A search auditor with a ledger habit

Three columns on paper taught me more than the first reporting tool I was given: client name, job promised, proof delivered. That was in the back room of a small printing office near Rennes, with two phones ringing and a wall calendar carrying half the business memory. I come from western France, and that early habit stayed with me. Before I name a problem, I want to see the mark it leaves: the phrase a customer uses, the page a machine cites, the wrong description that keeps returning even after a website has been edited.

For 17 years I have worked around search pages and the evidence behind them. Search audits, local landing-page diagnostics, multilingual content reviews, competitive SERP notes, editorial quality checks, internal reporting systems for service businesses — the unglamorous parts. I like the unglamorous parts. They show whether a business is understood by the systems that repeat its name. When AI answers entered the buyer path, I saw many teams carry over the old reflex: collect one good-looking answer, paste it into a slide, call it visibility. That is too fragile for decisions.

My work now is to build repeatable tests. I keep prompt ledgers with date, engine, language, location intent, cited source, answer position and description error in separate columns. French prompts and English prompts are tested separately because the evidence path often changes by language. Competitors stay in the set, because visibility without context flatters everyone. My stance is simple: an AI answer is useful evidence only when it appears often enough, names the right source often enough, and describes the business accurately enough to guide action. Everything else is weather talk.