What AI says about you, and how we know

We take the questions your customers ask, put them to ChatGPT, Perplexity and Gemini the way somebody near you would ask them, and keep what each one said. Then we read each answer for your business name.

That is a measurement, not work on your business. It is the only honest way to tell whether everything else we do moved anything.

Yours. You pick the searches you want to win, and up to ten of them are asked each month. Where you have picked more, they are ranked by what they are worth commercially and near-identical phrasings are collapsed, so ten different questions get asked rather than ten wordings of one.

Each one is anchored to your town before it is asked. The answer we record is the answer somebody near you would have got, and your page prints the town the question named, so you can tell whether we asked about your area at all.

• ChatGPT: asked in a real browser, logged out, and the rendered answer is screenshotted.
• Perplexity: the same, a real browser and a screenshot.
• Gemini: answered over its API. Text, no picture, because there is no headless capture for it.

Where a screenshot exists, your page shows the screenshot, because a photograph of the assistant saying it is stronger than our transcript of it. Where there is none, the answer is drawn inside that assistant’s own frame around the genuine recorded text, and the card says it is a recreation.

Named means the answer used your business name. Not that it linked to you, not that it described something like you. Your name, in the text the assistant produced.

We also keep every other business the answer named, in the order it named them, because the answer’s own order is the only ranking an assistant gives. That is where "who got named instead" comes from, and it includes competitors we hold no record of, because a rival we do not list is still the business taking that question.

Once per calendar month per question, so nothing is bought twice. Starting a paid plan also fires a capture straight away rather than making you wait for the monthly pass, so your answers page has real captures on the first day.

The page answers two questions. Are the assistants naming you more than they used to, and what did we do about it.

The first is a line per assistant: the share of the questions it answered that named you, one point a week. The scale runs from zero to a hundred and stays there, because a scale that floats to the peak turns four points between 8 and 11 percent into a mountain range. The first week we hold is the baseline, and the sentence beside the line names that week rather than implying the count started at zero.

The second is a dated list of everything delivered: the post that went up, the review answered, the service copy that reached your listing, the directory you went live on. Each row links to the live thing. Only delivered work is on it, so a draft waiting in your queue is not counted.

Running more than one location, there is a roll-up across all of them. It adds the questions and the wins before taking the share, rather than averaging the percentages, so a location asked forty questions is not outvoted by one asked four.

Your AI answers page is the evidence behind the line: every captured answer, per question, per assistant. The verdict, the quote around your name where there is one, the full answer text, and the screenshot.

A month with no capture stays empty rather than being filled with a guess.