Quick answer. We ask the AI engines the real questions your buyers ask, several times each, then measure how often you get cited, mentioned, and recommended. We roll that into a 0 to 100 score and show a confidence interval on it. No fake precision.
Step 1: we run real buyer prompts
We do not check vanity terms. We track the questions a customer actually types, like "best CRM for a small agency." Those are the moments where you either show up or a competitor does.
Step 2: we sample, not snapshot
One run means nothing. The same prompt can cite you at 9am and skip you at 9:05. So we repeat each prompt across engines and look at the pattern. More runs, tighter range.
Step 3: we score what matters
The score leans hardest on real citations (your domain used as a source), because those are earned and hard to fake. Then coverage (are you mentioned at all), prominence (how early), and sentiment (do they speak well of you). Mentions are easy to game, so they count for less.
Step 4: we show the uncertainty, every time
Every number ships with a 95 percent confidence interval. If the sampling is too thin to mean anything, we refuse to print a score and say so. And a week-to-week change only counts as real if the intervals do not overlap. No dressing up noise as a win.
Step 5: we show our work
You can see the raw answers behind every result. Not a black box that grades its own homework.
FAQ
Why does the score come with a confidence interval?
Because AI answers are not deterministic. Ask the same question twice and you can get different sources. We run each prompt several times and report a range, so you are looking at signal, not a single noisy roll of the dice.
What counts as a citation?
Your own domain showing up as a cited source in the engine's answer. We track that separately from a plain mention (your name in the text) and a recommendation (the answer actually tells people to pick you).
Which engines do you measure?
Perplexity, Google AI Overviews, Gemini, and ChatGPT via OpenAI web search. We label the ChatGPT one honestly as a web-search proxy, and we cover Copilot through Bing Webmaster signals rather than pretending we can read the app.
Can you guarantee I will get cited?
No, and anyone who says they can is selling something. We measure where you stand, show you who gets cited instead, and tell you the specific source to go earn. The rest is the work.