Measuring GEO results: why one test tells you nothing
By Jasper Paauwe. Updated on .
Why one test says little
Ahrefs asked five AI tools the same questions for 20 days and followed about 37,500 citations. For ChatGPT and Google AI Mode, about 60 percent had disappeared the next day. For AI Overviews it was about 50 percent, for Gemini and Copilot about 80 percent. About 75 percent of what disappeared came back within the 20 days.
If you do not see your brand today, it may be back tomorrow. If you do see it, that is no proof that your work has helped either.
How to set up a measurement
Setting this up takes an hour, and after that half an hour per week.
- Write down twenty questions that customers really ask, without your brand name in them.
- Pick fixed tools, for example ChatGPT, Perplexity, Copilot and Google AI Mode.
- Ask the same questions every week and note for each answer: named, cited or not.
- After eight weeks, look at the share of answers that include your brand, not at individual results.
What else you can log
Search Console does not show visits from AI crawlers. Your server log does: look for user agents such as OAI-SearchBot, Claude-SearchBot and PerplexityBot. If they never come by, that is a technical problem to solve first. Your GEO does not measure those series for you at the moment. The check assesses your site, not the answers themselves.
Sources
- How Stable Are AI Citations? (Ahrefs, 2026)
- Overview of OpenAI Crawlers (OpenAI, 2026)
- Does Anthropic crawl data from the web? (Anthropic, 2026)
- Perplexity Crawlers (Perplexity, 2026)