Skip to main content
Growth6 min read

Traffic from ChatGPT and Perplexity: measure AI as a channel

People ask an assistant, the assistant cites a page, and some of them click. Here is how to see those visits as their own channel, and what they are worth in money.

Vicente TorresFounder of Vero

Some research now starts in a chat window instead of a search box. Someone asks ChatGPT or Perplexity which tool to use, the answer cites a few pages, and a few of those readers click through. If your analytics files that visit under Referral next to a forum thread, you cannot tell whether assistants are sending you anyone, let alone customers. This post shows how to measure AI assistants as a channel of their own, where the measurement breaks, and how to read the channel in money.

Why do AI assistants deserve their own channel?

A channel is a group of sources that behave alike and that you would act on together. Search visitors arrive with a query in mind. Social visitors arrive from a feed, mid-scroll. A visitor from an assistant arrives after a conversation in which your page was offered as an answer, often next to one or two alternatives. That is a different intent from both, and it deserves a line of its own.

Lumped into Referral, the same visits disappear among blogs, newsletters and directories. You cannot see a trend, you cannot compare it with search, and you cannot tell whether the pages assistants cite are the ones that sell. Splitting it out costs nothing and answers a question many founders now ask: is anyone finding us through AI?

How does a visit from ChatGPT or Perplexity arrive?

When someone clicks a cited link in a browser, the request usually carries a referrer: the host of the page they came from. Vero reads that host and files it under AI when it is one of chatgpt.com, chat.openai.com, perplexity.ai, claude.ai, gemini.google.com or copilot.microsoft.com. The AI list is checked before the search list, so gemini.google.com counts as AI and not as Google search. A subdomain counts as its parent host. Some assistants also tag their links: ChatGPT adds utm_source=chatgpt.com. A utm_source named after an assistant (chatgpt, openai, perplexity, claude, gemini, copilot, or one of the hosts above) puts the visit in AI as well.

Two things make the picture less tidy. First, tags come before the referrer. Vero reads the channel from the utm_medium first, then the utm_source, then the referrer, so a link tagged utm_medium=referral or utm_medium=email lands in Referral or Email even when an assistant sent it. Second, desktop and mobile apps do not always send a referrer at all. A visit with neither a referrer nor a tag lands in Direct, because nothing in the request says where it came from. No analytics tool can recover a referrer that was never sent, so treat the AI row as a floor, not a total.

Crawlers are a separate matter. When an assistant fetches your page to read it, that is a bot request, not a visitor (OpenAI lists its crawlers and their user agents). A fetch carries no assistant referrer, so it never lands in the AI channel, and Vero drops requests whose user agent matches its list of known bots before anything is stored. The AI channel counts people who clicked. If you want to know how often your pages are crawled, your server logs are the place to look.

Read the AI channel in money, not visits

Visits from assistants are usually few next to search, which makes them easy to dismiss. The question that matters is what each visit is worth. Take the launch week in the Vero demo, a fictional SaaS on myapp.com: chatgpt.com, the only assistant in its referrers, sent 240 visitors and €400 of revenue. That is the smallest channel by visitors, about €1.67 per visitor, while the site as a whole made about €1.25 per visitor (€18,423 from 14,724 visitors). Google, by referrer, sent 2,876 visitors and €2,783, a little under €1 each.

One week and 240 visitors is a small sample, so the honest reading is not that AI traffic converts better. It is that the channel is too valuable to leave inside Referral and worth watching over several weeks. Revenue per visitor is the number to track, because it compares channels of very different sizes on equal terms. Vero shows revenue by channel and by referrer on Pro once your payment provider sends sales; the revenue attribution guide walks through the setup.

Set it up in four steps

1. Install the tracker on every page an assistant might cite, not only the home page: docs, explainers, pricing and the blog. The install guide takes three steps.

2. Open Sources and find the AI row in Channels. Then read the referrer rows behind it, so you know which assistant sends what. Visits that arrive tagged are already in the AI row, and the UTM report shows the tag they carried.

3. Connect Stripe, Lemon Squeezy or Paddle so each sale comes back with the visit that brought it. Without revenue, the AI row is one more traffic number. With it, you can put AI next to search and social and compare what each visitor is worth. The Stripe integration is one webhook and its signing secret.

4. Filter by an assistant’s referrer and open Pages to see where those visits start. That tells you which pages assistants actually cite, which is often not the page you would have guessed.

What makes a page worth citing

Nobody can promise a citation, and anyone selling a guaranteed ranking inside an assistant is guessing. What you can do is write pages that are easy to quote: a direct answer near the top, one idea per section, real numbers with their source, and headings that say what the section contains. Those are the same habits that make a page useful to a person who arrives from search.

Some sites also publish an llms.txt file, a plain-text summary of what the product does and where the key pages live. Vero publishes one. It costs an afternoon and states your facts in your own words, but there is no evidence yet that it changes how often a site gets cited, so treat it as a courtesy, not a lever.

What this measurement cannot tell you

It cannot tell you how often an assistant mentioned you without a click. A recommendation that ends with the reader typing your name into a browser shows up as Direct or as branded search, and no referrer will ever connect the two. It also cannot tell you what was asked. The referrer carries a host, never the conversation.

Those limits are the same ones search had before keyword data disappeared from referrers: you measure the visit and what it led to, not the intent behind it. That is still plenty to act on. If the AI row grows and its revenue per visitor holds, the pages it lands on deserve the same care as your best search pages. If it stays flat, you have lost nothing by watching it. And if a source sends visitors but never a sale, the reasoning in stop optimizing for vanity traffic applies here too.

Unsure of a term? Read the glossary

Vicente Torres
Vero is built by one person in Spain. I read every waitlist entry and pick the beta accounts myself.
Vicente Torres, Founder of Vero LinkedIn
Hand-picked beta, waitlist open

Ask for a place in the beta.

Anyone can join the list. The accounts we pick get Pro free for the whole beta.

By joining you agree to receive launch updates and accept our Privacy Policy.

See the demo