How we measure
Every number we show you comes from a fixed, repeatable method. Here it is in full.
What we measure
Mention rate: the share of answers that name your company, out of every answer we collected. We report it with the number of answers behind it.
Competitor share: the same measure for each competitor you name, from the same answers.
Accuracy: whether what the AI says about you is right (category, pricing model, key features, integrations).
Sources: the websites the AI cites or leans on for your category, because those are usually what has to change.
Why we don't report a 'rank'
AI answers change from run to run. In a January 2026 study by SparkToro and Gumshoe, the chance of getting the same list of brands twice for the same question was under 1 in 100. A position in one answer means little; how often you appear across many answers means a lot. So we count appearances, not positions.
Source: SparkToro, January 2026: AIs are highly inconsistent when recommending brands or products
The questions
Questions are written the way buyers ask them: 'best X for Y', 'alternatives to Z', 'X vs Y', and questions about your use case. We agree them with you, keep them fixed, and re-ask exactly the same ones each time we measure, so before and after are comparable.
Engines and runs
Free check: 10 questions on ChatGPT, Perplexity and Google AI, checked by hand in the consumer apps.
Diagnostic: 30 questions, each run 3 times on each of the 3 tools: 270 answers logged, with sources.
Growth: 40 to 60 questions, re-measured at weeks 4, 8 and 12 with the same protocol.
The instant check on this site uses the engines' developer APIs, one run per question. API answers can differ from the apps (a 2026 comparison by Surfer found only 15 to 32% overlap in the brands named), so we label it indicative and never use it alone for paid work.
How a mention is counted
A company counts as named if its name, or its website, appears in the answer or in its cited sources. Lookalike names are checked by hand in paid work.
Limits we tell you about
Answers vary by account, location, time and model version; we record the date, tool and settings for every run.
Nobody controls what AI systems say. We work on the inputs (your site, the sources AI cites, accuracy of your listings) and measure the outputs; we never promise a specific result.
Small samples are noisy. We show the number of answers behind every percentage and avoid claims a sample can't support.
Data handling
We ask for read-only access where possible and never need your customers' personal data. Raw answers are kept with your project so every number in a report traces back to evidence.