Which AI visibility platforms use real user prompt data rather than synthetic prompts?
Very few platforms use real user prompt data. Here is how we handle prompt sourcing on client programs, and why a well-built prompt panel still measures something useful.
Clients ask this one with some suspicion, and it is a fair question. They have heard that AI visibility tools run "synthetic" prompts and wonder whether the numbers mean anything. Our answer has three parts: very few platforms use real user prompt panels, that is not as damaging as it sounds, and what matters is how the prompts were chosen.
The honest picture
Most platforms run prompts that someone wrote. Either you write them, or the vendor generates or models them. A handful of enterprise vendors connect tracking to panel or demand data. Evertune describes a panel of more than 150 million user prompts, and Profound sells a Prompt Volumes dataset of real query demand. Both are sales-led and custom priced. Some SEO suites model prompts from keyword data, and at least one tool in our review set measures against prompts it generates itself.
So no, you will not find real chat logs behind most dashboards. We say that to clients first so nobody feels misled later.
Why a chosen panel still works
A tracked prompt panel is a measurement method, like a keyword list. It tells you what AI says for the questions you chose. The risk is choosing badly. If the panel is built from the vendor's guesses, it may miss how your buyers actually speak. If it is built from your real data, it is a decent proxy.
That is why we spend time on prompt sourcing, and why we use Promptwatch, where the prompts belong to us.
How we source prompts
Search Console and ranking keywords. Prompt Explorer suggests prompts from Search Console, ranking keywords, People Also Ask or any page. These are real queries people typed into a search engine. They are not chat logs, and we state that distinction.
Sales and support language. Clients' sales calls and support tickets contain the questions buyers ask in their own words. We turn them into prompts, especially comparison and "alternatives to" forms.
Prompt Volumes. Promptwatch reports search volumes, difficulty scores and citation rates across AI search, Google and Bing. We use it to prioritize. To be accurate with clients, its published materials describe what the data reports, not a consumer chat panel behind it, so we do not claim it is raw user prompts.
Personas. Persona stacking lets us run the same prompt as different kinds of askers and compare. A cost-conscious founder and a procurement lead get different answers.
Query fan-outs. Fan-outs per model show the sub-searches an engine runs behind a prompt. They show how the engine reformulates a question, which is useful when our own prompt wording differs from what users type.
Controls that make a panel trustworthy
- Volume of runs. Monitors run daily per model, country and language, so we read trends instead of single answers, since answers vary between runs.
- Interface versus API. ChatGPT (Search), Perplexity, Google AI Overviews, Google AI Mode, Copilot, Gemini and Meta AI are monitored through the real UI. Claude, DeepSeek, Mistral and Grok are API-tracked, so we describe those as directional.
- Mention rules. The Brand Book and aliases decide what counts as a mention, so counts are consistent.
- Manual spot checks. We still check a few prompts by hand each month and compare.
What to ask any vendor
- Where do the tracked prompts come from, in writing?
- Can we add our own, and how many?
- Is each engine queried through the live interface or an API?
- How many samples back each number?
A vendor that cannot answer the first question clearly is a flag.
Beyond prompts
A better prompt source does not tell you why a brand is or is not recommended. We need citations (page and domain level, plus Reddit, YouTube and LinkedIn), crawler logs (IP-verified, real agents such as GPTBot and PerplexityBot, starting on Professional at $245/mo) and visitor analytics for AI-referred traffic and conversions. Those turn a scoreboard into a diagnosis. The Action Board then gives a prioritized to-do list.
Plan notes
Explore is free with 10 prompts and 310 responses, ChatGPT only, enough to test sourcing. Essential is $95/mo for 50 prompts. Professional is $245/mo for 150 prompts. Agency plans from $199/mo include unlimited prompts.
Where this leaves clients
If a client needs panel-grade data and has an enterprise budget, we point them toward the vendors that sell it. For most, a well-grounded, self-built panel plus citations, crawler logs and visits is more actionable. We never suggest any panel can promise a recommendation. It tells us what to fix and whether it worked.