Measuring AI Search Visibility: Prompts, Citations and Traffic

By the FlyRank team, Manchester · Published 15 October 2026

TL;DR: You cannot rank-track AI answers the way you track keywords, because responses vary from run to run. Measure instead with a fixed set of prompts, repeated runs, and a log of whether your brand is mentioned, whether your URL is cited, and whether the facts are right. Add analytics for AI referrals and a “how did you find us” question. Report trends over weeks, not single results.

What should you measure?

Measure mentions, citations, accuracy, position and competitors across a fixed set of prompts. These replace keyword rankings as your core visibility signals.

A mention means the assistant names your brand. A citation means it links to or lists your page as a source. They are different, and a brand can be mentioned often while its pages are never cited.

A useful headline figure is citation presence across your prompt set: the share of runs in which at least one of your URLs is cited. Track it over time and by topic.

How do you build a prompt set?

Write around 15 realistic questions per topic that your customers would ask, covering awareness, comparison and buying stages, plus local variants if you serve an area.

Source the wording from real places: customer emails, sales calls, People Also Ask boxes, forum threads and your own site search. People type questions to an assistant differently from keywords, so keep them natural and specific.

Include prompts at each stage. “What is answer engine optimisation?” tests awareness, “AEO vs SEO for a small business” tests consideration, and “best SEO agency in Manchester” tests decision stage.

Freeze the list once agreed. Changing wording each month makes trends meaningless, so only add new prompts, never edit old ones.

How do you track and log results?

Run each prompt in a fresh session several times, on more than one assistant, and record the outcome in a spreadsheet with the date.

AI answers vary. The same prompt can produce different brands and sources on consecutive runs, so a single result is an anecdote. Repeat each prompt at least three times and note how often you appear.

1. Set up a clean test

   Use a logged-out or fresh session, with memory features off where possible, and web search on if you want to test cited answers.

2. Run and record

   For each prompt, log the assistant, date, whether you were mentioned, whether a URL was cited, the position, and competitors named.

3. Capture sources

   List which domains the assistant cites. Over time this shows which sites shape answers in your market.

4. Flag wrong facts

   Note incorrect statements about your business and assign someone to fix the source of the error.

Run the baseline before you change anything, then repeat on a schedule such as monthly. Allow weeks after a content change for crawlers to revisit pages, and do not draw conclusions from the first few days.

How do you measure AI referral traffic?

Create a custom channel or filter in GA4 that groups sessions by AI referrer sources, and check which referrers genuinely appear in your data before trusting the numbers.

Some assistants pass a referrer, and some mark links with a parameter. OpenAI’s bot documentation covers its crawlers, and the way ChatGPT tags outbound links is worth confirming in your own analytics rather than assuming.

Start with the Google documentation on default channel groups and build a custom group matching sources such as chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com. Test each pattern against real sessions, because broad patterns can match unrelated sources.

Traffic from AI Overviews and AI Mode is not separate. Google states in its AI features documentation that it counts within overall Search traffic in the Performance report, so you cannot isolate it there.

Many AI-influenced visits arrive without a referrer, as direct or branded search. Add “ChatGPT or another AI assistant” as an option on your enquiry form’s “how did you find us” field to catch some of that hidden demand.

Key takeaway: expect AI referral numbers to undercount. Combine analytics with form answers and prompt tracking.

Keep a note of each assistant and mode you test, such as search on or off, because behaviour differs between them and comparisons only work like for like.

Do you need a paid tool?

No. A spreadsheet and a disciplined routine are enough to start, and paid tools mainly save time at larger prompt volumes.

Dedicated AI visibility platforms run prompts automatically and chart mentions over time. They are useful when you track hundreds of prompts, many brands or several countries, but their methods and coverage differ, so test before committing.

Whatever you use, check how it runs prompts: which models, whether it uses web search, how often, and whether it repeats runs. A tool that reports a single run as fact is giving you a false sense of precision.

For a small UK business, 30 to 60 well-chosen prompts tracked by hand each month will reveal more than a dashboard of poorly chosen ones.

How should you report it?

Report trends, not snapshots: citation presence by topic, accuracy issues fixed, AI referral sessions and leads, always alongside normal organic results.

Show the baseline, the current month and the change. Pair numbers with the actions taken, such as a corrected business listing or a rewritten service page, so the link between work and outcome is clear.

Be candid about limits. You cannot prove that one change caused one citation, and results move because models and search features update. Say so in the report.

If you want this set up for you, our answer engine optimisation services include a baseline visibility check and a tracked prompt set. Read how AI Overviews choose sources and how ChatGPT finds and cites websites to understand what you are measuring.