Your customer asks an AI engine which tool to pick. You read the answer. What you do not see is what happened before it: the engine typed three searches of its own, opened the pages they returned, and wrote its answer from them.
If none of those searches leads to a page of yours, you are not in the answer. And until now, you had no way to know what those searches were.
What are AI searches, or fan-out queries?
AI searches are the web searches an AI engine runs on its own before answering a question. The industry calls them fan-out queries: one question fans out into several searches, each looking for a piece of the answer.
Ask an engine "which CRM for a 20-person sales team?" and it may search "best CRM small sales team", "HubSpot vs Pipedrive" and "Pipedrive pricing". It reads what those searches return, keeps what it trusts, and builds its answer. The question your customer typed is not the question the web was asked.
That is why ranking on your customer's question is not enough. You have to answer the questions the engine asks on their behalf.
What does the AI searches page show?
Mentionable records the searches the engines run on every answer to the questions you track, and the AI searches page turns them into something you can read in a minute.
Two numbers at the top. How many distinct searches the engines ran over the period, across how many themes. And the share of your tracked answers where the engine ran at least one visible search: 40% means four answers in ten were built on a search you can read.
Search themes. Similar searches grouped together, from the most searched to the least, each with a bar showing its weight against the top theme. Under each theme: how many times it was searched, how many variants it has, how many of your tracked questions trigger it, and the icons of the engines that ran it.
What each theme is after. A label tells you what the search looks for: a comparison, a price, reviews, a how-to, a definition or a local result. A filter keeps only one kind, for example every price search the engines ran about your market.
The detail of a theme. One click opens the exact searches inside the theme, with how many times each one was run, and the tracked questions that triggered them.
How are the searches grouped into themes?
Two searches belong to the same theme when they share at least half of their meaningful words. Case, accents, filler words and years are set aside first, so "best CRM 2026" and "best CRM" land together, and so do the same words in another order.
Each theme is named after its most frequent search, written the way the engine wrote it.
The grouping uses no AI model. The same searches always give the same themes, so a theme you saw last week is still there this week, with its numbers moved.
Which AI engines show their searches?
Not all of them, and not on every answer.
Some engines expose the searches they ran, others keep them to themselves. The same engine can search the web on one question and answer from memory on the next. Mentionable records every search an engine makes visible, and nothing it hides.
That is the reason for the second number at the top of the page. It tells you how much of the picture you are looking at, instead of letting a partial list pass for a complete one.
What do you do with a search theme?
You read it as a brief from the engine itself. A theme is a topic the engines need answered before they recommend anyone in your market.
- A comparison theme ("your brand vs a competitor", "best tools for…") asks for a comparison page. If you do not publish one, the engine reads someone else's.
- A price theme asks for clear pricing, on a page the engine can read.
- A reviews theme sends the engine to review sites and forums. That work happens off your site, on the sources the engines cite.
- A how-to or definition theme asks for a guide that answers it plainly.
The sites the engines read during these searches are on the Sources page, so you can see who answers each theme today. And when Mentionable drafts an article for one of your tracked questions, it starts from what the engines searched on that question.
Can I filter the searches?
Yes. Search a word inside the queries, pick one tracked question, change the period (the last 30 days by default), keep only one engine, or one country when your questions cover several.
Filtering by country matters more than it looks. The same question asked from France and from the US does not trigger the same searches, because the engine searches the web from where the question comes from. See AI visibility by country.
Can my own AI read the searches?
Yes, through the Mentionable MCP server. The list_fan_outs tool returns the raw searches, deduplicated and ranked by frequency, with the questions that triggered them, the engines that ran them, and when each one was first and last seen. Your own AI can then cross them with the rest of your data, for example to find the searches that no page of your site answers.
Which plans include AI searches?
All of them, and the 7-day free trial too.
The searches come from the scans Mentionable already runs on your tracked questions. There is no separate scan, no extra cost and nothing to turn on: the page fills up as your answers come in.