The whole method, published. What we ask, how we ask it, how we score it, and the rules we hold ourselves to. No email required — if you want to run it yourself, this is enough to start.
Every audit tests ChatGPT, Claude, Gemini, Perplexity, and Google’s AI answers. Those five carry the overwhelming majority of AI answer traffic. We report each engine separately and never average them together — they behave differently enough that a blended number hides the finding. Gemini, for instance, names brands far more often than it cites sources; grading it by Perplexity’s citation habits would produce a false failure.
Most tools guess at prompts. We build the question set from evidence: the questions your sales calls, support tickets, and reviews show people actually asking, plus the follow-up questions the engines themselves offer when someone asks about your category.
Questions are grouped in tiers, because they measure different things: branded (does AI know who you are, and are the facts right?), comparison (does it name you unprompted when someone asks for a recommendation?), problem-first (does it surface you when a buyer describes a need without naming a category?), and cost (who owns the pricing answer you’re absent from?).
AI answers are non-deterministic. The same question, asked twice, minutes apart, in identical conditions, returns different answers — and often a different brand list. Published research bears this out: run the same prompt three times in ChatGPT and only about 2% of cited sources appear in all three runs. Independent studies find you need roughly seven to eight runs before a brand’s mention rate is statistically stable.
So we don’t report a yes or a no. We run every question multiple times and report a rate — “named in 6 of 8 runs” — with the variance stated plainly. It is the difference between a measurement and an anecdote. If you re-run one question yourself and get a different answer than our report shows, that is expected, and the rate is what makes our number reproducible anyway.
Every probe runs logged out, with no history, no memory, and no personalization — in a fresh session, from your buyers’ geography. Checking your own logged-in ChatGPT gives you a flattering answer no prospect will ever see; the engine remembers you.
Some platforms simulate a buyer’s chat history to make answers feel “contextual.” We don’t. A simulated history can’t be reproduced or verified — you’d be measuring the engine multiplied by someone’s fiction of your buyer. The cold probe is the reproducible floor: any skeptic, including your own team, can re-run it.
A score tells you there’s a problem. The part that changes anything is the meaning-space map: we measure how closely each of your pages sits to the question a buyer actually asked, against the competitors the engines recommend instead. When a competitor sits at 82 and you sit at 52, the gap usually isn’t quality — it’s aim. Their page answers the exact question; yours answers a nearby one.
That single view converts “we’re invisible” into a build list: the specific questions your content doesn’t answer, ranked by how close each sits to a purchase.
Before any content work matters, the engines have to be permitted to read and use your pages. We read your live robots.txt, check every AI crawler by name, and look for Content Signals — the machine-readable line that says whether your content may be used in AI answers.
We find sites broadcasting ai-input=no while paying for visibility consulting. That’s a four-hour fix that gates everything downstream, and it belongs in Week Zero of any plan.
Any projection in our reports is labeled a forecast, modeled conservatively, with the assumptions shown. Nobody can guarantee placement inside an AI engine, and any firm that does is selling something.
What makes the number credible is that we go back: at day 60 and day 90 we re-run the same questions, on the same engines, with the same method, and report the measured movement — including when it’s less than we projected. Forecast replaced by evidence, not defended.
The method above is yours to use. If you’d rather have it run properly, with the transcripts, the map, and the 90-day plan, that’s the audit.
See the AI Visibility Audit Start with a $99 Snapshot