Brand claim audit before AI speaks

Collect the claim surface

AI will happily reuse whatever it can see. If the website still says 'trusted by 400 teams' after a quiet product reset, the graph will learn a stale boast. Start the audit with official pages, sales decks, bios, and the last 20 public posts. That is the real claim surface, not the brand book.

Classify every load-bearing sentence

Use four buckets. If a sentence can change a buyer's mind, it is load-bearing.

• Provable now: evidence exists and a person will stand behind it.
• Provable with work: the fact is likely true, but the source is missing or informal.
• Aspirational: directionally true as a goal, dangerous as a present-tense claim.
• Forbidden: legal, competitive, or reputational risk if a model says it fluently.

Audit checklist

Work through these with the Living Brand Graph open, or with a spreadsheet if you are starting cold.

1. 01List quantified claims: customers, percentages, time saved, rankings, years of experience.
2. 02List named proof: logos, quotes, case studies, certifications, partner names.
3. 03List category claims: 'the only,' 'first,' 'leading,' 'guaranteed.'
4. 04Attach a source or mark the claim as candidate or rejected.
5. 05Lock the claims that must not drift, then rerun strategy so pillars do not depend on rejected proof.

After the audit

Update the site and the graph together. Otherwise source-aware content will keep citing a page you no longer believe. A claim audit is one of the highest-leverage AI readiness tasks a founder can do in a single afternoon.

If you cannot find the evidence, the model should not be more confident than you are.