ChatGPT Search traffic: how we measure it
Referrer headers are sparse, but you can still see what's happening.
ChatGPT Search sends some referrer traffic, much of it dark. Measurement requires triangulation — no single signal tells the full story, but combined they paint a useful picture of how AI assistants are sending you visitors and what content they prefer.
The signals to combine
- Direct traffic spikes on long-tail pages with no obvious other source.
- Brand search lift after content cited by AI engines.
- The chatgpt.com and perplexity.ai referrers in raw server logs (often stripped by analytics tools).
- Bing Webmaster Tools impressions — surfaces some ChatGPT-driven queries.
- GSC's 'AI' search appearance filter for Google AI Overviews.
The instrumentation
Add a UTM-less branded query baseline in GSC. Spike-detect weekly. Cross-reference with Bing Webmaster Tools, which surfaces some ChatGPT impressions because OpenAI's index leans on Bing. Set up a Logflare or similar log pipeline to retain referrer headers stripped by GA4.
What the data tells you
Most AI traffic is informational, not transactional. Optimise for citation, not click — the citation builds brand authority even when the click never happens. Tracking citation rate is more useful than tracking referral traffic.
Reporting
Build a weekly 'AI presence' scorecard with citation count, branded search index and direct-traffic anomalies. Treat it as a leading indicator, not a revenue line — the revenue shows up in branded search 8–12 weeks later.