Check your marketing claims before regulators do

The launch page is in review. The headline leads with a number, the second section says the product is "powered by AI", and a line about health in the FAQ was borrowed from a competitor's site because it sounded right. Legal wants to know what backs each one. This page shows how to test the claims in your copy against sources before you publish, and how to turn the ones that fail into wording you can stand behind.

Regulators expect the evidence first

US advertising law has asked for this for decades. The FTC's 1984 policy statement says advertisers need a reasonable basis for advertising claims before they are disseminated, not after someone asks. What's new is the attention on AI claims, and the agency has been turning its warnings into binding orders.

In April 2025 the FTC said Workado had advertised its AI content detector as 98 percent accurate, while independent testing put its accuracy on general-purpose content at 53 percent. The final order, approved that August, bars accuracy claims unless the company has competent and reliable evidence for them. In August 2026 the agency finalized consent orders against Cox Media Group, MindSift and 1010 Digital Works over an "Active Listening" service, sold as an algorithm that picked up conversations from smart devices to target ads. The FTC said the service wasn't based on voice data at all. The three firms paid $930,000 between them.

The Workado case carries a lesson that reaches well beyond AI. According to the FTC, the model behind the detector had been trained or fine-tuned to classify academic content, yet it was sold for general use. A number that holds in one setting can mislead in another.

How a check works on a marketing claim

Paste a claim exactly as it appears in the copy. For a full landing page, brochure or sales one-pager, upload it as a PDF or Word file, or paste the text, and CiteJury lists every checkable claim so you can pick up to 25. Our page on checking a whole document covers that flow.

Each claim is split into its parts. CiteJury searches several independent search indexes for evidence for and against each part, favoring primary sources such as regulators, government statistics and official pages. The best passages are sealed into a single evidence pack. Claude, ChatGPT and Grok read that same pack and answer separately, every quote they use is checked word for word against its source, and a final review writes the verdict: Correct, Incorrect, Sources disagree or Can't confirm.

The part marketers use most is the safe way to say it: a sentence that says only what the sources support. It isn't finished copy, but it marks the boundary your copy has to stay inside. Here's a check of a claim you've seen on a hundred wellness pages.

The eight glasses claim

The claim: "You need to drink 8 glasses of water a day to stay healthy." It's the kind of line that gets into copy because everyone has heard it, so nobody asks for a source.

CiteJury split it into three parts and searched 96 results. The sources it found included Harvard Health, Dartmouth, the University of Rochester Medical Center, Tufts Medicine, McGill and PubMed. Pages it couldn't read, including Reddit and Facebook posts, were set aside before the AIs read anything, and 12 sources were locked into the evidence pack. Claude, ChatGPT and Grok all said Incorrect, and none of the three parts held.

The reason, in the result's own words: no scientific basis for "8 glasses", and food and other drinks count too. The suggested wording: "Most healthy adults can drink to thirst. Eight glasses is a rule of thumb, not a requirement." The check took under three minutes.

For a hydration brand, that isn't a dead end. It tells you what you can't say (that eight glasses is a health requirement) and what you can build on: drinking to thirst, and water from food and other drinks counting toward the day. A copywriter can work with that. What they shouldn't do is keep the old line and hope nobody checks.

Many claims that fail look like this one. They aren't invented. They're a familiar idea stated with more certainty or precision than the evidence carries, and the fix is usually to narrow the claim, attribute it, or drop the number.

Checking copy before it ships

1. List the objective claims. Numbers, comparisons ("twice as fast"), "clinically proven", "#1", health and environmental benefits, and anything about what your AI features do.
2. Paste each one as written, or upload the draft and pick from the claims CiteJury finds. Wording matters: "helps support" and "prevents" are different claims.
3. Choose a tier. Standard runs all three AIs plus a final review, for about 8 credits ($0.80), and usually takes one to three minutes. Use Deep for health, safety or environmental claims, where a wider search is worth the extra time. Quick, with one AI, is fine for an early draft.
4. Read the parts, not only the verdict. A claim can be half right, and the part that fails is the part to rewrite.
5. Rewrite from the safe wording. Keep your voice, stay inside what the sources support, and check the new line again if it makes a new claim.
6. Keep the record. Download the sources as Excel or CSV and file them with the approved copy, or share the check's link with legal. Anyone with the link sees the check and its sources.

A claims checklist for marketing teams

• Survey statistics: who ran the survey, when, how many people answered and who paid for it. Our guide to tracing a statistic walks through it.
• Comparisons: name what you're comparing against and on what measure. "Faster" needs a "than" and a test.
• AI features: describe what the feature does, not what AI could do in general. Both FTC cases above turned on claims about what a product could do.
• Borrowed numbers: a figure measured on one group, product or setting doesn't transfer to yours without evidence.
• Your own test results: if a claim rests on internal data, CiteJury won't find it, so expect Can't confirm. That's your cue to make sure the study itself is in your substantiation file.
• Evergreen copy: sources change. Check old claims again before a relaunch, not just the new ones.
• Superlatives: "leading", "#1" and "most trusted" each need a source and a scope, or they need to go.

What it won't do

CiteJury isn't legal advice. It tells you whether a statement holds up against the sources it found. It doesn't tell you whether a claim meets the FTC's standard, the UK's advertising codes or any other rulebook, and it doesn't judge whether a line is puffery or an implied claim. Those are calls for your lawyer or compliance team, and a sourced check gives them something concrete to review.

• It checks text, not images or video. A claim made by a before-and-after photo or a product video isn't checked, so paste the on-screen text or voiceover script instead.
• It can't see your internal tests, customer data or unpublished studies.
• Paywalled journals and sites that block automated access can't be read. The result lists what it couldn't open.
• A verdict is only as good as the sources found. Results can be wrong or incomplete, which is why every sentence links to its source.

Checking a claim before launch takes a few minutes. Rewriting it after a customer complaint or a competitor's challenge takes much longer. For claims about laws and rules themselves, see checking regulatory claims.

Questions people ask