Start with the questions you track.
Your property’s active prompts define the sample. A prompt is a question sent to an AI platform. Its wording and location affect what gets measured. Choose questions that reflect the decisions your buyers make, then keep that set stable when comparing periods.
A property tracks three platforms from ChatGPT, Perplexity, Google AI Overview, Google AI Mode, Gemini and Microsoft Copilot. Heading collects answers for your tracked questions. Copilot tracking reads the AI summary in Bing results. These observations are not a feed of private customer conversations or a census of every AI search.
Paid properties are scheduled for one collection per active prompt and tracked platform per UTC day. Free properties receive a point-in-time snapshot. A snapshot cannot establish a trend.
Explore prompt research →Count the answers with a detected mention.
Heading checks the answer text for your brand name, configured aliases and domain. Matching is case-insensitive and uses word boundaries. A matching answer counts once, even if your brand appears several times. A match does not establish a positive recommendation or a correct description of your brand.
Daily visibility, for one platform
Answers with a detected mention ÷ recorded responses × 100The denominator is the recorded responses for active prompts in that UTC day. It includes a recorded result with no AI answer. A fetch that fails without producing a response is absent from that denominator. Check collection coverage before comparing scores.
The daily aggregate across platforms pools the recorded responses. Period summaries average daily visibility values. The 7-day and 30-day rolling averages use days with data within each window; missing days are not filled with zero. Filters change the sample, so retain the same prompt, platform and date selections when comparing results.
Ten recorded responses. Six matches.
This is an illustrative calculation, not customer data. Assume one platform records a result for each of ten active prompts on one UTC day.
| Recorded result | Responses | Matches |
|---|---|---|
| Brand name appears; website is cited | 4 | 4 |
| Brand name appears; website is not cited | 2 | 2 |
| Website is a source; no name, alias or domain appears in answer text | 1 | 0 |
| Answer has no matching mention or website citation | 2 | 0 |
| Platform returns no AI answer | 1 | 0 |
| Total | 10 | 6 |
6 matching responses ÷ 10 recorded responses × 100.
The citation-only result contributes no detected mention. The no-answer result stays in the denominator. If that last result were instead a failed fetch with no stored response, the observed score would be 6 ÷ 9, or 66.7%, with incomplete coverage. That difference would not show improved brand performance.
A mention and a source link tell you different things.
A mention is a text match. A citation is a source URL supplied with the answer. Heading matches source URLs to the property’s domain, including subdomains, to identify citations of your website. A source can link to your site without naming your brand in the answer text. That link alone does not increase the mention-based visibility score.
Citation counts and visibility percentages have different units. One response can contain several source links. A cited third-party article may discuss your brand but is not a citation of your own domain. Inspect the original answer and source URLs to understand what was actually said and linked.
A citation does not prove that someone clicked, visited or bought. Referral traffic and conversions require separate evidence.
Explore citation tracking →A missing answer is different from missing data.
Some search results contain no AI answer. Heading records that outcome as a response with no surfaced brand. Collection errors are different: missed fetches are retried, and a result can arrive later. Daily results use the scheduled UTC run date, including when a fetch completes after midnight. Recent totals can change as collection and analysis finish.
Repeating the same prompt can produce a different answer. Heading’s daily sample does not measure every possible response. Your own signed-in session may also differ because of conversation history, personalisation, location or changes to the platform.
Read a change with its sample attached.
- Keep the comparison consistent. Adding prompts, changing platforms or changing location can move the score without any change to your website.
- Check the underlying answers. Literal matching can miss an unconfigured name or match a word used in another sense. Review aliases and inspect ambiguous matches.
- Allow for small samples. In the ten-response example, one match changes visibility by ten percentage points. A single day is not proof of a lasting change.
- Compare competitors on the same questions. Scores from different properties, prompt sets or industries are not directly comparable.
- Separate observation from cause. A higher score after a content update does not prove the update caused it. Platform changes and source selection can also affect answers.
Visibility is not market share, search volume, revenue or a forecast of ranking. Heading does not attach a statistical confidence interval to this score. Use it to find changes worth investigating, then return to the answer evidence.
