Where should we look?

What a Snapshot measures

A Snapshot tells you whether AI assistants name and cite your company when your buyers ask the questions they ask before they know who to call. That's the whole job. Nothing else gets measured.

We run each question on four engines: ChatGPT (GPT-5 through the OpenAI API, web search on), Claude (Claude Sonnet 5.5 through the Anthropic API, web search on), Gemini (Gemini 3.6 Flash, search grounding on) and Perplexity (Sonar). Those are the engines as of 30 September 2026, and every report states the model ids used, so you can see exactly what was asked of what. Fifteen questions across four engines gives you 60 answers.

Each question is asked once, in a fresh single-turn conversation, from the market's locale. That's London, GB by default, or New York, US on the US edition. So the answers reflect what a buyer in that market would plausibly see.

Three things are recorded per answer:

• Mentioned: your brand or an approved alias is named in the prose of the answer.
• Cited: one of your own domains appears in the answer's source list.
• Rank: the position of your first citation in that list.

Mention and citation are never blended into one score. They fail for different reasons, and a blended number would hide which one you've got. Every matched mention is printed verbatim and checked before it's counted, so you can read the sentence yourself rather than trust a tally.

How your 15 questions are chosen

We read your site and propose 15 prompts across 3 to 5 verticals. You edit every one. You approve every one. Nothing runs until you do, and you're charged only after that approval.

Why go in that order? Because the questions decide everything that follows. A badly chosen question produces a clean-looking number that means nothing, and the buyer is the only person who can spot that early. You know which questions your customers actually ask, and which ones we've guessed wrong from your website alone.

The mix is deliberate. Most of the prompts are unbranded: what a buyer asks before they know who to call. That's where visibility is won or lost, because nobody types your name into an assistant until they already know you. Two prompts are branded, there to check the engines know the company exists at all. One is framed as a comparison.

So if the draft list looks wrong, change it. Swap a question, reword it, drop a vertical. The approval step exists because the list is yours.

What the report shows

The report sets out the counts first: mentions, citations and rank for each vertical, on each engine. Then it gives a verdict per vertical, drawn from quoted claims in the answers themselves. The thresholds are printed next to the counts, so you can see why a vertical landed where it did.

There are three verdicts:

• Recognised, not cited: the engines name you but rarely cite you. That points to a page problem, not an awareness problem.
• Entrenched: one rival appears in the majority of answers. Displacing a name that fixed is a bigger job.
• Open: no vendor is named in more than about a quarter of answers. Nobody owns the question yet.

Open is the verdict people want, but it's also the least settled, since an empty field can fill. Recognised, not cited is usually the most fixable, because the model already knows who you are and the gap sits in which pages it reaches for.

Every claim behind a verdict is quoted. That makes the report longer. It also means you can argue with it, which is the point.

Snapshot or Deep Dive

Start with the Snapshot. It's 15 questions and 60 answers, enough to show where you stand in a handful of verticals and whether the problem is awareness, pages or competition.

The Deep Dive is 100 questions and 400 answers. It makes sense when the Snapshot has shown you a pattern worth testing harder, or when your business spans enough verticals that 15 questions can't cover the ground. Also, if you need a number you can put in front of a board, a bigger sample gives you more to stand on.

If you only want a rough read on one page, you may not need either. There are free tools that need no account and store nothing: an AI citation check for a single page URL, an AI readiness check, an AI-crawler robots.txt check and an llms.txt generator. There are also public industry benchmarks, which run the same method on a set of buyer questions per category, each stating its run date, market and prompt count.

What it cannot tell you

A Snapshot is a sample, not a census. The same question asked tomorrow can get a different answer, and API answers sit close to, but not identical to, what the consumer apps show. Treat the result as a reading of position, not a fixed fact.

It doesn't measure traffic. A citation is a source link, not a click, so a strong citation count doesn't promise visitors.

It doesn't explain why a model named you. We can show what was said and what was cited, but the reasoning inside the model isn't visible to us or to anyone else.

And it says nothing about stability over time. One run is one moment. If you want to know whether a verdict holds, you'd need to look again.