How to Measure GEO: Don't Rely on One AI Screenshot

Measure GEO on four levels: did AI read you (crawlers), did AI mention you (answers), did buyers come to your website from AI (visits), and did it bring enquiries (business). Looking at only one level easily leads to the wrong conclusion.

Why one screenshot proves nothing

A common habit is to ask ChatGPT once and take a screenshot: "AI recommends us" or "AI hasn't found us yet". These conclusions aren't reliable, because:

• The same question asked differently, in another language or on another day can get a different answer.
• Different AI platforms use different sources and priorities.
• Login state, chat history and location can all affect the result.

What you need is a repeatable test, not an occasional screenshot.

Four levels of measurement

Level 1 is the cause; the other three are results. When results are poor, go back to level 1 first: did AI never come, did it come and get blocked, or did it read the site but miss the key pages?

How to run buyer-question tests

1. Build a question set: from sales emails, enquiries and trade fair notes, collect 15–30 questions buyers really ask, covering selection, finding suppliers, comparison and company checks.
2. Fix the test conditions: record the platform, language, date and whether web search was on.
3. Record every answer: whether you're mentioned, whether it's accurate, which URLs are cited, which competitors are named.
4. Retest regularly: rerun the same questions monthly or quarterly and look at the trend.

Accuracy matters as much as being mentioned. Being named by an AI with the wrong specifications can be worse than not being named.

Blind spots in AI referral traffic

GA4's AI Assistant channel only counts clicks that carry referrer information. According to Google's documentation, Google's own AI Overviews and AI Mode are counted as organic search. Clicks from mobile apps, copied links, and visits where the buyer read the AI answer and then searched your name are also not credited to AI. ChatGPT automatically adds utm_source=chatgpt.com to its referral links, so that part can be identified more accurately.

For industrial websites, AI referral visits are usually modest; a few dozen a month is common. Judge them by quality: which pages they land on, how long they stay and whether they send enquiries.

What a monthly report looks like

• AI platforms that visited this month and how often, compared with last month
• Core pages read / not read
• Blocked requests and their causes
• Buyer-question tests: number of questions where you're mentioned, list of inaccurate descriptions
• AI referral visits and landing pages
• Content to fix and add next month

GEOmonitorai brings the first three levels together in one dashboard: crawler visits and blocks from the logs, AI referral visits from the tracking script, plus regular buyer-question tests. Level 4 is in your own hands. We suggest your sales team simply ask every new enquiry: "How did you find us?"

Further reading: How to Track Traffic from ChatGPT and Other AI Assistants · What Is an AI Visibility Tool? And What It Should Measure · Free AI crawler robots.txt checker

GEO Basics series, part 9 of 10. Next: A GEO Roadmap for Exporters: Start with an AI Audit