See What AI Says in brief

What it is

A one-time audit of whether AI assistants name your business when someone asks them a question you'd want to be the answer to. You get a PDF that is yours to keep and a web link that stays live for 12 months. It is not a subscription, not a dashboard, and not an agency retainer.

What it costs

$4.99, one-time, no subscription. Paid before the run starts.

What gets measured

Engine	How it's asked	Why it's in the set
ChatGPT	with web search	The assistant most of your customers actually use
Gemini	with Google Search	The default assistant on Android and across Google's apps
Claude	from memory, no web access	Heavily used by business buyers
Google AI Overviews	live search results	The AI answer your customers see without asking for it
Google AI Mode	live search results	Google's dedicated AI answer tab

Not live yet, and therefore not measured and not charged for: Gemini (from memory, no web access). This line disappears from the site and from every report the day these surfaces are switched on.

The question set

• 25 blind questions — your name is never mentioned. These are the ones that count, because they measure whether you get found rather than whether you get described. We spread them across four buying stages: 3 on category discovery, 8 on solution comparison, 9 on vendor selection, 5 on high intent.
• 5 branded questions — we include your name, to see what the engines already believe about you: identity, facts, pricing, competitor, reviews. The facts one is graded for accuracy against your own site.
• 3 of the questions are asked 3 times. You choose which. These systems are not deterministic, so a single run of a drifting system is a coin flip — repeat-sampling is the difference between a number and an anecdote.

The workload behind one report

• 30 questions
• 176 answers recorded and kept
• 126 graded reads — one per model answer
• 302 engine calls in total

The 176 answers are everything we recorded. 126 of them are model answers, each read and graded once; answers read off Google's AI surfaces are checked mechanically rather than graded by a model.

How an answer is scored

Every answer lands in exactly one of 5 buckets. The rule is published because a counting rule you can't reproduce is an opinion, and this category is full of opinions sold as measurements.

• recommended you
• named you in passing
• read your site, recommended someone else
• just repeated the name from the prompt
• did not mention you

Below 20 mentions we call a breakdown directional rather than diagnostic. Trakkr published that threshold, and we adopted it rather than invent a friendlier one.

The verdict

Each report opens with one of 4 answers to the question you actually paid to settle — should I spend money and time on this at all? They are fine, blocked, fixable, structural.

fine means nothing here is worth your money. It is a real outcome with a mechanical threshold, not a courtesy. Two reports with the same numbers and the same version reach the same verdict; moving a threshold is a version bump.

What this measurement cannot see

Printed here and in the appendix of every report, in the same words:

• Results are logged-out. A signed-in user may see something different.
• We measure whether AI names you, not how many people asked.
• We do not attribute revenue to AI mentions. Nobody currently can.
• Findings are true on the date stamped on them.
• Only your three key questions were asked more than once. The rest were asked a single time.
• No cross-day sampling — the whole run happens within one session.
• Google AI Mode is not available in every market.
• 30 questions is a sample, not a census. Treat the direction as real and the exact percentages as approximate.
• Gemini (from memory) was not queried. Everything here describes ChatGPT, Gemini, Claude, Google AI Overviews and Google AI Mode, and answers on one assistant do not reliably transfer to another.

What we will not do

• We never publish a customer's results. Your report is a private link served noindex. No leaderboard, no public index, no "top brands in AI search" post. Several companies in this category run exactly those.
• We don't claim a trick that doesn't exist. There is no file to upload and no markup to add that makes an assistant recommend you. Google says so directly and we quote them on the FAQ.
• We don't attribute revenue to AI mentions. Nobody currently can, and anyone showing you that number is modelling it.

Version

The version identifier is stamped on every report, so you can tell whether two reports are comparable. The benchmark is the same pipeline (the method) run against 34 brands you already know — it's published at /benchmark so you can check our numbers against companies you can judge for yourself.

Contact

[email protected]. More on who runs this and how to check our work.