Tracked prompts are more reliable than sceptics think

There's a recurring objection to AEO platforms: the data comes from tracked prompts, not real conversations people are having with ChatGPT, Gemini or Perplexity.

Most AEO tools run a curated list of prompts through AI models on a schedule, then check whether your brand gets mentioned or cited in the response. Sceptics point out that this isn't the same as seeing what real buyers type into ChatGPT.

True.

But the conclusion people jump to next, that tracked-prompt data is therefore unreliable, misses how those prompts are actually sourced.

The prompts come from real query data, not guesswork

Practitioners increasingly build prompt lists from data that already exists, rather than inventing questions from scratch. Some of that data is now genuinely conversational, not just tracked.

Three sources come up again and again:

Google Search Console. Long-tail, conversational queries already sit in GSC data. Marketers are pulling these directly to build prompt sets, reverse-engineering real buyer intent from questions people already typed into Google.

GA4 landing pages and on-site search terms. The pages people land on and the terms they search for on your own site reveal real buyer language, questions like "best estate agents in Bristol", "best hybrid bikes for commuting", or "best accountants for small businesses".

Licensed conversation panels. Profound's Prompt Volumes product licenses anonymised prompts from double opt-in consumer panels of real AI assistant users, hundreds of millions of prompts a month, scrubbed of personal information and compliant with GDPR and CCPA. That's not a tracked prompt at all. It's an actual conversation someone had with ChatGPT, Gemini, Claude or Perplexity.

Feed any of this into prompt selection and you're not guessing at buyer language. You're tracking prompts built from evidence of what people are already asking, in some cases from the conversations themselves.

There's no ceiling on scope

Scale isn't the constraint people assume it is.

• SE Ranking recommends starting with 20 to 40 prompts, run across 2 to 3 AI models, tracked for at least 30 days
• Profound sees most brands tracking 100 to 1,000 prompts, with a couple hundred being typical
• Ahrefs' Brand Radar runs on an index of hundreds of thousands of real ChatGPT prompts, alongside your own custom set

None of that points to a fixed, narrow list incapable of capturing real buyer behaviour. The constraint is effort and budget, not the methodology.

Why getting the prompt set right matters commercially

Brand visibility behaves differently depending on which stage of the buyer journey a prompt sits at. One analysis found 94% brand mention rates at the point of conversion, but only 48% retrieval and 48% citation. At the awareness stage, mentions sat at 60%, while retrieval and citation were close to zero.

That means the same tracked-prompt approach can look completely different depending on which stage your prompt set covers. A vendor blending awareness and conversion prompts without separating them will hand you a skewed picture, whatever the prompts are built from.

Getting this right is worth the effort. Semrush found that visitors arriving from AI search sources such as ChatGPT convert at 4.4x the value of traditional organic search visitors. Yet government research published in February 2026 found only 16% of UK businesses were using at least one AI technology, with 80% neither using it nor planning to.

Most brands are still working this out. The ones that get their prompt methodology right now have room to build a real lead.

The question worth asking instead

The old framing, tracked prompts versus real conversations, is getting weaker by the month. Some vendors now license anonymised real conversation data from opt-in consumer panels, rather than relying on tracked prompts alone.

The real question is how rigorously the prompt set is built, how much of it is grounded in evidence or real conversations, and how transparent the vendor is about where the data comes from.

A platform pulling from licensed conversations, Search Console data, GA4 behaviour and buyer-stage segmentation is doing something meaningfully different to one running a generic, off-the-shelf template list with no visibility into its sourcing.

FAQs

Isn't tracked-prompt data just guesswork?

Not when it's built properly. Prompts sourced from Search Console queries, GA4 data and buyer-stage research are evidence-based, not invented.

Why don't AEO platforms just use real ChatGPT conversations?

Some now do. Profound's Prompt Volumes licenses anonymised prompts from double opt-in consumer panels of AI assistant users, rather than relying on tracked prompts alone. It's still the exception rather than the norm across the category, and coverage varies by engine and region, but the assumption that real conversation data is completely off-limits is out of date.

Does this mean visibility scores are unreliable?

No, it means the quality of a vendor's prompt methodology matters more than most buyers realise. Ask how prompts are sourced, how many are tracked, and how they're split by funnel stage before comparing visibility numbers between platforms.