Robot VisibleJuly 11, 2026 · 7 min read
How to track AI visibility over time
A measurement framework for tracking AI mentions, citations, cited pages, competitors, technical changes, referrals, and business outcomes without relying on one opaque score.
Quick answers
- How do I track changes to my website's AI visibility over time?
- Monitor a stable set of customer questions and save the provider, date, full answer, brand mentions, cited URLs, and competitors for every run. Combine this with technical crawl history and a website change log, then compare several runs before and after meaningful changes.
Build a query panel before a dashboard
Start with the questions a buyer asks before they know your brand: category discovery, problem diagnosis, alternatives, comparisons, and purchase questions. Keep a stable core panel so one run can be compared with the next, while reviewing and expanding it when products or customer language change.
For every run, preserve the exact query, answer engine, date, locale, answer text, brand mentions, cited URLs, and relevant competitors. AI answers vary, so a screenshot or a single percentage without this context is weak evidence.
Track the metrics in layers
- Technical eligibility: successful fetches, index status, canonical coverage, sitemap health, and material regressions.
- Query coverage: the percentage of monitored questions with a valid captured response.
- Mention rate: the percentage of answers that name the organization or product.
- Citation rate: the percentage of answers that cite at least one URL on your domain.
- Cited-page distribution: which pages earn citations and whether the intended page wins each topic.
- Competitor share: who appears or is cited for the same questions, tracked at query level.
- Outcomes: referrals, engaged visits, scans, sign-ups, leads, or revenue attributable to AI discovery where measurable.
Connect changes to outcomes
Maintain a change log for meaningful content and technical releases. Record the URL, date, problem, change, and questions expected to improve. Compare several runs before and after the change; do not credit a single fluctuating answer as proof.
When a citation changes, inspect the cited passage and competing sources. A lost citation can come from a crawl regression, stale information, a stronger competing page, a rewritten query, or normal answer variation. The evidence should guide the next test.
Use a cadence that matches the decision
- Run technical checks after deployments and on a scheduled monitor for regressions.
- Run the stable citation panel weekly or monthly, using the same settings for comparable trends.
- Review high-value buying questions separately from broad awareness questions.
- Report new and lost citations alongside the underlying evidence, not only a net score.
- Review the query set quarterly so it continues to represent customer demand.
Use consistent metric definitions
| Metric | Calculation | What it answers | Caution |
|---|---|---|---|
| Mention rate | Answers naming the brand ÷ valid answers | How often is the brand part of the response? | A mention may be neutral, negative, or unsupported |
| Citation rate | Answers citing the domain ÷ valid answers | How often does the domain support an answer? | One citation can support only a narrow claim |
| Query coverage | Questions with a valid captured result ÷ monitored questions | Is the panel producing comparable evidence? | Provider errors should not be counted as misses |
| Competitor share | Answers naming or citing each competitor ÷ valid answers | Who repeatedly wins the same demand? | Define the competitor set and denominator |
| Citation retention | Previously cited questions still citing the domain ÷ previously cited questions | Are gains persisting? | Normal answer variation requires several observations |
| AI referral conversion | Desired outcomes from identified AI referrals ÷ AI referral sessions | Does observable traffic create value? | No-click visibility is not represented |
A useful change log entry
| Field | Example |
|---|---|
| Question | What tools can improve a website's AI visibility? |
| Target URL | /resources/ai-visibility-tools |
| Observed gap | Cited pages include criteria and examples; target page does not |
| Change | Added a tool-category table, pitfalls, sources, and selection scenarios |
| Published | Exact deployment timestamp |
| Expected result | Target URL becomes eligible and appears in later citations |
| Review window | Several comparable runs after crawl and index confirmation |
Interpret gains and losses carefully
- A new mention without a citation can indicate better entity recognition but does not show which source supported the claim.
- A citation gained for one question is evidence for that question and provider, not proof of a site-wide ranking increase.
- A lost citation should be checked across repeated runs before triggering a rewrite or rollback.
- A different customer URL can be a positive result if it is more relevant, current, or conversion-ready than the expected page.
- Search-console impressions, AI referrals, engagement, and conversions provide independent evidence beyond answer monitoring.
Sources and further reading
- AI Performance in Bing Webmaster Tools — Microsoft Bing. Describes citation counts, cited URLs, and citation trends available for Microsoft AI experiences.
- Publishers and Developers FAQ — OpenAI. Explains that ChatGPT referrals include the utm_source=chatgpt.com parameter.
Continue learning
See where your website stands
Run a free scan and get your AI visibility score across all six categories, with the gaps to fix first.