Real answers, checked and re-checked

What we actually check

We ask ChatGPT, Claude, Gemini and Perplexity the questions your customers actually ask — “best X providers”, “who should I use for Y”, “is Z legit” — and record the answer each one gives. The same answer a real customer would see, word for word. Not estimates, not a simulation, not a score invented in a spreadsheet.

We sample what Google itself shows for the same questions too — both its AI Overviews and its AI Mode answers, sampled from your market's Google results. Like every AI answer, they vary between asks and are shaped by context we can't see, so we report them the same way as the rest: a sampled rate, never a guarantee.

From every answer we record who got named, who got recommended, whose website the answer leaned on, and whether anything it said about your business is actually true — checked against the facts you've confirmed with us. We also check whether AI's research can find your site at all, because you can't be in an answer you weren't found for.

Why we never trust a single check

Ask an AI platform the same question twice and you'll often get two different answers. That's not a flaw in our measurement — it's how these systems work. Any tool that shows you one green tick from one check is hoping you don't know that.

So we don't check once. We ask each question again and again across the month, worded the different ways real customers word it, and tell you how often you appeared: “named in 7 of 20 answers this month, up from last month.” And when we don't yet have enough answers to give you a fair percentage, we say “early read” — we'd rather tell you it's early than dress up a guess. AI answers vary between asks, so we report sampled rates and trends, never guarantees — and the trend is the number worth watching.

How often we check

Your core topics are checked every week, with a wider set of buyer questions checked through the month. Some tools check daily and charge accordingly. We think that's mostly theatre: asking the identical question every day measures the same noise seven times over. Weekly checks with varied wording tell you more, for less of your money — so that's what we do.

When something genuinely changes — you stop appearing for a topic you'd been winning, a new competitor starts showing up, an answer gets a fact about you wrong — we email you, with the fix attached. And we'll never panic you over one odd answer: a change has to hold before we call it a change.

Being named isn't being recommended

There's a world of difference between an answer that says “options include A, B and you” and one that says “for a business like this, use you”. So we don't just tell you whether you appeared — we tell you how. These are the exact words you'll see on your dashboard, and on the free check on our homepage:

• Top pick— AI put you forward first, as the best choice.
• Recommended— AI suggested you as a good option, though not the top one.
• In the mix— AI listed you as one option to compare among others.
• Just named— Your name came up, but not as a suggestion.
• Absent— AI answered the question and did not mention you at all.

Separately, we flag Described incorrectly — AI stated something untrue about you — because that's a different problem from not being there, and it needs a different fix. Plus your share of all the names given, and where in the list you tend to sit.

One thing we deliberately don't do: “sentiment scores”. Tools that claim to grade whether AI feels positive about you misfire constantly, and a false alarm sends a small business chasing a problem it doesn't have. Whether you were recommended, merely named, or missing — that's what you'd actually act on.

What we don't claim

• No guarantees. Nobody can guarantee AI will cite or recommend a business — anyone who says otherwise is selling something. We measure, we show you the trend, and the work either moves it or we say so.
• No forever. Appearing in an answer this week doesn't mean appearing forever. AI re-reads the web constantly — which is exactly why we track the trend rather than framing any single result.
• No guessed demand. We don't estimate how many people ask each question, because we don't have that data and a dressed-up guess would poison everything else on this page. Connect your Google Search Console instead and we'll use your real numbers.