How we measure search and answer-engine visibility
The principles behind every number in SERP Checker: what we measure, how a single check runs, how the score is computed — and what we will not do.
01What we measure
One visibility pipeline, three signal families. Search positions: where your site ranks in Google and Bing for each tracked keyword, per location, language and device. Answer-engine mentions: whether ChatGPT, Gemini, Claude, Perplexity, Copilot and Google Overview name your brand in their answers, and which sources they cite. Reputation signals: how your brand shows up on review platforms, forums and media results.
The pipeline stores the three families side by side, each dated, so a ranking and the mention it competes with are read against each other — not in separate tools with separate truths.
02How a measurement runs
Every check fixes four things before it runs: the date, the engine, the exact prompt and the model version. The answer is stored verbatim, together with its citations and a screenshot, so what we scored is what you can read.
If a check fails — a provider error, a CAPTCHA, an unreachable endpoint — the run is marked no_data and is never counted as a zero position or a lost mention. A failure is not a measurement. Reports state coverage explicitly, as "N of M", so a gap is disclosed rather than averaged away.
03The visibility score
The visibility score is position-weighted: a mention near the top of an answer weighs more than one buried at the bottom, and not being mentioned weighs nothing. It is computed from the stored raw answers — the same rows you can open and read — so it is reproducible: same data, same score.
Every number on a report page is recomputed from the raw stored data when the page renders, never copied from a cached figure. The raw answers are available through the public API documentation, so an outside system can reproduce any published number.
04What we do NOT do
We do not use clever prompt tricks to force answer engines to rank a brand. We do not sell paid placements in answers — ours or anyone else’s. We do not guarantee placement: answers vary by date, model and even wording, so anyone promising a fixed spot is selling a screenshot, not a pattern.
What we sell is the repeatable pattern: measurements across dates and models, with the variance shown instead of hidden.
05How we keep ourselves honest
Every figure we advertise must survive our own audit. The release gate runs scripts/verify_numbers.py, which recomputes each advertised number from its formula on frozen inputs and fails the release if any number moved.
Reports disclose when measurements were taken and what they covered — "N of M" — so you can tell a measured drop from a missed check. If the evidence behind a number cannot be shown, the number does not ship.