Measurement
Measuring AI traffic
You can see the clicks. You cannot see the mentions that did not produce a click. Plan around that gap rather than pretending it is not there.
What you can measure today
Assistants that link to sources send referral traffic with identifiable hostnames. In Google Analytics 4, build an exploration on Session source / medium and filter for these:
chatgpt.com chat.openai.com perplexity.ai claude.ai copilot.microsoft.com gemini.google.com you.com
Save it as a segment. In most accounts this traffic is small in volume and noticeably better in quality — longer sessions, higher intent — because the visitor arrived already half-convinced by a recommendation.
In Google Search Console, AI Overview impressions are folded into your normal Search performance data rather than split out, so you can see the effect on impressions and clicks but cannot fully isolate it. Watch for the pattern of impressions holding steady while clicks decline on informational queries: that is usually your answer being read without a visit.
Server logs: the underrated method
If you have access to raw logs or a CDN log stream, filter by user-agent for GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and the rest. This tells you something analytics cannot: which of your pages are being fetched, and how often. If your most important page is never crawled, that is a finding you can act on immediately.
What you cannot measure
Be honest with yourself and with anyone you report to:
- Mentions without clicks. If an assistant recommends you and the user acts on it directly — calls you, searches your brand name later — you will never attribute it.
- Share of voice. How often you are named versus a competitor is not visible from your own data. Third-party tools like Semrush's AI Toolkit, Ahrefs Brand Radar or Otterly.AI sample this by running prompts repeatedly. It is sampling, not census.
- Why you were dropped. No assistant explains why it stopped naming you.
Track direct and branded-search traffic alongside AI referrals. A steady rise in people searching your name specifically, with no campaign to explain it, is often the visible shadow of mentions you cannot see.
A measurement plan that survives contact with reality
- Baseline your assistant referral segment now, before you change anything.
- Log-check which pages the AI crawlers actually fetch, monthly.
- Pick five questions a customer would genuinely ask and check the answers manually once a month. Write down whether you appear. It is manual and it is sampling, but it is real.
- Watch branded search volume as the long-run indicator.
Do not build a dashboard that implies more precision than exists. The gap is structural, not a tooling problem you can shop your way out of.
Take this to your assistant
Paste it into ChatGPT, Copilot, Claude or Gemini and apply it to your own website.
Nothing is sent anywhere. The text is copied to your clipboard.