# Marketing Claim Substantiation > Reviews marketing or client-facing copy claim by claim, classifies each claim type, demands a named and dated source, and returns a verdict of SUPPORTED, NEEDS-QUALIFIER, or UNSOURCED — never a rewrite that quietly weakens an unsupported claim. Use before publishing web copy, ads, decks, PR, or sales collateral; trigger on 'check these claims', 'is this copy compliant', 'can we say this', 'substantiate', 'legal review of the copy', 'we claim we are the fastest'. Not for writing the campaign that will carry the claims (use marketing-campaign-brief), not for search-intent content planning (use marketing-seo-content-brief), and not for verifying a metric's calculation logic (use data-analytics-metric-definition). - Skill: `alihusains/marketing-claim-substantiation` (Agent Skill) - Install (CLI): `npx skillmds@latest add alihusains/marketing-claim-substantiation` - Raw SKILL.md: https://api.skillmd.com/api/skills/alihusains/marketing-claim-substantiation/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Marketing & Growth - Author: alihusains (https://skillmd.com/u/alihusains) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/alihusains/marketing-claim-substantiation --- # Claim Substantiation Review ## Purpose Copy reaches publication carrying numbers nobody can trace and comparisons nobody tested, and the fix applied under time pressure is to blur the wording rather than find the evidence. This skill separates the two: it identifies what type of claim each sentence makes, states what evidence that type requires, and returns an explicit verdict per claim. It does not rewrite unsupported claims into safer sounding ones. ## Prerequisites - **Inputs:** the copy under review, complete, in its final intended context — headline, body, footnotes, and any on-screen or spoken variants. Partial copy produces partial review. - **Access:** the evidence sources the author intends to rely on: research reports, internal analytics or benchmarking results, third-party test results, certifications, customer data with consent, or published pricing. - **Also required:** the market or jurisdiction the copy will run in, and the audience type (consumer or business). Both change what is permissible. If the intended market is unknown, name it and stop. The same claim is compliant in one jurisdiction and prohibited in another, and reviewing against the wrong one gives false assurance. ## Data classification **Internal** during review. Evidence files may be **Confidential** — cite them by title, owner, and date rather than pasting their contents into shared documents. If substantiation depends on customer data, confirm the consent basis before it is used and never reproduce individual records, account numbers, or positions in the review output. ## Claim types and evidence required | Claim type | Example shape | Evidence required | | --- | --- | --- | | Performance | "cuts processing time by 40%" | Named measurement, method, sample, date, scope | | Comparative | "faster than the leading alternative" | Like-for-like test against the named comparator, dated, methodology available | | Superlative | "the best", "the only", "number one" | Verifiable ranking from an independent source, with the ranking criteria and period | | Outcome or result | "customers save X per year" | Actual observed results with the population defined and typicality stated | | Testimonial | quoted customer praise | Written permission, genuine, attributable, not edited to change meaning | | Certification | "ISO certified", "regulated by" | Current certificate or register entry, with number and expiry | | Pricing | "free", "from $X", "no fees" | Full conditions disclosed at equal prominence to the price claim | | Forward-looking | "will reduce", "guarantees" | Almost never substantiable — reframe as a capability, or drop | | Subjective preference | "beautifully designed" | None required; not a factual claim. Do not spend review time here | ## Procedure 1. **Extract claims atomically.** Split the copy into individual factual assertions. One sentence can carry three claims; a bulleted feature list usually carries one each. Number them so the verdict list maps back to the copy. 2. **Classify each claim** against the table above. Classification determines the evidence bar, so do this before looking at any evidence. Subjective preference claims are marked and set aside. 3. **Request the source per claim.** For each, capture: source title, owner, date, population or sample, and method. A link alone is not a source; an internal dashboard screenshot is not a source unless the metric behind it is defined and owned. 4. **Test scope match.** The most common failure is a real source that does not support the claim as written — measured on one product but claimed for the range, measured in one region but claimed globally, measured in a controlled test but claimed as typical experience. Compare the claim's scope to the evidence's scope explicitly and write the comparison down. 5. **Check currency.** Evidence older than the last material change to the product, pricing, or market does not support a present-tense claim. Record the evidence date next to the claim and let the approver judge; do not silently accept it. 6. **Assign one verdict per claim.** | Verdict | Meaning | What happens next | | --- | --- | --- | | SUPPORTED | Evidence matches claim in scope, method, and date | Publishable; record the source reference in the claim register | | NEEDS-QUALIFIER | Evidence supports a narrower claim | Add the specific qualifier and footnote the source at equal prominence. Re-review the qualified wording | | UNSOURCED | No evidence supplied, or evidence does not support it | Blocked. Flag to the author and the approver. Do not publish | | NOT-A-CLAIM | Subjective, non-factual | No action | 7. **Never launder an unsourced claim.** Rewriting "40% faster" as "dramatically faster" removes the number, not the assertion, and is still unsupported. If a claim is UNSOURCED, the options are: find evidence, state the narrower supported claim, or remove it. Say this plainly rather than offering softer phrasing. 8. **Produce the claim register.** Output a table of claim number, claim text, type, verdict, source reference, evidence date, and the qualifier where one was added. This register is the artefact that survives the review — attach it to the copy so the next reviewer does not repeat the work. 9. **Route blockers.** Send UNSOURCED claims to the author and to whoever holds the legal or compliance gate, with the specific evidence that would resolve each one. "Needs substantiation" without saying what would substantiate it returns unchanged. ## Failure modes - **Aggregating footnotes.** One asterisk covering six claims makes none of them clear. One qualifier per claim, adjacent to the claim. - **Prominence mismatch.** A headline claim with conditions in eight-point type at the foot of the page is a failed disclosure regardless of accuracy. - **Reviewing copy without its layout.** Prominence and adjacency cannot be judged from a text file. Request the artwork or page. - **Second-hand evidence.** A vendor's own marketing quoted as the source for your comparative claim is not independent evidence. - **Silent scope creep at translation or localisation.** Every localised variant re-enters this procedure; a claim compliant in the source market may not be in the target. ## Boundaries - Building the campaign that will carry these claims — use `marketing-campaign-brief`. - Planning content against search demand — use `marketing-seo-content-brief`. - Confirming that an internal metric is calculated correctly in the first place — use `data-analytics-metric-definition`. - Answering a customer or regulator complaint about a published claim — that is an incident, not a review; escalate to the compliance owner. ## Hand-offs - **Receives from:** `marketing-campaign-brief` (claim list from step 5), `sales-proposal-assembly` (claims used in proposals and RFP responses). - **Routes to:** `data-analytics-metric-definition` when a claim rests on an internally computed number whose definition is unclear, and back to the originating skill with the claim register attached.