// execute

Claims & Proof Check

Use when the user wants to check marketing copy for claims that need proof or permission before it is published - a landing page, ad, email, deck or case study. Also use when the user mentions claim substantiation, "can we say this", proof points, testimonials, "HIPAA compliant" or "SOC 2" badges, AI accuracy claims, "clinically proven", superlatives like "#1" or "best", or FTC advertising rules. Flags every claim sentence by sentence, with what evidence or consent to keep on file and the regulator's own guidance.

ExecuteengineStarter
bash
$npx skills add sarojkjha/aaj-marketing-skills --skill claims-and-proof-check
Or run it in Claude or ChatGPT through the AAJ engines connector

Claude Code

/plugin marketplace add sarojkjha/aaj-marketing-skills
/plugin install aaj-marketing-skills@aaj

OpenAI Codex

codex plugin marketplace add sarojkjha/aaj-marketing-skills
codex plugin add aaj-marketing-skills@aaj

Claude or ChatGPT — no install

Add this as a custom connector:

https://aaj-engines-mcp.sarojkjha10.workers.dev/mcp
Recipes: vendor data in, a verdict out

// when to use it

Use when the user wants to check marketing copy for claims that need proof or permission before it is published - a landing page, ad, email, deck or case study. Also use when the user mentions claim substantiation, "can we say this", proof points, testimonials, "HIPAA compliant" or "SOC 2" badges, AI accuracy claims, "clinically proven", superlatives like "#1" or "best", or FTC advertising rules. Flags every claim sentence by sentence, with what evidence or consent to keep on file and the regulator's own guidance.

// what you give

The copy to check, the audience (general, B2B SaaS, AI, EdTech or MedTech), and any evidence or permissions already on file

// what you get

A verdict per sentence (needs proof, needs consent, review wording, OK), every flagged claim quoted with why it needs support, what to keep on file and the regulator's guidance link, and a summary count

Engine spec

// decision it supports

Which claims in this copy need proof, consent or a wording change before it goes live?

// inputs

The copy to check, the audience (general, B2B SaaS, AI, EdTech or MedTech), and any evidence or permissions already on file

// output

A verdict per sentence (needs proof, needs consent, review wording, OK), every flagged claim quoted with why it needs support, what to keep on file and the regulator's guidance link, and a summary count

// acts or analyzes

Analyzes only

Makes no network calls, writes no files and changes nothing in your tools.

// where your information goes

How you run itWhere your inputs go
Skill installed in your agent, or the local MCP serverThe engine runs on your machine and AAJ receives nothing. Your AI agent still sends what you give it to its model provider (Anthropic, OpenAI and so on), under your own account's terms.
Hosted MCP serverYour agent's model provider, as above, plus AAJ's server on Cloudflare, which receives the engine inputs over HTTPS, runs them in memory and keeps no copy. Cloudflare keeps error logs (no inputs) for up to 7 days.

No engine needs customer names, emails or other personal data; leave them out.

// limitations

  • It does not replace legal or regulatory review. It flags claims that need support and never says a claim is compliant or legal.
  • It matches wording in the text you paste. It can flag a phrase that is not a claim in context, and it misses implied claims, images, charts and the page's overall impression.
  • An evidence note clears a flag if it quotes the flagged words or names the flag id. The engine does not check that the evidence exists or would hold up.
  • Its guidance links are to US regulators (FTC, HHS, Department of Education) and the EDPB only, and its audience notes cover general, B2B SaaS, AI, EdTech and MedTech.

// review before acting

Legal counsel decides whether each flagged claim can run. The founder or marketer confirms the evidence or signed consent is actually on file, and a reviewer checks the visuals and implied claims the engine cannot see.

// setup

install
$npx skills add sarojkjha/aaj-marketing-skills --skill claims-and-proof-check
hosted mcp
$https://aaj-engines-mcp.sarojkjha10.workers.dev/mcp

Agents listed: Claude Code, Cursor, OpenAI Codex, Windsurf, Cline. Tested and recorded: Claude Code.

Full spec on GitHub

Claims & Proof Check

Find every claim in a piece of copy that a buyer, a competitor or a regulator could ask you to back up, and say what to keep on file before it goes live. The principle: a claim you cannot support is a liability, not a selling point.

When to use

Before publishing a landing page, ad, launch email, sales deck, case study or social post that makes claims about results, rankings, competitors, security, compliance, health or AI. Also when a reviewer asks "can we say this?".

Before you start

  1. Read the brand/product context first. If .agents/product-marketing.md (or .agents/aaj-brand.md) exists, read it for the product, the audience and the proof points already recorded. If none exists, ask the user what the product is and who it is sold to.
  2. Gather inputs: the exact copy (not a summary of it), the audience - "general", "b2b-saas", "ai", "edtech" or "medtech" - and any evidence or permissions already on file (studies, reports, signed releases, review exports).
  3. Confirm the objective: a list of claims to support, reword or cut before publishing. This skill flags; it never decides what is legal.

Method

Read the copy one sentence at a time and treat every checkable statement as a claim. Sort each claim by what it needs: proof (numbers, results, rankings, comparisons, security, compliance, health and AI claims), consent (testimonials, named customers, logos, creators) or a wording review (absolutes, guarantees, "free" and deadline offers). For each, name the evidence that would hold up and point to the regulator's own guidance - not a summary of it. I would rather a founder cut a strong-sounding line than publish one they cannot show.

Workflow

  1. Run the engine on the copy with the right audience (see Run the tool). Pass anything already on file as evidenceOnFile.
  2. Read every flag. Confirm it is a real claim in context; the engine matches wording, so a phrase can occasionally be flagged where it is not a claim. Drop those and say why.
  3. Look for what pattern matching misses: implied claims (a before/after image, "join the teams who stopped losing deals"), claims carried by visuals, and the overall impression of the page.
  4. For each open flag, choose: support it (collect the evidence named in "keep on file"), get consent, reword it to what you can show, or cut it.
  5. Re-run with the new evidence notes until every remaining flag is either on file or rewritten.
  6. Present the result with the disclaimer line, and tell the user to check the legal consequences with counsel before publishing.

The checklist

Claim typeNeedsExample trigger
Specific number, percentage, multiplierproof"40%", "3x", "saves 10 hours", "500+ schools"
Result or outcomeproof"increase revenue", "improves reading scores"
"Proven" / "studies show"proof at the level the words imply"scientifically proven", "research-backed"
Superlative or rankingproof of who ranked you"best", "#1", "industry-leading", "award-winning"
Comparisona dated, sourced side-by-side"faster than X", "unlike other tools"
Rating or reviewthe source export, collection method"4.9 out of 5", "1,200 reviews"
Testimonial, endorsement, influencerconsent, disclosure, typical resultsquoted praise, "trusted by", "as seen in"
Compliance badgethe report or certificate, its scope and date"HIPAA compliant", "SOC 2 Type II", "FERPA"
Security or privacy promisethe control behind it"bank-grade encryption", "we never sell your data"
Health or clinicalcompetent and reliable scientific evidence"clinically proven", "reduces readmissions"
AI capability or accuracythe test behind it"AI-powered", "99% accurate", "replaces your team"
Absolute or guaranteewording review, or the written terms"only", "guaranteed", "100%", "zero downtime"
"Free" or limited-time offerwording review, terms next to the offer"free trial", "risk-free", "ends Friday"
Environmentalspecific evidence"eco-friendly", "carbon neutral"

Sentence verdict = the strongest open need in the sentence: needs proof, then needs consent, then review wording; a sentence with no open flag is OK. These are next steps, never a legal verdict.

Run the tool

node resources/claims-check.js --demo
node resources/claims-check.js '{"text":"Rated 4.9 out of 5. HIPAA compliant.","audience":"medtech"}'
node resources/claims-check.js --help

check(config) takes { text, audience, evidenceOnFile } and returns sentences (each with a verdict), flags (quoted text, claim type, why it needs support, what to keep on file, regulator sources), a summary count, unmatchedEvidence, and the disclaimer. An evidence note clears a flag when it equals the flag id (for example "s3-number") or quotes the flagged words; the flag stays listed as "on file". Audience adds notes where they apply: HIPAA authorization for patient testimonials, FERPA and COPPA for student data and under-13 users, the absence of official "HIPAA certification", and the test standard for AI comparisons with people. The same engine runs in the free Claims & Proof Checker at aajconsult.com/tools/claims-proof-checker.

Reference

Every source the engine links is the regulator's own page: FTC (advertising substantiation, comparative advertising, Advertising FAQs for small business, Endorsement Guides FAQ, Consumer Reviews and Testimonials Rule Q&A, Health Products Compliance Guidance, "Free" guide, Green Guides summary, Start with Security, COPPA FAQ, two AI enforcement releases), HHS (HIPAA marketing; no recognised private Security Rule certifications), the U.S. Department of Education (What is FERPA?) and the EDPB (register of approved GDPR certification schemes). URLs and the line relied on live in SOURCES in resources/claims-check.js.

Present the result

  • Summary: sentences checked, and how many need proof, need consent, need a wording review, or are OK.
  • Per flagged sentence: the sentence, its verdict, and each flag: quoted words, claim type, why it needs support, what to keep on file, the guidance link.
  • Evidence on file: what was noted, and any note that matched nothing (it may point at a claim that was reworded).
  • The disclaimer, verbatim: "This flags claims that need support. It is not legal advice: check the legal consequences with counsel before publishing."

Guardrails & common mistakes

  • Never say a claim is compliant or legal. Say what evidence or consent it needs and link the guidance; the legal call belongs to counsel.
  • Never invent the evidence. If the user has no study, the answer is to get one, narrow the claim, or cut it - not to describe a study that might exist.
  • "Compliant" is not a certificate. Several standards have no official certification at all; the wording must match what the company actually holds.
  • A testimonial does not launder a claim. If you could not say it yourself without proof, a customer cannot say it for you.
  • Check the whole page, not only the sentences. Images, charts and the overall impression can make a claim the words do not.
  • Pattern matching has limits. It finds wording; it cannot see context, so read each flag and the unflagged sentences too.

case-study-and-proof · aaj-copywriting · brand-voice-governance · ai-marketing-governance · creator-partnerships · landing-page-brief

// faq

// install

$npx skills add sarojkjha/aaj-marketing-skills --skill claims-and-proof-check

// worked example

Get the worked example for Claims & Proof Check

One email. The filled-in config, the output, and how to read it.

No spam. Unsubscribe anytime.

// human in the loop

Need a human in the loop?

AAJ builds the strategies, tools, and playbooks these skills are made of — with engagements for growth-stage and enterprise teams.