The short version
AI answers change from run to run and from country to country, so one check proves nothing. Build a fixed list of 20 to 50 realistic prompts per market, run each several times on a schedule, and record whether you're mentioned, whether you're cited with a link, and who else appears. Report ranges with dates. Then compare with real data: Search Console, GA4 AI-platform referrals and leads.
Why one screenshot isn't a measurement
AI systems generate answers fresh each time. Sources can change with the prompt wording, the user's location, the model version and plain randomness. So we never report AI visibility as one number. A fair statement looks like "mentioned in 6 to 8 of 10 runs between 12 and 19 October," not "we rank #2."
Step 1: Write prompts that sound like real buyers
Skip keyword lists. Write the questions customers type or speak:
- Category: "What's the best accounting software for a small restaurant?"
- Comparison: "Shopify vs WooCommerce for a store with 500 products"
- Problem: "How do I fix duplicate Google Business Profile listings without hiring an agency?"
- Brand: Ask whether your own brand is any good and what customers say, using your real business name and city.
- Buyer research: long, specific prompts, like "we're a US-based software company and want to be recommended in ChatGPT; which agencies specialise in this?" People really do ask this way.
Step 2: Build one prompt set per market
Don't translate one list. People in the US, UK, Canada and Australia use different words (the UK and Australia say "optimisation"; the US and Canada usually say "optimization"), different competitors, and different local sources. For each market:
- Write prompts in local phrasing and, in Canada, in each language you serve.
- Set the location in the tool where it allows, and record it.
- Add local-intent prompts ("best accountant in Manchester / Toronto / Sydney / Austin").
Report each market first, then a combined view. A blended number hides which market is working.
Step 3: Choose engines and run on a schedule
Start with the engines your customers use: ChatGPT, Google AI Overviews and AI Mode, Gemini and Perplexity. Add Copilot if shopping or enterprise audiences matter. Some data providers offer API access to LLM answers and mention data. Anything you check by hand is labeled "manual" with the date.
Run each prompt at least three times per session, monthly (weekly during a campaign). Keep a stable core set so you can compare over time, and add new prompts without deleting old ones.
Step 4: Record five metrics
| Metric | Meaning | Log |
|---|---|---|
| Mention rate | % of runs where your brand name appears | Y/N per run |
| Citation rate | % of runs where your URL is linked as a source | Y/N + URL |
| Share of voice | Your mentions ÷ all brand mentions in the set | Count competitors |
| Sentiment | Positive, neutral or negative description | Short note |
| Accuracy | Are facts right (services, location, pricing)? | Flag errors |
Mention and citation are different. An answer can name you without linking, or link you without naming.
Step 5: Add the data that is actually yours
Prompt tracking is a sample. Your analytics are real, and the two should never be blended into one number.
- GA4: Look at session source/medium for AI platforms such as chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai. Referrer names change, so confirm them. Some AI visits arrive with no referrer and show as direct, so GA4 undercounts.
- Search Console: Google says AI Overviews and AI Mode activity is counted under the Web search type with regular results, so there's no separate filter. Watch impressions and clicks for question-style queries and use the Countries filter for each market.
- Leads: Ask "how did you hear about us?" on forms and calls, and include "ChatGPT / AI assistant" as an option.
Step 6: Connect visibility to business results
Visibility only matters if it leads somewhere. Track, per market:
- Leads or demo requests from organic and AI referrals (GA4 key events)
- Branded search impressions and clicks over time (Search Console)
- Pipeline or revenue from those leads, from your CRM
- For ecommerce: organic revenue, product impressions and clicks, and Merchant Center issues
Our GEO ROI calculator can help you estimate a business case. Treat its output as a model, not a result. For how visibility compounds into pipeline, see this case study breakdown.
Ecommerce teams: what to add
Ecommerce questions we see include "which tool tracks my product visibility in Copilot shopping?" and "how do I measure AI visibility ROI for products?" A practical approach:
- Track product-type prompts ("best waterproof jacket under $150") separately from brand prompts.
- Record which retailers and review sites the engines cite, since your product may appear through them.
- Watch Merchant Center diagnostics for item issues.
- Compare AI-referral revenue in GA4 with your total, expecting small and incomplete numbers at first.
No tool sees every AI conversation. Ask any vendor what their data covers and what it misses. See also ecommerce AI search.
Step 7: Turn numbers into actions
- Not mentioned at all: check crawler access and entity basics first. See technical SEO for AI and entities and the knowledge graph.
- Mentioned but not cited: add short direct answers to key pages.
- Competitors cited from sites you're not on: plan to earn a real presence there. See AI citation building.
- Described wrongly: correct your own pages and profiles, then third-party listings.
What we'd skip
- A single AI visibility "score" with no method behind it
- Daily checks (the noise swamps the signal)
- Promising rankings or citations to clients
- Treating llms.txt as a measurement or ranking lever; there's no proven effect
Frequently asked questions
What is AI search visibility?
It's how often, and how favorably, your brand appears in answers from AI tools like ChatGPT, Google AI Overviews, Gemini and Perplexity, including whether they link to you.
How many prompts should I track?
Start with 20 to 50 per market, covering awareness, comparison and purchase stages.
Why do results change every time?
Answers are generated each time and vary with wording, location, model version and chance, so run each prompt several times and report ranges with dates.
Can I see AI traffic in GA4?
Partly. Filter by source/medium for AI platforms, but expect undercounting because some visits have no referrer.
Does Search Console show AI Overviews separately?
Google's documentation says AI Overviews and AI Mode are counted in the Web search type, so there's no separate report.
How do I measure AI visibility across countries?
Use a separate prompt set for each market with local phrasing and location settings, then use Search Console's country filter and GA4 geography. Report each market first, then combine.
How do I measure ROI from GEO or AEO work?
Track leads and revenue from organic and AI referrals, add a "how did you hear about us" field, and compare against a baseline. Expect an imperfect picture and say so in reports.
Can an agency guarantee AI citations?
No. They can improve the conditions you control and measure the results.