How to Track ChatGPT and Perplexity Traffic, and Find Out if It Converts

People now ask an AI assistant questions they used to type into a search engine. Some of the answers include a link to your site, and some of those people click it.

That traffic is in your analytics today. In most reports it sits inside "referral" or "direct", where nobody looks at it. This guide shows you how to find it, what you cannot find, and how to answer the question that matters: do these visitors convert?

How much AI traffic is there#

Less than the headlines suggest, and growing fast.

A 2026 study by SE Ranking puts AI referrals at about 0.32% of all website traffic, up from 0.02% in 2024. ChatGPT sends about three quarters of it. Gemini, Perplexity, Copilot, and Claude share the rest. The same study reports that AI visitors spend about 68% more time on site than visitors from organic search.

So the share is small and the visitors look engaged. That is a good reason to measure it on your own site and not rely on averages.

How each AI assistant identifies itself#

A visit carries two pieces of source information: the referrer, which the browser sends, and UTM parameters, which are part of the link. AI assistants use one or both.

To build one segment for all of them, filter sessions where the referrer domain or the UTM source contains one of these names:

chatgpt|openai|perplexity|gemini|copilot|claude

Save that segment. You will use it for every step below.

What your analytics will not show you#

Three kinds of AI traffic are hidden, and no tool can fully recover them. It is better to know this before you report a number to your team.

Clicks that arrive with no source#

Some AI apps open links without a referrer and without a UTM parameter. The visit arrives with no source information, so it is counted as direct. One sign of this is a rise in direct visits that land on deep pages, such as a long blog post or a documentation page. Few people type those addresses by hand.

Google AI Overviews and AI Mode#

A click from an AI answer inside Google search carries a Google referrer. It looks the same as a normal organic click. You cannot separate the two in your own analytics.

Visitors who never click#

Many people read the AI answer and stop there. If the answer named your brand, some of them search for you later or type your address. That visit is counted as branded search or direct. The AI assistant gets no credit in any report.

AI crawlers are a separate topic#

Bots that read your pages to train or ground a model are not visitors. They do not run your analytics script in most cases, and they never buy. Look for them in server logs, not in your analytics.

Find out if AI visitors convert#

Now use the segment. Compare AI visitors with organic search visitors over the same period, on four measures.

1. Landing pages. Which pages do AI visitors arrive on? This shows you which of your pages the assistants cite. It is often a comparison page, a guide, or documentation, and less often the homepage.
2. Conversion rate. Signup, lead, or purchase rate for the AI segment against organic search.
3. Depth. Pages per session and time on site. A visitor who already read a summary of your product often goes directly to pricing.
4. Where they stop. The last page before they leave. If AI visitors land on a guide and leave from it, the guide has no clear next step.

Be careful with small numbers. If you have 80 AI visits and 3 conversions, the rate can double or halve next month for no reason. Report the count next to the rate, and wait for a few hundred visits before you make a decision from it.

Why cookie consent makes this harder#

In most analytics tools, a visitor who declines the cookie banner is not counted. For a source this small, that is a serious problem. If you get 200 AI visits a month and 40% decline the banner, you are left with 120, which is too few to compare anything.

Cookieless measurement does not have this limit, because there is no cookie to decline. Every visit is counted, with its referrer and its UTM parameters. Read cookieless tracking explained for how that works.

What to do with the answer#

The result of your comparison tells you where to put effort.

AI visitors convert well. Find the pages they land on and look at what those pages have in common. They are usually specific, factual, and easy to quote. Write more pages like them. Check that each one has a clear next step, because these visitors arrive deep in the site and never see your homepage.

AI visitors read and leave. The assistant sent them for information, and the page gave it to them. Add the next step to that page: a short product explanation, a link to pricing, a way to try the product.

You have almost no AI visitors. The assistants do not cite you yet. They build answers from pages they can find and quote: comparison articles, review sites, community threads, and clear documentation. That is a content and visibility task, and your analytics can only tell you when it starts to work.

How this looks in GrainQL#

GrainQL stores the referrer domain and the UTM source for each visit, so the AI segment is one filter. You can also skip the filter and ask Kai:

• "How many visitors came from ChatGPT and Perplexity last month?"
• "Compare the signup rate of ChatGPT visitors with organic search visitors."
• "Which pages do visitors from AI assistants land on, and where do they leave?"

Kai runs the comparison and shows the sessions behind the numbers, so you can watch what these visitors do on the page.