How Findabl Measures AI Citations

What we ask

Generated unaided buyer questions name a need or a category and filter out your brand, so a mention is something the engine chose. Questions are generated from your website and can be edited. An optional answerability check tests whether an engine can answer them; Google Suggest checks for similar search wording. These checks are signals, not proof that people ask the exact question.

Which engines answer

Perplexity (Sonar Pro), an OpenAI model with live web search, and Google Gemini. We call these models through APIs with web search enabled; answers can differ from the consumer apps, which have their own account context, tools and system prompts. Claude is available as an add-on. Google AI Overviews are tracked separately on Growth and Pro and never mixed into the AI citation rate.

What counts as a citation

Completed scans save a representative answer for each question and engine, along with cited source domains. A citation means the answer mentions your brand or website, after brand-identity checks. What each engine says when asked about you by name is shown apart from the category questions, with an accuracy label (correct, hallucinating, wrong entity, no information) and a warning when AI may be confusing you with a same-named company.

How to read changes

AI answers vary between identical asks; our June study measured that directly. After four comparable scans with measured citation rates, the dashboard shows typical scan-to-scan variation as a plus-or-minus band based on the sample standard deviation. A change inside that band is within typical variance, not proof of progress or decline. Engine labels summarize recent comparable scans as stable or volatile, with a one-sample state until more history is available. Deep Scan on Growth and Pro asks every question three times and uses a majority vote to decide whether your brand was mentioned; the displayed answer is a representative sample, not a merged answer.

Cadence

Citation monitoring is opt-in. Eligible monitored projects are scheduled weekly. Standard free scans use Perplexity only with up to five questions.

What we do not claim

We do not model visibility from search data, we do not claim an engine's answer is the same for every user, and we do not claim a change caused a citation. We show the answers and the sources.