How to Find Prompts That Mention Your Brand on AI Platforms

Can you see the real prompts people type?

Not from the AI platforms. AI companies do not share users' conversations with the businesses mentioned in them, for privacy reasons. Two kinds of data come close. Prompt databases built from opt-in consumer panels or consented clickstream data, such as Profound's Prompt Volumes and Semrush's AI Visibility Toolkit, show samples of real prompts in your category and which ones produce answers that name you; they cover panel members only, and their volumes are modelled. And platform reports and your own data show traces:

• Bing's grounding queries. Bing Webmaster Tools' AI Performance report shows the search phrases Copilot used when it retrieved and cited your pages. These are not users' prompts, but they show how the engine interpreted questions that led to you.
• Search queries in Search Console, which include AI Mode queries mixed in with ordinary searches, unlabelled, along with long question-style searches; some long queries come from rank trackers.
• AI referral landing pages in your analytics. ChatGPT adds utm_source=chatgpt.com to links clicked in its search results, which shows which pages drew clicks.
• What customers tell you when asked what they searched for or asked an assistant.

Everything else is inference and testing, so treat any prompt list as a sample, not the full picture.

Where do candidate prompts come from?

Assistants themselves can help brainstorm variations of a question, but their suggestions show what is plausible, not what people actually ask.

How do you ask customers what they asked AI?

Your own customers are the most direct source of real prompts about your brand, and most will tell you if asked simply:

• On sign-up or enquiry forms, add "AI assistant" to "How did you hear about us?" and, if chosen, an optional box: "What did you ask it?"
• In sales calls, ask early: "How did you research options? Did you use ChatGPT or another assistant? What did you ask?"
• In onboarding surveys, ask what alternatives they considered and where they heard about them.

Record the wording exactly, without tidying it up. Even twenty real prompts collected this way often reveal needs and phrasing that no keyword tool shows.

How do you turn candidates into a tested prompt set?

1. Collect and deduplicate candidate questions from the sources above.
2. Rewrite them conversationally, as a buyer would ask an assistant: "I run a 12-person design agency; what's a good project management tool that integrates with Xero?"
3. Add realistic constraints, such as budget, team size, location, industry or a must-have feature.
4. Group by intent and stage: learning, shortlisting, comparing, checking, buying.
5. Include unbranded and branded prompts, tracked separately. Branded prompts inflate visibility figures, as covered in whether brand-name prompts inflate AI visibility scores.
6. Run each prompt several times in each engine, with fixed settings.
7. Label results: named, recommended, cited, absent, and which competitors appear.
8. Keep a core set of 30–100 prompts for regular tracking, and a wider exploratory set for occasional checks.

For companies with many products, a structured taxonomy helps, as covered in how to create an AI search prompt taxonomy for a multi-product company.

What should each prompt record contain?

Store each prompt with its source, the date added, its intent and stage, the buyer type it represents, whether it names your brand or a competitor, and the products or pages it relates to. After each test, add the engine, model, run count, mention rate and cited sources. This makes it possible to see later which sources produced the most useful prompts, and to filter reports by product, stage or market.

What about prompt volume figures from tools?

Some visibility tools estimate how often prompts are asked. These are models, usually built from search keyword data, clickstream panels or survey data. They are not counts from the AI platforms. Use them to rank topics roughly, not as precise demand. When a tool shows "prompt volume", ask how it is calculated and how often it is updated.

How do you find prompts where competitors are named but you are not?

• Run competitor-branded prompts, such as "alternatives to [competitor]" and "[competitor] vs...", and record whether you appear.
• Run category prompts and list every brand named across runs.
• Look at the sources cited when competitors are named. They show which pages and third-party sites carry the evidence you lack.
• Rank the gaps by business value, and plan content or outreach for the top ones.

How do you keep the prompt set current?

Review it quarterly. Add prompts for new products, features and markets, and for questions sales and support hear more often. Retire prompts nobody asks any more, but keep a stable core for trend reporting. After launches or rebrands, add prompts that use the new names and check old names too. Alerts for changes in how you are mentioned are covered in how to set up alerts for brand mentions in AI responses.

What might an illustrative case look like?

Consider an illustrative accounting software company whose first prompt set was 40 variations of "best accounting software". It is mentioned in most of them, and the team assumes it is well covered. Rebuilding the set from sales call notes, support tickets and Bing grounding queries produces different prompts, such as "accounting software that handles construction retentions" and "which accounting app works with my bank in Ireland". Testing shows the company is absent from most of these, while two niche competitors appear. The team adds pages for construction retentions and Irish bank feeds. It also tracks the new prompts monthly and reports them separately from the generic ones.

What do people ask about finding brand prompts?

Can I buy a list of real ChatGPT prompts?

Not from the AI companies, which do not sell user prompts. Some tools sell access to prompts from opt-in panels or consented clickstream data; ask how the panel was recruited and how volumes are estimated, and treat them as a sample.

How many prompts should we track?

A core of 30–100 for regular reporting, depending on how many products and markets you cover.

Should prompts include personal details?

Include the kind of context buyers give, such as team size or location, but not real personal information.

Do prompts need to be long?

Include the constraints buyers typically mention. Many real prompts are longer than search queries.