A mention is different from a source link
An AI answer may name your business without linking to it. It may also cite an article on your domain without recommending your brand. Record these outcomes separately. A third context column shows whether the answer presents you as a suitable choice, an alternative or inaccurately.
Define a simple coding rule in advance. Count your business at most once per answer as a mention. Count source links separately. Save the answer or a reference to the measurement output. This prevents one answer naming you three times from outweighing three separate answers.
| Measurement question | Record | What remains unknown |
|---|---|---|
| Is my business mentioned? | Answer, brand name and context | Whether anyone visited your website |
| Is my page cited? | Source URL and associated question | Whether the source led to an enquiry |
| Does it bring customers? | Visit, enquiry or order | How much came from one content change |
Build a question set that reflects your customers
Include exploratory, comparison and purchase-focused questions. Use broad questions without your brand name and a smaller, separately reported branded set. Otherwise, visibility may appear high simply because you steer the answer towards your business.
An online shop might compare questions about a quiet coffee grinder for a small flat. A service business might start with an installer for a home with limited outdoor space. Make the situation concrete without inventing preferences that only your offer can satisfy.
Record the language, region, platform, date and search mode. Save failed requests too: an error is not an answer without a mention. If you change the question set, report the new series separately or measure both sets alongside each other.
Download the measurement log (CSV)Calculate the share and show uncertainty
Mention share is the number of answers mentioning your business divided by the number of valid answers. With 24 mentions in 100 answers, the share is 24%. Six mentions in 20 answers gives 30%, but the smaller sample provides much less certainty.
Our calculator adds a 95% Wilson interval. For 24 out of 100, this is approximately 16.7% to 33.2%. The interval describes sampling uncertainty under a model of independent, comparable observations. It does not correct for biased question selection, repeated near-identical prompts or platform changes.
Do not call this share a universal GEO score. Report its scope alongside it: which questions, platforms and period. Compare platforms separately and avoid market-wide conclusions from a self-selected list.
Calculate the share and intervalUse Google and Bing data too
Google now documents a separate Generative AI performance report in Search Console, rolled out globally from 31 August 2026. It shows impressions in AI Overviews and AI Mode, including page, country and device dimensions. The report may be absent if impressions are insufficient. It is not a record of all prompts or ChatGPT mentions.
Bing Webmaster Tools offers AI Performance for citations in supported Microsoft experiences and partner integrations. It includes cited pages and grounding queries. Bing explicitly describes this as aggregate citation activity, not a ranking or a complete log of every AI answer.
Use prompt measurements to investigate specific customer questions and platform reports to track broader patterns on your own domain. Keep definitions separate in your report; a Bing citation and a Google impression are not interchangeable units.
Turn the report into a work plan
After each measurement round, select a limited number of questions. Check inaccurate business information first, then missing answers around important offers. Review recurring source pages and what they explain. Do not copy their text; identify the evidence or explanation you can add.
Record the owner, changed URL and review date for each action. Include visits, enquiries and orders where you can measure them reliably. Increased visibility without relevant enquiries may warrant reviewing the question selection.
Keep raw answers alongside the summary. You can then investigate whether an increase reflects better context, different source selection or merely more isolated name mentions.
Explore cognito's measurement approachFrequently asked questions
How many prompts do I need?
There is no universal minimum. Choose enough questions to cover your important offers and customer decisions, and report the sample size. More near-identical prompts do not automatically make a better sample.
Does the calculator measure my website automatically?
No. You enter the number of valid answers and mentions yourself. The calculator computes the share and statistical interval; it does not run AI queries.
