How to find the AI prompts your brand shows up for (when Search Console only gives you impressions)

Questions this guide answers

• How can SEO teams find the queries where their brand appears in AI search?
• Can I see which ChatGPT prompts mention my brand?
• Does Search Console show AI Mode queries?
• What regex finds AI Mode queries in Search Console?
• What are grounding queries in Bing Webmaster Tools?

Direct answer

You can't download a list of the AI prompts your brand appears in, because no engine publishes one. Search Console's generative AI report shows how often your pages appeared in AI Overviews and AI Mode, with no queries. Bing Webmaster Tools shows a sample of the search phrases Copilot used to find your pages. ChatGPT, Claude, and Perplexity offer site owners no prompt report.

So work in reverse. Collect candidate questions from what you can see: long and question-form queries in Search Console, Bing's grounding queries, and the questions your sales and support teams already hear. Rewrite them the way a buyer would ask an assistant. Then run them on the engines your buyers use, several times each, and record whether the answers name you, recommend you, or cite your pages. The prompts you keep become your prompt set: the fixed list of questions you measure over time.

Why there is no AI keyword report

A keyword report works because search queries are short and repeated. Millions of people type the same 3 words, so a ranking for that string is a stable fact you can report.

Prompts don't behave that way. Google says the average AI Mode search is triple the length of a traditional search (May 19, 2026). In a UC Berkeley study of 24,069 conversations with search-enabled AI models, prompts averaged 57.1 words. Long, specific questions are rarely asked twice in the same words.

That has 2 consequences. The engines have little reason to report prompt strings, and so far none do for site owners. And when a prompt does reach a report, it is often too rare to show: Search Console's Performance report documentation says that "to protect user privacy," it might not track "some queries that are made a very small number of times."

What each source can and can't tell you

Checked against each platform's documentation on Oct. 14, 2026. For a fuller tour of the free reports, see what Search Console, Bing, GA4, and Merchant Center show.

Step 1: Mine Search Console for conversational queries

Google confirms that AI Mode activity is in the standard Performance report. Its counting rules say a follow-up question in AI Mode is "essentially performing a new query," and John Mueller said in August 2026 that Search Console "includes information on AI Overviews and AI Mode in the general performance report," when asked about rows such as "yes, pricing" in the queries table. There is no filter that separates them, but 3 regex filters on the Query dimension narrow the list to likely candidates. Search Console uses RE2 syntax; the (?i) prefix makes a pattern ignore case.

• Read them by page. Add the Page dimension and compare with the pages that show impressions in the generative AI report. A long question-form query landing on a page with AI impressions is a strong candidate.
• Group by intent. Rare queries are withheld, so the report shows the more common phrasings. Group what you find by the intent and persona behind it.
• Watch for automated probes. Monitoring tools send scheduled prompts to Google too. Oddly specific long queries with steady impressions and no clicks may be machines checking a brand, not buyers.
• Don't over-read a match. A 12-word question can be an ordinary search. The filters find candidates; testing in Step 4 decides what matters.

Step 2: Read Bing's grounding queries

Bing Webmaster Tools' AI Performance report shows grounding queries, which Microsoft describes as "the key phrases the AI used when retrieving content that was referenced in AI-generated answers." They are the AI's search, not the user's prompt, and Microsoft says they represent "a sample of overall citation activity." Still, they are the only first-party list of query phrasings tied to AI citations of your pages.

On Oct. 13, 2026, Microsoft added Topics Portfolio (Beta), which "groups the grounding queries where your site appears into broader thematic clusters" and lists Top Opportunity Queries: "the strongest queries across your whole portfolio where other sites appear and yours does not." Microsoft shows some volumes in the new view as ranges and asks you to treat them "as directional signals for planning, not as a precise ledger."

Use both lists as raw material. A grounding query such as "payroll software contractor payments multiple countries" points to the buyer question behind it, and an opportunity query names a topic where you are absent.

Step 3: Add the questions you already hear

The richest source of prompts isn't a search tool. It's the questions buyers already put to your team, in roughly the words they would type into an assistant.

• Sales calls and demos: the comparison and "will it work for us" questions asked before a deal.
• Support tickets and onboarding: the setup, integration, and limit questions people ask after buying, which other buyers ask AI before buying.
• Lost-deal and closed-won interviews: what the buyer was looking for, and which alternatives they considered.
• Public communities: the threads where your buyers ask each other for recommendations, and the reviews they leave.

Step 4: Rewrite, then test on the engines

Turn each candidate into a full question with the context a buyer would give: role, company size, constraint, and what they are comparing. Keep a mix of types: category questions asked before the buyer knows your name, comparison and use-case questions, and a few branded questions to check accuracy. AI search prompt set covers the 7 prompt types worth including.

Run each prompt more than once

Run every candidate on the engines your buyers use, and run it several times. Answers vary from run to run, so one check is one sample (see how many runs an AI-visibility number needs). For each answer, record whether it names you, whether it recommends you, which brands it names instead, and which pages it cites.

Keep the prompts with real buyer intent, including the ones where you never appear. A prompt at zero with real demand is the most useful gap in the set, not a prompt to delete (why). Once the set is fixed, read it per prompt rather than as one average (how).

Step 5: Keep the set current

New questions appear when a market changes: a product launch, a pricing change, a news story, a new integration. Repeat Steps 1 to 3 monthly and add the new candidates. When you add or retire a prompt, start its own history, so the trend for the rest of the set isn't distorted.

What to skip

Four shortcuts look tempting and mislead.

• Asking an assistant which prompts it mentions you for. It has no access to other people's conversations, so its answer is a guess.
• Ranking by prompt volume alone. Volume figures for prompts are modeled from other signals. Use them to prioritize intents alongside persona fit and buyer value, as long as the method is disclosed.
• Tracking only branded prompts. They almost always name you and tell you little about discovery.
• Deciding from one run. A single answer is one draw; read a prompt over several runs.

What may change

These reports are moving quickly, so date anything you write down. Google added a multimodal search-type filter to the generative AI report on Sept. 24, 2026, and Bing's topic view arrived on Oct. 13. In the UK, the Competition and Markets Authority's interpretative notes on its publisher conduct requirement (June 3, 2026) expect Google to give publishers click-through data from links in its AI features, "fully disaggregated from other elements of general search," through a platform such as Search Console. Query-level data is not on that list, so plan as if the prompt itself will stay hidden.

How SolCrys helps

SolCrys starts from the same reverse method. Each workspace's category gets a Golden Prompt Set built from search demand, the questions people ask around it, and community discussion where available, and prompts carry a type, a persona, and a demand band. You add your own questions alongside them. Prompt Pulse is our public dataset of 1,012 buyer-style questions across 16 industries, if you want a starting list for your category.

Then the prompts run on a schedule across ChatGPT, Google AI Overviews, Gemini, Perplexity, and Claude, depending on plan, and Prompt Analysis shows for each one how often answers name you, which brands they name instead, and which sources they cite. Signal, the weekly market brief, ties each new content idea to a tracked prompt and flags ideas aimed at questions you don't measure yet, which points to prompts worth adding.

To test 10 buyer questions on ChatGPT now, Start Free: 10 ChatGPT prompts and 1 content audit, free, with email verification and no credit card.

What we don't claim

The regex filters find candidates, not proof that a query came from an AI feature; Google provides no way to tell them apart. The reports show phrasings that already reach you; a prompt set measures the intents and personas behind them, so the two work together. And the platform details above are as of Oct. 14, 2026; check each platform's documentation before relying on them.

Sources

• Google: AI Mode U.S. insights (May 19, 2026)
• UC Berkeley: Search Arena, arXiv:2506.05334 (June 5, 2025)
• Search Console Help: Performance report (privacy filtering and row limits)
• Search Console Help: How impressions, position, and clicks are counted (AI Mode)
• Search Console Help: Generative AI performance report (Search)
• Search Engine Roundtable: Google Search Console AI Mode queries in the performance report (Aug. 6, 2026)
• Google Search Central Blog: Web multimodal search in Search Console (Sept. 24, 2026)
• Microsoft Bing: Introducing AI Performance in Bing Webmaster Tools Public Preview (Feb. 10, 2026)
• Microsoft Bing: Introducing Topics Portfolio (Beta) in AI Performance (Oct. 13, 2026)
• UK CMA: Publisher conduct requirement interpretative notes (June 3, 2026)

FAQ

Related guides