How to track traffic from ChatGPT and other AIs

Traffic from ChatGPT and other AI assistants can be tracked like any other referrer, as long as your analytics tool knows the hostnames the assistants send people from and groups them as a source of their own. ChatGPT links arrive with chatgpt.com as the referrer, Perplexity with perplexity.ai, Claude with claude.ai, and Gemini with gemini.google.com. Tools that do not recognise those hosts show the visit as "direct" or as a random hostname, which is why many founders believe they get no AI traffic at all.

Why AI traffic hides in your reports

When someone clicks a link inside ChatGPT, their browser sends the address of the page they came from in the Referer header, the same way a click from Google does. The difference is what the analytics tool does with that hostname. Google Analytics and most privacy-friendly tools keep a list of known search engines and social networks; a host that is not on the list is shown as a plain referral, buried among hundreds of others, or, when the assistant strips the referrer, the visit becomes "direct".

Two things make it worse. Assistants open some links in an in-app browser that sends no referrer at all, and people often copy a link from an answer and paste it into a new tab, which also arrives as direct. So the AI number you can see is a floor, never the whole story. That is a reason to track it well, not a reason to ignore it.

The hostnames to recognise

These are the referrer hosts each assistant sends people from. A tracker matches a host when it equals the entry or ends with a dot and the entry, so subdomains are covered.

Glai ships this list as its traffic source catalog, with Meta AI, You.com, Poe and Phind as well, and files every one of them under the source type ai. The list is data, not code, so a new assistant is one row added, and your old visits are not renamed.

How to see AI visits by name

1. Install a tracker that groups AI sources. With Glai that is one script tag in the head of your pages; the collector reads the referrer on the first pageview of each visit and stores the source name and type with the visit.
2. Open the sources breakdown and filter by the type "ai". Each assistant appears as its own row with visitors, pageviews, bounce rate and, if you have connected payments, revenue and customers.
3. Check the landing pages. Filter by source and look at the entry pages. Assistants usually link to a specific answer page, a comparison or your pricing page, not your home page, which tells you what question the person asked.
4. Watch it for a month before judging. AI traffic is spiky: a mention in a popular answer brings a burst, then nothing, then another burst when the model's index refreshes.

What the numbers usually look like

Across the dashboards we have looked at, visitors from assistants are a small share of the total, often a few percent, but they arrive further along. They have already asked a question, read a summary and chosen a link, so they bounce less and convert more than the average search visit. That is the reason to measure the AI row against revenue rather than against pageviews: the attribution post explains how a payment is tied back to the visit that brought it.

None of this needs a consent banner: the referrer is read on the server and the visitor is counted the way the privacy post describes. Do not expect clean UTM parameters. Assistants link to the page as it appears in their sources, so the only reliable signal is the referrer. If you want to know which answer a person read, the entry page and the time of day are usually enough to work it out.

Getting more of it

• Answer questions plainly on your pages. Assistants cite pages that state a fact in a sentence a model can quote: what the product does, what it costs, what it does not do.
• Keep your pricing public. "How much does X cost" is one of the most common questions people ask an assistant, and a page with the number wins the citation.
• Write comparison pages that are fair about the alternatives. Models prefer a source that mentions the competition over one that pretends it does not exist.
• Let the bots in. Check your robots file does not block the crawlers the assistants use to read pages; a blocked page cannot be cited.

What Glai does with AI traffic

Every source in the catalog has a type, and ai is one of them, so the dashboard shows an AI row beside search, social, email, paid and referral, with the share of visitors and the conversion rate of each. A click on the row filters every other breakdown: the pages AI visitors read, the countries they are in, the goals they complete and the payments they made. The stats API returns the same breakdown as JSON, with source_type=ai as a filter, if you want it in a spreadsheet or a Slack message.

Ready to see your own AI row? Create a free account, paste the script, and the first visit shows within seconds.

Questions people ask

Does ChatGPT send a referrer when someone clicks a link?

Usually yes. Clicks from the ChatGPT website arrive with chatgpt.com as the referrer, and the collector names them ChatGPT. Links opened from the mobile app's in-app browser sometimes arrive without one and count as direct, so the number you see is a minimum.

Can I use UTM parameters to track AI traffic?

Only on links you control, such as those in your own documentation that an assistant may quote. Assistants link to pages as they found them, so the referrer is the reliable signal and UTM tags are a bonus when they happen to be present.

Is AI traffic worth optimising for?

Measure it against revenue first. In most dashboards the AI row is small in visitors but strong in conversion, because the person already asked a question and chose your link. If the revenue per visitor is above your average, it is worth the effort.

How quickly does a new assistant appear in the reports?

As soon as its hostname is in the source catalog. The catalog is data, so a new assistant is one row, and visits after that moment carry its name. Visits recorded before it was added keep the hostname they arrived with.