when AI produced the copy, not a human. It makes no difference to the FTC who has typed that sentence, since the firm behind the ad bears responsibility. AI visibility for each client roster needs a way to track which claims keep surfacing within AI-generated output, using per-client reporting and granular controls so account teams can catch comparative language early, before it compounds over many accounts. Those records give staff proof if someone questions a claim down the road.

Building a substantiation-ready process before comparative claims go to market

Substantiation must exist before the ad runs. It must be in place before the ad goes out, and can't be assembled once a complaint lands. This point should shape how teams set up ad writing, yet most do it in reverse, seeing proof as forms to collect later instead of a step every message must clear before anything else.

Lawyers, marketing staff, plus study teams each handle part of it, and most failures come from leaving one of them out during drafting. A list for checking claims makes that joint work real. Begin with naming each stated point and implicit message made by the ad, plus what its pictures and setting tell an ordinary reader. Sort the claims and put each one to the right level of proof. For comparison ads, check that side-by-side proof is on file, fits the item as sold, names the proper rival, and reflects everyday practice. Set statistical significance beside consumer relevance, because a finding can matter in data but be meaningless for consumers and still fail at NAD. Spell out each disclosure duty by pinning down what missing details would do that, and make sure no disclosure gets subtly undermined by the ad's overall feel. If brand's advertising has amplified any AI-generated comparison, count that citation as its own claim and use an identical standard.

Keeping comparative claims fresh doesn't stop once they're live. Substantiation records require updating if one item in the comparison alters materially, since Competitor offerings and formulas evolve. Keeping good records carries the same weight: put the file together as the claim takes shape, rather than scrambling to assemble it once a problem shows up.

Agencies that teach account teams these rules, rather than handing over a checklist for rubber-stamping, become the go-to partner clients rely on for defensible marketing in the AI era. It cuts risk above all. Buyers who already understand that danger see a clear advantage too.