Skip to main content

What the AIs searched for

The web searches assistants ran to answer your prompts — grouped into topics, split by AI platform, and what it means when one didn't search at all.

When an AI assistant answers a question about your category, it often searches the web first — and what it searches for is not your prompt. The Searches page shows you what it actually typed.

Where to find it

Searches, in the left sidebar under your brand. It reads the same date range, tags and prompts as the rest of your dashboard, so what you filter elsewhere you filter here.

Your prompts versus their searches

You track "What are the best running shoes for marathon training?" To answer it, an assistant might search best marathon running shoes, carbon plate running shoe reviews, and Hoka vs Nike marathon. Three searches you never wrote, from one prompt you did.

That gap is the point of the page. Your prompt is the question your buyer asks. These are the queries that decide which pages the assistant ever sees — and so which pages can be cited at all. A page written to match your prompt's wording, but matching nothing in this list, never enters the answer it was meant to win. See What makes a page get cited by AI.

Topics

Assistants rarely search the same phrase twice. best marathon running shoes, best running shoes for marathons and top marathon shoes are one intent typed three ways, so MentionBird groups near-identical searches into a topic and reports the topic.

Each row carries how often the topic was searched, its share of all searches in range, and which of your prompts triggered it. Open a row to read the individual searches inside it.

The long tail — searches too varied or too rare to group — collects in a terminal Other searches row rather than disappearing. Finding it near the top tells you something real: assistants in your category are searching widely rather than converging on a few phrases.

Topics are mined from your brand's recent history overnight. Until that has run, the page says its topics are still being built and lists one row per search in the meantime. The counts are complete either way; only the grouping is pending.

Splitting by AI platform

The list can be read one AI platform at a time, using the filter at the top. The same prompt often sends different assistants down different paths — one searching for reviews, another for head-to-head comparisons — and that difference is actionable in a way the blended view hides.

One thing to read correctly: an AI platform that doesn't publish the searches it ran shows no rows here rather than a zero. Its answers still count everywhere else in MentionBird, and the figures on this page describe the assistants that report their searching.

Answered without searching

Some answers arrive with no search at all — the assistant replied from what it already knew. The page reports that share as answered without searching.

It is the most strategic number here. When an assistant searches, fresh pages can change the answer, which is the opening ranking advice works in. When it doesn't, the answer reflects what the model absorbed long before you published anything, and a page you put up this week cannot reach that run. A prompt with a high share here is one you win slowly, through mentions earned across the web, rather than through a page of your own. See Why AI answers change between platforms and days.

One language at a time

If you track prompts in more than one language, searches are grouped per language and a chooser appears at the top of the page. Spanish prompts produce Spanish searches, and folding those into one list would group phrases that don't mean the same thing. Pick the language you're working on; every total on the page is scoped to it.

Language is read from the prompt itself, and is a separate thing from the country you chose for it — Global is a country setting. See Choosing AI platforms and countries.

Exporting

Export gives you the page as a CSV, filtered exactly as you are looking at it — the same date range, tags, prompts, AI platform and language. The topic column carries the topic's readable name rather than an internal key, so the file stands on its own. It works like your other exports; see Exporting citations and sources.

Where else searches appear

An individual run card lists the searches that ran for that one answer, in the order they ran — useful when you are already reading an answer and want to know how it was assembled. See Reading the full AI answer for a prompt run.

Your assistant can reach the same data through MCP and the API. See What you can ask through MCP.

Did this answer your question?