Quick answer: Track AI visibility with a fixed set of customer prompts and a log of the exact responses. Record the platform, date, language, settings, brand mentions and linked URLs. Repeat observations, separate branded from unbranded prompts, and report results as a sample. Keep citations, referrals and qualified enquiries as distinct measures.
In this guide: Define what counts before collecting responses · Build a balanced prompt set · Copy this tracking-sheet structure · Calculate rates with visible denominators · Account for variability · Connect monitoring with the editorial backlog · Frequently asked questions
A screenshot of one favourable answer can be encouraging, but it is not a monitoring system. A useful system lets another person repeat the method and understand what changed.
You can begin with a spreadsheet. A specialist tool may help at larger scale, but the underlying definitions still matter. A dashboard cannot correct an unclear prompt set or a misleading denominator.
Define what counts before collecting responses
A mention should identify the intended business, not an unrelated organisation with a similar name. A citation should include a link to the relevant domain or page. A recommendation should be recorded separately because an answer can mention a brand as an example, caution or comparison without recommending it.
Write simple coding rules. For example, record mention context as “informational,” “shortlisted provider,” “negative,” or “ambiguous.” Keep the original response so a judgement can be reviewed later.
Do not count an assistant repeating your brand name from the prompt as evidence of unsolicited discovery. Branded accuracy checks have value, but they answer a different question.
Build a balanced prompt set
Use questions from actual customer research. Include the problems people are trying to solve, comparisons they make and questions used to assess a provider. Avoid a list consisting only of “best agency” prompts with minor wording changes.
For a small business, an initial set could contain five problem questions, five solution comparisons and five provider-selection questions, with a separate group for brand accuracy. These counts are a manageable starting suggestion, not a statistical requirement.
Version the set. If you replace half the prompts next month, do not attribute the entire resulting difference to a content improvement. Keep the original set available for comparison.
Copy this tracking-sheet structure
| Column | What to enter |
|---|---|
| Prompt ID and version | A stable identifier for the exact question. |
| Intent | Problem research, comparison, provider selection or brand accuracy. |
| Platform and mode | The product and visible model or mode used. |
| Date and time | When the response was collected. |
| Language and location context | The language and any location stated in the prompt. |
| Search and account context | Relevant visible settings and whether a fresh conversation was used. |
| Brand mention | Yes, no or ambiguous, with the context. |
| Citation URL | Every relevant linked source, preserving the exact URL. |
| Recommendation | Whether the business was actually suggested as a provider. |
| Evidence | Saved response text or screenshot reference. |
| Notes | Errors, uncertainty and unusual conditions. |
When using automated collection, check the platform's current terms and the capabilities of the approved tool. Do not assume an API response represents the consumer app experience.
Calculate rates with visible denominators
An observed mention rate equals responses containing a verified brand mention divided by the number of eligible recorded responses. If nine of 60 responses mention the agency, the rate is 15% for that sample.
Calculate citation rate separately. If six of the same 60 responses link to the domain, the observed citation rate is 10%. A response containing several links to the same domain still counts as one cited response under this definition; you can also report unique cited pages separately.
Exclude failed runs only under a documented rule and disclose them. Keep branded and unbranded results separate. Report platform and language breakdowns before combining them into a total.
Account for variability
Repeat prompts when practical and retain all runs. Record changes to models, modes, account settings or location context. These can make periods less comparable even when your website is unchanged.
If a rate rises from two mentions to four in a small sample, report the counts. Calling that a 100% improvement without context overstates the precision. A manual sample can reveal useful patterns, but it does not measure every user or establish a fixed rank.
Review which pages are cited and for which questions. A single relevant buying guide may provide more actionable insight than a larger number of loosely related mentions.
Connect monitoring with the editorial backlog
For each recurring information gap, identify the page that should address it. Improve factual clarity, evidence or accessibility where appropriate. Record the publication date and specific change.
Then repeat the monitoring method without assuming causality. If visibility changes, consider other explanations and continue observing. Pair the log with identifiable referral traffic and CRM outcomes to understand commercial relevance.
Frequently asked questions
Is “share of voice” the same as market share?
No. A tool's share-of-voice metric usually depends on its selected prompts, platforms and counting method. It is not the share of all customer demand unless a defensible methodology establishes that.
Should we monitor competitors?
You can record which other businesses appear in the same relevant responses. Apply consistent rules and use the findings to understand information gaps, not to invent claims about their performance.
Can a spreadsheet be enough?
Yes, for a manageable prompt set. Clear definitions and saved evidence are more important than the complexity of the dashboard.
Sources and further reading
Platform guidance checked on 12 October 2026. Frameworks and illustrative examples are editorial guidance; they are not platform ranking formulas or reported client results.
Continue reading
- How Do You Measure Marketing Results When Customers Discover You Through AI?
- How Can Your Business Become More Discoverable in ChatGPT Search?
- How Can Your Business Improve Its Visibility in Google Gemini?
Plan your next step. Discuss your website and marketing priorities with Stefano Marketing Agency. We can clarify your goals and agree on the appropriate scope.