Measuring AI visibility means repeating the same questions, on the same platforms, against the same competitors, and recording what comes back. A screenshot from last Tuesday is not a baseline.
This is the method AO Agency uses in audits and monitoring. It is imperfect, because the systems are imperfect. It is still more useful than guessing.
Prompt selection
Start from how buyers actually ask: category, use case, industry, integrations, alternatives and “best for” questions. Include the messy wording people type, not only polished keywords. Keep the set large enough to see patterns and small enough to re-run.
Intent groups
Group prompts so results can be read: commercial, comparison, alternative, informational, and product-specific. A brand that appears on definitional queries but never on “which vendor should we shortlist?” has a different problem from the reverse.
Brand vs non-brand prompts
Non-brand prompts show whether you enter the consideration set. Brand prompts show whether the system knows who you are and describes you accurately. Both belong in a baseline. Mixing them into one percentage hides the story.
Competitors
Name the competitors you actually lose deals to, plus the brands that already occupy AI answers in your category. Track them on the same prompts. Competitor share is often the number leadership understands fastest.
Repeated testing
Run the set more than once before you treat a gap as structural. Re-run after major content or PR work, and on a schedule if you are monitoring. Record date, platform, prompt, mention, citation URLs, and a short note on accuracy.
Platform differences
ChatGPT, Google AI experiences, Gemini and Perplexity do not share one citation UI. Some answers have no links. Some retrieval is more web-fresh than others. Do not average incompatible interfaces into a fake “AI ranking”.
Citations
When sources appear, capture the URL and the domain. Over a prompt set you will usually see a short list of repeated publications, docs, directories or blogs. That list is the source landscape — and often the real brief for digital PR.
Share of answers
Mention rate is how often you appear. Share of answers compares that rate with competitors on the same set. Neither should be presented as a guaranteed ranking factor. Both are descriptive statistics of observed outputs.
Limitations and variability
Answers change. Session context, model version, location and sampling all introduce noise. Sentiment scoring is not perfectly objective. Claims about proprietary algorithms are not measurement.
Used honestly, this method still answers the question leadership is actually asking: when buyers use AI to research our category, do we show up, and is the description right?
Get a measured baseline
AO Agency builds a prompt universe, tests selected platforms and reports mentions, citations, competitors and source patterns.
