Inside the Compliance Scanner: Every Check and the Policy Behind It

Why we publish our rule list#

CopyForge does not treat a pattern score as a black box. The current registry version, rule count and recorded review date are shown in the summary above, read directly from the scanner registry. The dated explanations below preserve the release history; the linked live registry shows current scope. Every rule has a severity, channel scope, review/effective date and policy or guidance source. When a rule matches, CopyForge returns the exact matched phrase and the source behind that rule.

The canonical live list is also rendered on the About page from the same rule registry the scanner executes. That matters because scope is part of the rule: an ad-only check is not silently advertised as Amazon coverage.

How a scan runs#

A scan is a deterministic text pass, not a model opinion. For the selected channel it loads only rules whose declared channel scope includes that channel. A match returns the rule id, label, severity, every bounded occurrence with exact offsets, deterministic context classification where supported, source link and guidance. Results also report the engine version, registry version, applied rule IDs and whether bounded coverage completed. The same text, channel and versions produce the same deterministic core matches.

A human still decides what to publish. The scanner does not inspect images, product eligibility, account history or every category-specific requirement, and it is not legal advice or marketplace approval.

Rule library explained — 15 October 2026 snapshot#

This dated snapshot describes ruleset 2026.10.15 with 19 deterministic text rules. Use the live rule list for the current registry and channel scope.

1. Ranking or popularity claim · high · all channels Flags patterns such as “#1”, “best-selling”, “top-rated” and similar popularity/ranking wording. Amazon title policy and advertising substantiation are the source context.

2. Unverifiable social proof · high · all channels Flags unsupported review, rating, customer or sales-count language. The source is FTC reviews and endorsements guidance.

3. Regulated health or medical claim · high · all channels Flags disease, treatment and health-outcome wording that needs appropriate substantiation and may trigger additional regulation. The source is FTC Health Products Compliance Guidance.

4. Regulatory action or endorsement claim · high · all channels Separately flags “FDA approved”, “FDA-cleared”, “FDA registered”, “FDA tested”, “FDA certified”, “FDA endorsed” and equivalent configured wording. Registration or inspection is not automatically product approval or endorsement.

5. Antimicrobial or public-health claim · high · all channels Flags wording such as “antimicrobial”, “disinfects”, “antibacterial” or “kills germs” for review against the applicable EPA registration, treated-article or market-specific rules.

6. Absolute guarantee · medium · all channels Flags absolute guarantee wording such as “risk-free”, “never fails” or “100% satisfaction” unless the written policy really supports the promise and its limits.

7. Scarcity or urgency claim · medium · all channels Flags static or absolute scarcity/deadline language that must be real and current when displayed.

8. Unsupported superlative · medium · all channels Flags objective comparative superlatives such as “best on the market”, “best on Amazon”, “unbeatable” or “world-class” when they imply evidence against a market.

9. Absolute durability or lifetime-performance claim · medium · all channels Flags wording such as “unbreakable”, “lasts forever” or “never wears out”. State a tested specification or real warranty instead.

10. Reference price or discount claim · medium · all channels Flags former-price and discount patterns that need a genuine, appropriately recent basis.

11. Evidence-dependent safety or certification claim · medium · all channels Flags claims such as CE certification, UL listing, BPA-free, food-grade and similar evidence-dependent statements.

12. Competitor or trademark reference · medium · Amazon Flags selected competitor/compatibility wording that can create accuracy or intellectual-property risk in an Amazon listing.

13. Off-platform contact or link · high · Amazon Flags URLs, email addresses, phone/WhatsApp directions and similar diversion text in Amazon listing copy.

14. Vague emotional filler · low · Amazon A copy-quality heuristic for low-information wording. It is not presented as a legal conclusion.

15. Before-and-after or transformation claim · high · Ads An ad-channel rule for before/after and specific transformation framing. It is not advertised as an Amazon rule.

16. Implied personal attribute · high · Ads An ad-channel rule for wording that asserts or implies sensitive or personal attributes about the viewer.

17. Unsubstantiated proof claim · high · Ads, Shopify, SEO, Email Flags “clinically proven”, “scientifically proven”, “lab tested” and related proof language under UK CAP/ASA substantiation principles.

18. Unqualified environmental claim · medium · all channels Flags generic environmental claims such as “eco-friendly” or “climate-neutral” where qualification/substantiation is required. The rule records its own effective date.

19. Unsubstantiated lowest-price claim · medium · Ads, Shopify, SEO, Email Flags “cheapest”, “lowest price”, “best price” and similar price-superiority wording that must remain provable when shown.

What the result states mean#

Review required means one or more configured patterns need human review. No configured patterns found means the applicable configured rules found no match; it is not clearance. Check partially completed means a bounded match limit was reached and the result is incomplete. The pattern score is a secondary summary, not a probability, accuracy rating or compliance certificate.

Direct negation and explicitly introduced quoted instructions or examples can be labelled as context review. Quotation marks alone do not clear a marketing claim, and qualified claims such as “may cure” retain their evidence-review severity. Complex grammar can still defeat a deterministic context check, so the surrounding sentence remains a human decision. Images, restricted-product status, landing pages, evidence quality and account history sit outside this text layer.

Rulesets change#

Each saved scan carries a ruleset version. When a rule changes, CopyForge increments the version rather than quietly changing the meaning of an old result. Product and Assurance workflows can therefore tell when stored content was scanned under an older registry and needs review again.

Reproduce six boundary cases#

The controlled examples below show exact inputs and results computed by the current scanner. They isolate a direct health claim, an invisible separator, direct negation, quotation and a health-related phrase the patterns miss, alongside a product-details control. Copy one input into the Amazon checker and compare its matched phrase, severity and context.

This is a deliberately selected demonstration set, not a representative dataset or a customer case study. It provides no measured accuracy percentage, sales result or marketplace approval. The example with zero matches and unsupported health wording is included to make the coverage limitation visible.

Run the channel you actually publish to#

Use the Amazon listing compliance checker for Amazon-scoped checks, or generate inside CopyForge with the target channel selected. The source-linked live rule list is the authoritative public scope for the current release.

A channel-scope example to reproduce#

Paste Lowest price for this ceramic mug. into a channel-specific scan. Under registry 2026.10.15, Shopify includes the uk-lowest-price rule; Amazon does not include that rule. This is a constructed scope test, not a claim that Amazon permits price-superiority wording. Amazon has its own title/listing rules and other patterns may apply to a different sentence.

Record the input, selected channel, applied rule IDs and matched rule IDs. A result saved without the channel can mislead the next reviewer. Our regression test checks this exact sentence against the executable registry; the live rule list shows the declared scope.

Scanner implementation update — 15 October 2026#

Ruleset and engine 2026.10.15 retain the 19-rule library and its last full policy-source review date of 2 October. This implementation update corrects ranking coverage across channels, clinical/regulatory wording, Meta personal-attribute matching and context handling. Quotation marks, an unrelated “no”, or a qualifier such as “may” do not clear an affirmative objective claim. Only explicit reference examples and direct negation receive the contextual lower-priority receipt.

Specialist claim-check scores apply one penalty per distinct matched rule and keep the most severe occurrence. Thin Shopify descriptions and single-word Amazon titles receive copy-completeness warnings. These are drafting heuristics, not marketplace minimum-length rules. A short-passage notice distinguishes missing detail from a clean risk-pattern result. No score verifies the truth of the product details or predicts sales.

Primary sources and heuristics#

The FTC substantiation statement, FTC health guidance, Amazon title requirements and Meta ad standards explain important source contexts. Rule 14 is a CopyForge copy-quality heuristic, not an external policy requirement. The live registry distinguishes source context from our check activation date.