A growing share of website visits are coming from AI search products. ChatGPT, Claude, Perplexity, Google AI Overviews. The marketing question every small business owner is asking is the obvious one: how do you measure it?
The honest answer is: better than a year ago, still imperfect.
This is what is actually measurable today, what is still dark, and what to set up so you can see what is possible to see.
What is measurable today
Referrer-based attribution from AI products that link out. Perplexity, ChatGPT's web search feature, and Claude with web access all send users to source pages via clickable citations. Those click-throughs arrive with HTTP referrer headers that identify the source. Setting up the right analytics filters captures them.
In practice:
- Perplexity sends traffic with
perplexity.aiin the referrer. - ChatGPT search sends traffic with
chatgpt.comorchat.openai.comin the referrer. - Claude with web access typically sends with
claude.aireferrer. - Google AI Overview clicks have the same referrer as regular Google search traffic, but appear in Google Search Console with new query patterns and impression types you can filter on.
Brand search lift. Even when AI products do not send the click directly, they often drive brand search later. If someone learns about your business in a ChatGPT answer and then googles your brand name, that brand search shows up in Google Search Console and Analytics. Watching brand search volume over time gives you a directional signal on AI-driven awareness.
Direct traffic patterns. AI products that strip referrer headers send traffic that arrives as "direct" in analytics. Watching direct traffic patterns, especially to pages that are not normally direct destinations (deeper service pages, FAQ pages), gives you indirect evidence of AI-driven traffic.
What is still dark
ChatGPT non-web-search answers. When ChatGPT answers a question from its training data without searching the web, you cannot tell. The user gets the answer, no click happens, and the only evidence is the brand search you might see later.
Google AI Overview impressions that do not click. The user sees your business cited in the AI Overview, reads the answer, and never clicks. You showed up in their answer. There is no analytics event for it.
Voice assistant attribution. Siri, Google Assistant, Alexa. The user asks a question, gets an answer that may have used your content, and you have no visibility into it.
What to set up
Three things, in this order:
- Filter your analytics for AI referrers. Configure custom segments in Google Analytics 4 for the major AI referrers (perplexity.ai, chatgpt.com, claude.ai) so you can track them as their own channel.
- Watch brand search trends. Set up a monthly review of brand search query volume in Google Search Console. Sudden growth in branded queries without an obvious cause is often AI-driven awareness.
- Audit your content for citation friendliness. Pages that are structured, factual, and clearly attributed (with author bylines, sources, and dates) are more likely to be cited by AI products. Structured data helps too.
The measurement infrastructure for AI traffic is going to keep improving. The businesses that set up the tracking now will be the ones who can prove the ROI of AI visibility work over the next 18 months.