You measure AI search visibility with a fixed set of prompts your buyers would actually ask, run on a schedule across ChatGPT, Perplexity, Google AI Overviews and Gemini, with every answer recorded: whether you are mentioned, whether you are cited with a link, how you are described, which page was quoted and which competitors appear. Keep the prompts and method constant and the numbers become comparable month to month. Change them and you are measuring noise.
Why rankings are not enough any more
A page can rank first and still lose the click. Pew Research Center found that when Google showed an AI summary, users clicked an ordinary result on 8% of visits, against 15% without one. Ahrefs estimated in February 2026 that the top result’s click-through rate is 58% lower when an AI Overview appears. Rankings tell you where you sit in a list that fewer people read. Citation tracking tells you whether you are in the answer they do read.
Step 1: build the prompt set
Write 30 to 60 prompts in five groups. Use your buyers’ words, not your marketing words:
| Group | Example for a Shopify agency | What it shows |
|---|---|---|
| Category | “best Shopify agency for D2C brands in India” | Whether you are on the shortlist |
| Comparison | “Shopify vs WooCommerce for a saree brand” | Whether your content shapes the answer |
| Problem | “how do I set up GST on Shopify” | Whether your guides are cited |
| Local | “Shopify developer in Dubai with Arabic stores” | Market-specific visibility |
| Brand | “is Maison Global Agency any good” | How the engines describe you |
Write the prompts as people speak to AI: full questions, often with context (“I sell handmade jewellery on Instagram, which platform should I use for a store?”). Run each market in its own language where buyers search that way.
Step 2: run it the same way every time
- Same engines and modes each run: ChatGPT with search on, Perplexity, Google with AI Overviews, Gemini.
- Logged out or in a clean profile, so personal history does not skew results.
- Same location settings for each market.
- Each prompt run three times, because generated answers vary; record how many of the three runs mention you.
- Same day of the month, so seasonal news does not distort comparisons.
Step 3: record the right fields
For every prompt and run, log the date, engine, prompt, whether your brand is mentioned, whether your site is linked, the URL cited, the position of the mention in the answer, the description used, its accuracy, and the competitors named. A spreadsheet works; the discipline matters more than the tool.
Step 4: turn it into a few numbers
- Mention rate: share of prompts where your brand appears at all.
- Citation rate: share of prompts where your site is linked as a source.
- Share of voice: your mentions divided by all brand mentions across the prompt set.
- Accuracy: share of descriptions that are correct and current.
- Top cited pages: which of your URLs the engines rely on, which tells you what to protect and improve.
Step 5: score why, not just what
Citation numbers tell you where you stand; a site score tells you why. We score four dimensions out of 25 each:
- Evidence: specific, quotable facts and first-party numbers in the opening lines of key pages.
- Structure: question-led headings, direct answers, schema that matches the text.
- Authority: reviews, listings, press and mentions on other sites, surfaced on yours.
- Crawlability: AI search crawlers allowed, pages indexed and eligible for snippets, fast server-rendered HTML. See robots.txt for AI crawlers.
The multi-city law firm we audited scored 25 out of 100 at baseline: slogans where facts should have been, no schema, and nothing a model could quote. The 30/60/90-day plan that followed went after evidence and structure first, because those are fastest to change.
Measure the same prompts, the same way, every month. Anything else is a screenshot, not a metric.
Step 6: connect it to traffic
Citations matter because of what they send. Three places to look:
- Search Console. Google says traffic from AI features is included in the overall Web search reporting, so watch clicks and impressions for the question queries in your prompt set.
- Analytics referrals. Visits from chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com show up as referrals; ChatGPT usually adds a
utm_source=chatgpt.comtag to its links. Build a channel group for AI assistants in GA4. - Enquiry forms. Add “How did you hear about us?” with “ChatGPT or another AI assistant” as an option. Many AI-influenced buyers never click a link; they search your name later.
A monthly rhythm that works
- Run the prompt set on the same day each month.
- Update the five numbers and the four-dimension score.
- List citations gained and lost, and the pages behind them.
- Pick the next month’s fixes: the pages engines quote for competitors but not for you.
- Report it alongside Search Console and GA4, so AI visibility sits next to the traffic it produces.
Want your baseline without building the spreadsheet?
Our free check runs ten buyer prompts across ChatGPT, Perplexity and Google AI Overviews and sends you the results and three fixes within two working days.
Get the free checkFrequently asked questions
Is there a tool that measures AI search visibility automatically?
Several tools track brand mentions across AI engines, and they save time. They still depend on the prompt set you give them, so the method above applies either way; check that a tool lets you fix prompts, markets and run frequency.
How many prompts are enough?
Thirty to sixty covers most brands across the five groups. Fewer than twenty makes the numbers swing too much from month to month.
How often should AI visibility be measured?
Monthly. Weekly runs mostly measure the engines’ own variation; quarterly runs are too slow to connect changes to results.
Can AI visibility be tracked in Google Search Console?
Partly. Google includes traffic from AI Overviews and AI Mode in Search Console’s Web reporting, but does not separate it out, so pair Search Console with a prompt set and AI referral tracking in analytics.
Related: What is GEO?, GEO services, free AI-search visibility check.