How to measure your brand's AI search visibility The four-part system.

Why rank tracking misses it

Your rank tracker tells you where a page sits in a list of links. It says nothing about whether an AI assistant named you, quoted you, or recommended a competitor while you held position three. That gap is now where a growing share of buying decisions happens, and it is invisible in every standard SEO report.

The early symptom usually shows up in Search Console first: impressions holding steady while clicks soften, with rankings unchanged. That pattern means the click became an answer. We unpacked the mechanics in the AI Overviews guide. The fix starts with measuring the thing properly, which is what the rest of this guide builds.

Step one: build a fixed prompt set

Everything hangs on this. Write down 20 to 30 prompts your buyers would genuinely type into an assistant, in the words they would actually use, across four buckets:

• Category prompts: "best X for Y" and "who should I use for X". These decide shortlists.
• Comparison prompts: "X vs Y" and "alternatives to X". These decide switchers.
• Cost prompts: "how much does X cost" and "is X worth it". These decide budgets.
• Problem prompts: "how do I fix X" and "why is my X not working". These build early trust.

Two rules matter more than the list itself. Fix it: never edit prompts mid-quarter, or your trend line becomes meaningless. And write prompts as buyers, not marketers: nobody asks an assistant for “omnichannel lifecycle solutions”, they ask who is good, what it costs and whether it works.

Step two: run the monthly prompt log

Once a month, run every prompt across ChatGPT, Gemini, Perplexity and Google’s AI Overviews, and log three things per prompt per engine: were you mentioned, were you linked as a source, and who was cited instead. A spreadsheet does this perfectly well. Screenshot the answers that matter, because assistants change their minds and you will want the receipts.

The third column is the one most brands skip and the one worth the most. Who gets cited instead of you, and which page of theirs earned it, is a content brief handed to you on a plate. Every competitor citation is a page you now know you need to beat, with the exact question it answered.

Step three: track assistant referral traffic

In GA4, build a channel group or exploration that isolates sessions from assistant domains: chatgpt.com, perplexity.ai, copilot.microsoft.com, gemini.google.com and claude.ai. Volumes will look small next to organic. Ignore the volume and watch the behaviour: assistant referrals arrive pre-sold, and they convert at multiples of standard organic in most accounts we see.

Treat this as the revenue proxy for the prompt log. Citations you cannot click are brand building; citations that send sessions are pipeline. Both count, and keeping them separate is what stops you over- or under-valuing the channel.

Step four: watch the AI crawlers

Your server logs or Cloudflare analytics will show you GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot and Google-Extended fetching pages. Two checks matter. First, that they are allowed in at all: a robots.txt written in 2023 may be blocking the bots you now want. Second, that they are hitting your money pages, not just the blog. If the crawlers never reach a page, no assistant will ever quote it. The full crawl-access checklist is in how to get cited by ChatGPT, Gemini and Perplexity.

The one number that matters: share of answer

Roll the prompt log into a single metric. Share of answer is the percentage of your fixed prompt set where your brand is mentioned or linked. Twenty-five prompts, eight appearances, 32% share of answer. Its companion is citation rate: the percentage where you are linked as a source rather than just named, which is the number that turns into traffic.

Trend share of answer monthly and judge it quarterly. Individual months bounce around because answers are probabilistic; the quarter tells you whether the entity, content and citation work is compounding. If you are new to the terminology, the plain-English GEO guide and the GEO vs SEO comparison set the scene.

What to do with a bad month

Work the diagnosis in order. Lost a citation you held? Check whether the page that earned it changed, was redirected, or got slower to crawl. Never held it? Read the page the assistant cited instead and beat it properly: a more direct answer, better evidence, cleaner structure. Dropped everywhere at once? Check crawl access and indexing before touching a word of copy.

What you should not do is rewrite everything or change the prompt set. AI visibility moves in quarters, and the brands that win are the ones that keep the measurement boring and the fixes specific.

If you would rather not run any of this yourself, it is the reporting layer of our SEO & AI Visibility subscription, from £1,250 a month plus VAT: prompt set agreed in month one, checked across four engines, and the share-of-answer trend in your monthly report with the citation sources attached. The full scope is on the AI SEO page.

Questions we get asked

Written by

Sam, co-founder, growth and paid media

Runs growth and paid media at Mesmerise Marketing. Ten years of paid and organic campaigns for challenger brands across DTC, B2B services and hospitality.

Meet the team