Which questions could your page answer?
People do not search assistants with keywords, they ask questions. This check derives the questions where your page could be the answer, scores how much of each answer your page already carries, and puts them to real models.
Real questions.
Real answers, on the record.
The shape of a run, not a measurement of anyone. Each row is a question your page could ground, the grounding score beside it, and whether the models that answered named you or named somebody else.
Example output. Your run reports your own questions and answers, at their own link.
Not a sample.
What the engines said about us.
The example above shows the shape. This is the real thing, from the same tracker a paid project gets: 90 of our own questions put to 3 engines over the last fortnight. We publish it because a tool that measures whether AI names you should be willing to say whether AI names it, and today the answer is that 216 of 1,987 answers named us, across 24 of the 90 questions. Ask about us by name and the models know us. Ask the category question and they name somebody else, which is the gap this product exists to measure.
| Question | Named us |
|---|---|
| lantad vs profound Gemini, ChatGPT, Perplexity | 14 of 54 |
| lantad vs otterly Gemini, ChatGPT, Perplexity | 29 of 52 |
| what is lantad Gemini, ChatGPT, Perplexity | 2 of 52 |
| is lantad legit Gemini, ChatGPT, Perplexity | 25 of 51 |
| best tool to check if my website is readable by ai crawlers Gemini, ChatGPT, Perplexity | 0 of 53 |
| best tools to track brand mentions in chatgpt Gemini, ChatGPT, Perplexity | 0 of 53 |
| ai visibility tools for uk businesses Gemini, ChatGPT, Perplexity | 0 of 52 |
| best free ai visibility checker Gemini, ChatGPT, Perplexity | 0 of 52 |
Published 11 September 2026, from the fortnight before it. The table is a sample by a stated rule: the 4 most-answered questions where a model named us, then the 4 most-answered where none did. Read-only and cached, so looking at this page runs nothing and costs nothing. All 90 questions are at /api/demo/prompts.
Read the page.
Screen, score, ask.
Three steps, and every one of them is reported. You see the questions, the scores, the engines, and who each answer named.
-
We read your page
The same crawler-eye extraction the scanner uses. Questions are derived from what is actually on the page, never invented.
-
Every question is screened and scored
A question survives only if your page can ground it. The score says how much of the answer your page already carries.
-
Real models answer, on the record
Each surviving question is asked live. You see who was named, who was named instead, and the sources a grounded answer used.
Four things per question
Not a visibility index and not a share-of-voice chart: four facts per question, each of which can be checked against the answer that produced it.
The question itself
Derived from your page's own text and screened, so every question on the list is one your page could fairly be judged on.
The grounding score
How much of that answer your page already carries, measured against your page's own text. Low means the page needs work before an assistant could build an answer from it.
Who was named
Whether the answer named you, and which other brand names it produced instead. The rival names are extracted from the answer, not supplied by you.
The engines, named
An open-weight model answering from training data alone, and a grounded engine answering from a real retrieval pass with its source pages listed.
An open-weight model
Answers from training data alone, with no web access. This measures whether your brand made it into what models already know. Open-weight models are published for anyone to run, so no consumer assistant is involved.
Gemini with Google Search grounding
Answers from a real retrieval pass, with the source pages it grounded in listed on every answer. This measures what AI search surfaces about you today.
We only claim engines we actually query, and every answer names the engine that produced it. Engines are named by vendor, and each one is that vendor's own API rather than their chat app: an API call carries no account history, no custom instructions and none of the app's own system prompt, so an answer here can differ from the one you get at your keyboard. No panels, no estimates, no invented volume numbers: measured claims with receipts.
Two people this is built for
The free run is a snapshot. Paid plans add the repetition that turns a snapshot into a signal.
Content and SEO teams
You want to know which questions your page can already win, and which ones it would need more material to answer at all.
- Finding the questions a key page could plausibly win
- Seeing which competitors get named instead
- Watching the naming change after a content fix
Founders and marketers
You want to know whether the category question gets answered with your name in it, and if not, whose name is there.
- Before committing budget to AI visibility work
- When a buyer quotes an assistant back at you
- When your positioning depends on being the named example
The questions come from your page, not from a keyword panel. We extract the readable content the way an AI crawler receives it, derive candidate questions from that text, and screen each one: a question survives only if your page could plausibly ground the answer. Nothing is invented.
We do not report prompt volume. Nobody publishes reliable volume for AI assistants, so any number we printed would be an estimate dressed as a measurement.
Common questions
Which engines answer the questions?
Two, and both are named on every answer. An open-weight model answers from training data alone with no web access, which measures whether your brand made it into what models already know. Gemini with Google Search grounding answers from a real retrieval pass and lists the source pages it grounded in, which measures what AI search surfaces about you today.
Why does this check need my email when the others do not?
Because every run puts 6 questions to real models live, the 5 we derive for your page plus one branded control, and each one is answered by every engine on the run. That costs real compute, so the result link goes to your inbox rather than being handed out anonymously. One run per page per day, three runs per day per address.
Where do the questions come from?
From your page's own text. We extract the readable content the way an AI crawler receives it, derive candidate questions from that, and screen each one: a question survives only if your page could plausibly ground the answer. Nothing is pulled from a keyword panel and nothing is invented.
What does the number beside each question mean?
The grounding score: how much of that answer your page already carries, measured against your page's own text. A high score means the material is there and the question is a fair one to be judged on. A low score means the page would have to be improved before an assistant could build an answer from it.
Do you report how many people ask each prompt?
No. Nobody publishes reliable prompt volume for AI assistants, so any number we printed would be an estimate dressed as a measurement. We report what we actually did: the questions derived, the grounding score, which engines answered, who they named, and the sources a grounded answer used.
What do paid plans add?
Repetition, which is the part that turns a snapshot into a signal. Paid plans track up to 25 prompts per page with scheduled re-runs beside your crawler monitoring, weekly and twice weekly on Business, keep the naming history, rank the competitors models reach for instead, and alert you the week an answer changes.
See the questions you could own.
One page, 6 questions put to real models, and a result you can check line by line. If the grounding scores are low, the page is the thing to fix first.