The 25 buyer questions AI engines get asked about you — and how to add your own

Your AI visibility score is only as good as the questions behind it. Ask the engines the wrong things and you'll measure the wrong reality — reassuring yourself with vanity prompts while real buyers ask questions you never tested. So this post shows you exactly what our scan asks: how the 25-question probe is built, why the mix is what it is, where a generated set honestly runs out — and how you can now add the questions only you know to ask.

Why questions are the unit of measurement

Buyers don't interact with AI engines through keywords; they ask full questions — "what's the best CRM for a five-person sales team that hates data entry" — and get a decisive shortlist back. So the honest unit of visibility isn't a keyword ranking, it's a question: for this specific question, did any engine name you? That's why the report shows results question by question, and why the headline number reads "named in X of Y buyer questions." What goes into that question set therefore matters enormously.

How the 25 get built

When you enter a domain, the scanner first reads your site and extracts a brand profile: your name, your category, a one-line description of what you sell, and the language your buyers use. From that profile it generates 25 questions written the way a real potential customer would type them into an AI assistant — natural, specific language ("best chatbot widget I can drop into my Flutter app"), not directory-speak ("leading software provider"). A French bakery gets its questions in French, because that's what its buyers ask Gemini.

The questions also deliberately vary buyer context — budget, team size, urgency, location where it's relevant — because "cheapest decent option" and "best premium choice" surface different shortlists, and you want to know your standing in both.

The mix, and why each slice exists

The 25 aren't uniform. The set targets a specific blend:

• Around 14 category and discovery questions — "best X," "recommend an X," "top X for this use case." This is the bulk of the set because it's where deals are decided: the buyer knows what they need and is asking for names. There's no page two; you're in the shortlist or you don't exist.
• Around 4 problem-first questions — the buyer describes the pain and asks what to use, without naming a category. These are the hardest to win and the earliest in the funnel: whoever gets named here frames the whole search that follows.
• Around 4 comparison and alternative questions — "X vs Y," "alternatives to X." The shortlist-narrowing stage, and the shape of question engines answer best from comparison-shaped content.
• Around 3 questions that name your brand — "Is X legit?", "X vs competitors." Here's the honest part: these are excluded from your headline score, because an answer that contains your name only because the buyer typed it can't measure unprompted visibility. The scoring post explains the arithmetic. We still show these answers separately, because what an engine says when asked about you directly — a stale description, wrong pricing, a product you killed — is a real problem you'd otherwise never see.

One rule is absolute across all 25: no placeholders. A question containing "[Competitor A]" or "[City]" isn't a question any buyer ever asked, and a single one poisons the measurement. Generated questions are validated against exactly that.

Where a generated set honestly runs out

A generated probe is a strong proxy — 25 realistic questions, in your buyers' language, across the funnel. But it's a proxy. We build it from what your website says about you; we can't know the exact phrasing from your last lost deal, the objection your sales team hears every week, or the regional wording your market uses that your homepage doesn't. Those questions exist in your call notes and support tickets, and no scanner can guess them.

Until recently that was simply a limit of the product. Now it's an input.

Adding your own questions

Every paid plan — Solo ($19/mo), Starter, and Agency — can now add up to 10 custom questions per brand, alongside the generated 25. They behave exactly like the standard set, on purpose:

• They ride along with every scan of that brand, including the weekly re-scans, so you get a per-question answer trail over time for the questions you actually care about.
• Same four engines, same parsing, same honesty rules. A custom question runs through the identical pipeline — and if it names your brand, it's automatically classified as branded and excluded from the headline score, same as our own brand questions. You cannot buy yourself a higher score by asking the engines about yourself ten times. That would make the number worthless, and the number is the product.
• Same anti-garbage guardrails. Questions must be 8–200 characters of real words; template brackets are rejected; duplicates (case-insensitive) are rejected so a slot is never wasted asking the same thing twice.
• Editable. Rephrase or delete a question anytime; the next scan picks up the change.

What to add first

If you're staring at ten empty slots, start here: the exact question from the last deal you lost to a rival. The "you vs [the competitor your prospects actually raise]" comparison. The phrasing your non-English market uses. The use case sales hears weekly but marketing never wrote a page for. Each of those is a question where the answer is currently being given to buyers, every day, without you in the room — and now you can read it, verbatim, on every scan.

See the standard 25 first

The free scan runs the full standard probe: 25 generated buyer questions across ChatGPT, Claude, Gemini, and Perplexity, with every answer shown question by question — whether you're named, who's named instead, and the exact sentence each engine returned. No signup, about a minute. Run it, read which questions you're losing, and then you'll know precisely which ten of your own are worth asking.