How do I know if AI chatbots are sending my store any traffic?

Check referrers in your analytics for assistant domains such as chatgpt.com and perplexity.ai, and for UTM parameters (ChatGPT adds a utm_source parameter to many of its links). Some visits arrive with neither and show as direct traffic. Then check server logs for assistants that fetch pages on a user's behalf. Most of those fetches do not run JavaScript, so they are missing from analytics reports. Agents that drive a real browser do run it and appear as ordinary visits.

There are two different things to measure, and they live in different places. The first is people: someone reads an assistant's answer, clicks a link and lands on your site. The second is machines: the assistant requests your page in order to write that answer. The first appears in analytics. The second mostly appears only in server logs.

For visits from people, start with the referrer. Build a segment or channel in your analytics tool that matches the domains of the main assistants, and keep the list current as new ones appear. Add a rule for the UTM source that ChatGPT attaches to many of its links. Together these capture the visits that identify themselves.

Expect an undercount. Links opened from an app, copied into a browser or passed through a privacy setting can arrive with no referrer and no parameter. They are recorded as direct traffic. A rise in direct visits does not prove assistant referrals; missing attribution has several causes.

For machine requests, look at the logs from your server or content delivery network. Search for the user agents assistants use when they retrieve a page for a user, such as ChatGPT-User, and count requests by page and by day. Because a user agent can be forged, check the source addresses against the ranges the attacker publishes before relying on the numbers. The pages fetched most often are the ones assistants consult when answering questions in your category.

Compare the two sets. If assistants fetch a product page frequently and few referred visits follow, the assistant may be answering the question without sending the reader on.

Then look at quality. Compare conversion rate and order value for assistant referrals with your other channels. Use a sufficient sample before drawing conclusions about order quality.

A simple supplement is to ask. A question at checkout or after purchase about how the customer found you captures referrals that left no technical trace.

How Kairal helps

Agent Traffic Intelligence, currently in early access, checks agent identity and purpose to support your access policies. A recognised agent is not proof of a customer’s authorisation; keep account, purchase and payment checks on the operations it can request.

Agent Traffic Intelligence