Truth-to-Answer Loop
A fact protocol, not an AI visibility score.
The method separates observed answers, approved product truth, root-cause hypotheses, controlled repairs, and recheck observations so one cannot silently stand in for another.
01
Buyer Question
Confirm five questions that come from sales, support, reviews, search, or real customers.
02
Observed Answer
Preserve what AI products, search, the official site, and relevant third parties currently say.
03
Approved Truth
Break product facts into atomic claims and obtain written approval from a named fact owner.
04
Discrepancy
Mark each difference as inaccurate, outdated, conflicting, missing, unsupported, confidential, or cannot determine.
05
Controlled Fix
Change only an approved page, document, or visible structured-data field the client controls.
06
Distribution
Update relevant internal links and sitemap within the approved repair scope. Third-party corrections require an agreed scope and a client-controlled approval path; bulk correction requires a separate quote. Propagation timing is not guaranteed.
07
Recheck
Repeat the same questions and disclosed conditions 14 days after the approved changes are live.
08
Change Log
Record the fact version, approver, evidence, implementation, QA, date, and recheck observation.
Measurement boundary
What the evidence can—and cannot—say.
- Five prompts × two or three AI products × three repeats is an observation set, not a statistical ranking.
- Mention, answer position, citation, and factual accuracy are recorded separately.
- A small number of outputs cannot represent stable model knowledge or brand perception.
- The recheck reports changed, unchanged, mixed, or cannot determine. It does not claim causality.
- No 0–100 summary score is generated because it would hide the sample and evidence limits.
Root causes remain hypotheses until evidence supports them.
Own-site conflict
Missing fact
Stale third party
Entity collision
Model fabrication
Prompt ambiguity
Cannot determine