Can AI visibility tools see the prompts people actually type?

No AI visibility tool can see the prompts real people type into ChatGPT, Gemini, Perplexity or Claude. The engines don't share that data with anyone. What a tracker does is send its own prompts to an engine, usually through an API, and record what comes back.

That makes every prompt-based score a sample of a sample:

• The prompts are chosen by the tool or by you, not drawn from real users.
• The engine may or may not search the web for a given prompt. Retrieval is discretionary on every major platform.
• When the engine does search, it rewrites the prompt into its own queries first.

The third point is the one most buyers miss. Even a perfect copy of a real user's prompt is not what the retrieval system searches for. The question of whether AI searches for the exact question a user types has a documented answer from the vendors, and it's no.

Three vendors describe the rewriting step in their own documentation.

• OpenAI: its help page on ChatGPT search says ChatGPT "typically rewrites your query into one or more targeted queries" that it sends to search providers. It adds that, with memory enabled, ChatGPT may use saved memories when rewriting the query.
• Google: AI Overviews and AI Mode may use a "query fan-out" technique, issuing multiple related searches across subtopics.
• Gemini API: Grounding with Google Search "automatically generates one or multiple search queries" and runs them.

So one user prompt can become several queries, shaped by context the tracker never sees. Memory and custom instructions make logged-in answers hard to reproduce at all.

What this means for you: a tracker's prompt list measures a proxy for an input nobody outside the engine can observe. That doesn't make prompt tracking useless. It makes it a directional test, and you should read its numbers that way. If you're deciding how to build that list, see whether to write your own prompts or let a tool generate them.

Bing Webmaster Tools added an AI Performance report in public preview on 10 February 2026 and expanded it in June 2026. It includes grounding queries, which Microsoft describes as "the key phrases the AI used when retrieving content that was referenced in AI-generated answers." Coverage spans Microsoft Copilot, AI summaries in Bing and select partner integrations.

This is different in kind from anything a tracker produces. It is the engine reporting the queries it actually ran, not a vendor guessing at what users might ask. The report also shows total citations, cited pages, intents, topics and a Citation Share metric. Microsoft frames Citation Share as observational, not a ranking or a competitive scoreboard.

Three limits apply:

• It's Microsoft's surfaces only. It tells you nothing direct about ChatGPT, Gemini, Perplexity or Claude.
• It only shows queries that led to your content being cited. Queries that retrieved a competitor instead are invisible to you.
• The grounding queries data is a sample. Microsoft scopes that caveat to grounding queries specifically. See whether Bing's AI Performance data is complete or sampled.

As of the evidence we hold (August 2026), the report had no API, so the data lives in the web interface and exports. Check whether there is an API for Bing's AI Performance report for the current state. You can get started with the free Copilot citation data in Bing Webmaster Tools.

Google Search Console has a generative AI performance report for AI Overviews and AI Mode. Its dimensions are pages, countries, dates and devices. It reports impressions only, with no clicks, no CTR, no position, and no query breakdown.

That means Google gives you no first-party view of which queries surfaced your pages in its AI features. Google also states that AI Overviews and AI Mode "may use different models and techniques," so one surface's behavior doesn't stand in for the other.

The gap between tracker samples and engine counts can be large. In OtterlyAI's 2026 experiment, Microsoft's AI Performance report credited one test site with 42,817 citations, while Otterly's own prompt tracker logged 3,934. The time windows may not match, so it's not an exact ratio. It does show that an engine's own count sees more than any prompt sample.

Treat prompt tracking and engine data as two instruments that check each other. A side-by-side view of prompt-based tracking versus engine-reported data shows what each can and can't tell you.

1. Verify your site in Bing Webmaster Tools. It's free, and it's the only place to see real grounding queries.
2. Export your grounding queries and group them by intent and topic.
3. Compare them with your prompt set. Where they overlap, your tracker is testing something close to reality. Where they don't, your prompts are measuring questions the engine isn't running. The playbook on comparing grounding queries with your prompt set walks through it.
4. Report each engine separately. Bing data says nothing about ChatGPT, and a blended score hides that.

If you want a quick snapshot before any of that, the free checker on this site runs up to three buyer questions and shows whether AI engines name you. It's a snapshot, not continuous tracking, and it carries the same prompt-proxy limits described above.

Do AI visibility tools buy or collect real user prompt data? None of the major engines publish user prompts, so a tool can't get them from ChatGPT, Gemini, Claude or Perplexity directly. Some vendors build panels or clickstream samples, but those are third-party estimates on a chosen population. They are not the engine's own record, and they still miss the rewritten queries the engine runs behind each prompt.

Are grounding queries the same as keywords? Not exactly. Grounding queries are the phrases the AI used to retrieve content that it then cited. They often look like search keywords because the engine is issuing searches. But they are generated by the model from a user's prompt and context, so one conversation can produce several of them, and you can't target them the way you would a single keyword.

Can I see grounding queries for ChatGPT? Not as of October 2026. OpenAI documents that ChatGPT rewrites prompts into targeted queries, but it doesn't publish those queries to site owners. Bing's grounding queries cover Copilot, Bing AI summaries and select partner integrations. Don't assume they mirror what ChatGPT searched for, even though OpenAI says ChatGPT search sometimes uses other search providers.

If prompts are only a proxy, is prompt tracking worth paying for? It can be, if you read it correctly. A stable prompt set tracked over time shows whether answers about your brand change and whether engines get your facts right. It can't tell you your share of real demand. Pair it with Bing grounding queries and check how well the two overlap before you trust any trend.

Does the prompt I test match what a logged-in user sees? Often not. OpenAI says that with memory enabled, ChatGPT may use saved memories when rewriting a search query, and product results can reflect memory or custom instructions. A tracker testing logged-out, memory-off sessions sees a cleaner but different answer. That's a reason to treat any single answer as a sample, not a verdict.