How do you measure GEO and AI visibility?
Measure AI visibility by running a fixed set of buyer questions across ChatGPT, Perplexity, Gemini, Copilot and Google's AI features on a regular schedule, and recording how often you are named or cited, where in the answer, and how accurately. Back that up with two hard data sources: server logs, to confirm AI crawlers are fetching your pages, and analytics, to count visits referred by AI platforms. Track the trend month to month, because individual answers vary between runs.
What should you measure?
| Metric | What it tells you | How to get it |
|---|---|---|
| Mention rate | How often your brand is named across the prompt set | Prompt runs |
| Citation rate | How often your pages are linked as a source | Prompt runs |
| Share of voice | Your mentions versus named competitors on the same prompts | Prompt runs |
| Position | Whether you are named first, in the middle or at the end | Prompt runs |
| Accuracy | Whether the description of you is correct and current | Prompt runs, reviewed by a person |
| Crawler activity | Whether AI search and training crawlers fetch your pages | Server or CDN logs |
| AI referral visits and enquiries | Whether visibility turns into visits and leads | Analytics |
Mention rate and citation rate are different. An engine can recommend you by name without linking to your site, or cite your guide as a source without recommending you. Track both.
How do you build a prompt set?
- Write 20 to 50 prompts in the words buyers use. Mix:
- category prompts ("best corporate travel agencies in Dubai")
- problem prompts ("how do I cut business travel costs for a 50-person team")
- comparison prompts ("X vs Y for corporate travel")
- brand prompts ("what does [your brand] do", "is [your brand] any good").
- Name three to five competitors you want to be measured against.
- Freeze the set. Changing prompts every month destroys the trend line. Add new prompts as a separate group.
- Fix the conditions. Same engines, same mode (for example ChatGPT with search on), logged-out or clean sessions where possible, same location settings.
- Run each prompt more than once if you can. Answers vary, so two or three runs per prompt give a steadier rate.
- Log everything: date, engine, prompt, brands named in order, cited URLs, and a short accuracy note.
How do you use server logs and analytics?
Logs show which crawlers fetch which pages. Look for user agents such as OAI-SearchBot and ChatGPT-User (OpenAI's search crawler and user-triggered fetches), GPTBot (OpenAI's training crawler), PerplexityBot, ClaudeBot and Bingbot, which feeds Copilot. Google's AI features use the normal Googlebot crawl. If a crawler never appears, check robots.txt and your CDN's bot settings before anything else.
Analytics shows the visits. In GA4, create a custom channel group or exploration that matches referrers such as chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com. ChatGPT often adds utm_source=chatgpt.com to links, which helps. Then look at landing pages and conversions for that channel, not just sessions.
One gap to know about: clicks from Google AI Overviews and AI Mode are counted inside Google's normal search data and are not reported separately in Search Console.
What are the common mistakes?
- One-off snapshots. A single check tells you very little because answers vary. The trend is the signal.
- Logged-in testing. Your own chat history and memory can bias answers towards your brand.
- Counting mentions without checking accuracy. Being named with last year's pricing or the wrong services is a problem to fix, not a win.
- Only measuring brand prompts. Anyone can get named when asked about themselves by name. Category and problem prompts are where buyers find you.
- Ignoring conversions. AI referral traffic is often small but well qualified. Judge it by enquiries, the same way you would when measuring SEO ROI.
What does this look like for a real business?
Here is a hypothetical. Say you run a B2B catering company in Dubai and build a 30-prompt set covering office catering, event catering and dietary-requirement questions, measured against four competitors.
Month one baseline: you are named on a small number of category prompts, all on Perplexity, and never on ChatGPT. Your logs show PerplexityBot visiting regularly but no OAI-SearchBot at all. The cause turns out to be a CDN bot-protection rule. Fixing it is a settings change. Over the following months you track whether ChatGPT mentions start appearing, whether your share of voice on "office catering" prompts rises as new menu pages go live, and whether the AI referral channel in GA4 produces enquiries.
How WebBox approaches this
Our AI visibility check builds the prompt set with you from your sales conversations, fixes the competitor list and conditions, and runs the baseline across ChatGPT, Perplexity, Gemini, Copilot and Google. You get mention rate, citation rate, share of voice and an accuracy review, with the cited sources that beat you listed by engine.
Ongoing measurement is automated with the same tooling we use in AI Automation: scheduled prompt runs where the engines allow it, results written to a shared tracker, log checks for AI crawler activity and a GA4 AI referral channel, all rolled into one monthly report. A person still reviews accuracy, because a model describing you wrongly is something a dashboard will not catch.
Related questions
Something else? Book a call and ask us directly.