Why we publish our rule list
Most "compliance checkers" return a score and a colour. That design has a problem: a score you cannot inspect is a claim, and this entire product exists because uninspectable claims are the problem. So the scanner is the opposite of a black box — every check is a documented rule, every flag quotes the exact phrase it matched, and every rule links the policy source it derives from. The ruleset carries a version (currently 2026.08.2, reviewed 19 August 2026) so a scan is reproducible: the same text and the same ruleset version always produce the same result.
This post is the full transparency exercise: all sixteen checks, what each one flags, and the published policy behind it. If you disagree with a rule, you can read the same source and judge.
How a scan runs
The scan is a deterministic text pass, not a model opinion. Each rule is a pattern with a severity (high, medium or low), applied to the copy you generated or pasted. When a rule matches:
- The exact phrase is quoted, not paraphrased.
- The rule and its reason are shown.
- The source is linked, so the policy text is one click away.
- Channels matter: some rules apply everywhere, others only to the channel being written — Amazon listing, Shopify description, SEO metadata, ad copy or email.
A human still decides what to do. The scanner surfaces risk; it does not pretend to be a lawyer.
The universal claim rules
These apply on every channel, because the underlying policies apply to any commercial text.
1. Ranking or popularity claims (high) — "#1", "best-selling", "top-rated", "award-winning". Amazon prohibits promotional ranking language in titles outright, and the FTC's substantiation standard requires popularity claims everywhere else to carry current evidence. Rank is volatile; the words age badly.
2. Unverifiable social proof (high) — "over 10,000 happy customers", "4.9 out of 5". Under the FTC's reviews and endorsements guidance, these are endorsements that must reflect real, documentable data. If the number cannot be produced from your own records within the hour, it does not belong in the copy.
3. Regulated health or medical claims (high) — "cures", "treats", "prevents", "anti-bacterial", "FDA approved", "clinically proven". The FTC's Health Products Compliance Guidance sets the bar: competent and reliable scientific evidence for health claims. One match here can move a general product into a regulated category — the single highest-severity pattern the scanner looks for.
4. Absolute guarantees (medium) — "lifetime guarantee", "100% satisfaction", "risk-free". A guarantee must match what you actually honour, with material terms nearby. Absolute wording attached to conditional policy is the classic pattern.
5. Scarcity and urgency claims (medium) — "only 3 left", "offer ends today", "last chance". Manufactured urgency is deceptive when the constraint is invented; real constraints need to be real, and dated.
6. Unsupported superlatives (high) — "world-class", "best on the market", "unmatched". Superlatives are comparative claims without a comparison, and comparisons need evidence.
7. Reference price and discount claims (medium) — "was $49, now $29", "50% off". The FTC's Guides Against Deceptive Pricing make a fictional "original" price a deceptive reference price. The rule flags the pattern so the bargain is anchored to something real.
8. Evidence-dependent safety and certification claims (medium) — "CE certified", "UL listed", "food-safe". Certifications are checkable facts: either the certificate exists for this product or the words do not belong in the copy.
9. Before-and-after and transformation claims (high) — "look 10 years younger", "from stained to spotless in one wash". Transformation narratives are regulated advertising claims on ad platforms in particular, and rarely substantiable for general merchandise.
10. Implied personal attributes (high) — copy that implies the reader has a negative attribute ("are you balding?", "tired of being overweight?"). Prohibited or restricted under Meta's Advertising Standards and disliked everywhere else.
The marketplace rules
11. Competitor or trademark references (medium) — naming another brand, or "compatible with [brand]" phrasing that implies endorsement. Amazon's title requirements restrict brand claims to brands you are authorised to use; a match here prevents an IP complaint before it exists.
12. Off-platform contact and links (high) — URLs, emails, phone numbers, QR codes, "contact us for a better price". Under Amazon's seller conduct policy, diverting buyers off the platform is a violation, and because it is machine-detectable, it is also one of the most-enforced rules on the platform.
13. Vague emotional filler (low) — "experience the difference", "unleash your potential", "elevate your lifestyle". Not a legal risk; a copy quality flag. Filler words spend the reader's attention and signal that a specific, checkable fact was available and not used.
The UK and EU rules
Three rules exist because UK and EU advertising law is stricter than the US baseline, and copy written for those markets is checked against it.
14. Unsubstantiated proof claims (high) — "clinically proven", "scientifically tested", "dermatologically tested". Rule 3.7 of the UK CAP Code, enforced by the Advertising Standards Authority, requires advertisers to hold documentary evidence for objective claims before the ad runs — and ASA rulings routinely find the evidence did not test what the copy claimed.
15. Unqualified environmental claims (medium) — "eco-friendly", "green", "sustainable", "climate neutral". Directive (EU) 2024/825 bans generic environmental claims unless backed by recognised excellent environmental performance, applying from 27 September 2026. The EU green claims framework is the reference; the fix is always specificity ("box made from 92% recycled cardboard") over adjectives.
16. Unsubstantiated lowest-price claims (high) — "cheapest price guaranteed", "lowest price in the UK". A price-superiority claim requires the comparison to be provable at the moment the ad runs — under both the CAP Code and EU consumer law, this is among the most-complained-about claims in small-business advertising.
What the scanner deliberately does not do
Boundaries, stated honestly: the scanner reads text. It does not inspect images, does not evaluate restricted-product status, knows nothing about your account history, and does not review landing pages. A clean scan means the text carries no flagged claim patterns — not that the listing is legally cleared. Human review stays in the loop because policy pages end with the same sentence: the seller is responsible.
Rulesets decay, so they carry a version
Marketplace policy changes — the 75-character title rule arrived with two months' notice in 2026, and the EU green-claims directive applies from a fixed date. Each rule records the date it was last reviewed against its source, and the ruleset version is printed with every scan result. When a policy changes, the rule changes, the version increments, and the change is visible — the same standard we ask sellers to hold their copy to.
Run a scan yourself
The fastest way to see all of this working is to paste a listing into the compliance checker and read what comes back: the flagged phrase, the rule, the severity, and the source link. No account needed — and if a flag ever seems wrong, the source is one click away, which is the point.
Written by Nabeel Ali
Brand designer — 12+ years, 50+ brands across 8 countries. Founder of CopyForge AI.