LLM traffic in GA4: measuring visits and leads from ChatGPT

7 min read
LLM traffic in GA4: measuring visits and leads from ChatGPT
7 min read

Your board has started asking what AI search is worth to the business, and most marketing teams cannot answer with numbers. The visits exist, but LLM traffic from ChatGPT, Perplexity, Gemini and Copilot has tended to sit scattered across Referral, Direct and Organic Search in GA4, so nobody can say how many leads or sales it brings.

This article sets out how we measure it: the GA4 channel setup, the landing page and conversion reports we build, and the limits of referrer data. It should help you decide what to ask your team for, what to put in front of leadership, and how much weight the number can carry.

Why measuring AI referral traffic matters to the business

Without a number, budget decisions about AI search are made on anecdote. One side of the room points to a competitor named in a ChatGPT answer, the other points to flat organic sessions, and the spend either stalls or goes ahead with nothing to judge it against.

With a baseline, the conversation changes. You can compare the conversion rate and revenue of AI referral traffic with organic search, see which pages AI assistants send people to, and judge GEO work on leads and revenue rather than on screenshots of chatbot answers.

It also changes where the content budget goes. When the landing page report shows assistants sending buyers to a pricing page or a comparison guide, those pages earn the next round of investment, and pages that attract no AI visits are judged on their organic search value alone.

There is also a reporting risk to manage. Visits are only the part of LLM visibility that produces a click. Plenty of AI answers mention a brand without sending anyone to the site, so a small visit count does not mean a brand is absent, and a growing one does not tell you why. Setting that expectation early saves a difficult board conversation later.

How we set up LLM traffic tracking in GA4

We run the same seven steps on every GA4 property. The setup is quick; the value comes from keeping the reports running every month so there is a trend to read.

Before any of it, we check the basics: that the GA4 tag fires on every template, that key events are defined for the actions that matter (a demo request, a quote, a purchase), and that internal traffic is filtered out. AI referral numbers are small enough that a broken tag on one template can hide most of them.

  1. Check the AI Assistant channel GA4 already gives you. Since May 2026, Google's default channel group has included an AI Assistant channel: when a referrer matches Google's list of assistants, GA4 sets the medium to ai-assistant and the campaign to (ai-assistant). Google names ChatGPT, Gemini, Deepseek, Copilot and Grok as examples and states that AI Overviews and AI Mode are excluded. We open the Traffic acquisition report and note what sits in that row.
  2. Add a custom channel group you control. Google does not publish its full list of assistants, and its release note does not say whether older data is reclassified. Custom channel groups apply to reports retroactively, and a standard property can have two of them with up to 50 channels each. We copy the default group and add an AI assistants channel where the source matches an explicit list of hostnames, such as chatgpt.com, chat.openai.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai.
  3. Put the AI channel above Referral. GA4 assigns a session to the first channel whose rules it matches, so an AI channel placed below Referral loses most of its visits to the referral rule, and Google's own help page gives the same instruction. We avoid loose patterns such as gpt or ai on their own, which sweep in unrelated sources, and we review the hostname list each quarter as new assistants appear.
  4. Cross-check the sources. In June 2026 GA4 added a Source Group dimension that consolidates source values for common platforms, with built-in grouping for ChatGPT and Perplexity, populated retroactively. We use it to check that our channel and Google's grouping agree. OpenAI also says ChatGPT adds utm_source=chatgpt.com to referral links (OpenAI publishers FAQ), so those visits carry chatgpt.com as the source even when no referrer is passed.
  5. Build a landing page report for AI visits. In Explore, we set landing page as the row, filter to the AI assistants channel and add sessions, engaged sessions, key events and revenue or lead value. This shows which pages assistants actually send people to. The list is often short and tends to include pages nobody chose for AI search, such as an old guide or a comparison table, which tells the content team what is already being cited.
  6. Report conversions by channel every month. We compare key events, conversion rate and revenue for the AI assistants channel with Organic Search and Paid Search over the same period. One limit to know: Google says custom channel groups cannot be used in the Key events paths report, so for assisted conversions we read the default group, where the AI Assistant channel now appears.
  7. Add what GA4 cannot see. Clicks from AI Overviews and AI Mode arrive as Google organic traffic, so we pull Search Console's generative AI performance report, available to all sites since 31 August 2026. It shows impressions by page, country, device and date, but no clicks or queries. For answers that never produce a click, we add prompt tracking: a fixed set of buyer questions checked in each assistant on a schedule, chosen using the approach in how to choose which prompts to track.

How to measure it and what to report upwards

We report LLM traffic as one page in the monthly pack, with five numbers: AI assistant sessions and their share of all sessions, key events and conversion rate against organic search, revenue or pipeline value, the top ten landing pages, and Search Console AI feature impressions for priority pages.

What good looks like is set against your own data, not an industry figure. We have not seen a published benchmark for AI referral conversion rates that we would put in front of a board. The useful questions are whether the trend is rising, whether those visitors convert at least as well as organic search visitors, and whether they land on the pages you want them to find.

For businesses that sell through a pipeline, we also pass the session source into the CRM with each lead, so sales can see which deals started with an AI referral. That is the figure boards respond to, and it survives the gap between a form fill and a signed contract.

Tell leadership plainly that the number is a floor. Some apps and browsers pass no referrer, and GA4 files a visit with no referrer and no campaign tags under Direct. Assistant answers that mention you without a link do not appear at all. For the wider picture, pair this report with the approach in how to measure performance in AI search.

When it does not fit, and the mistakes we see most

If AI assistants send a handful of sessions a month, monthly reporting produces noise rather than insight. In that case we set the channel up so the history is there, report quarterly and put the effort into visibility work first.

  • Treating AI referral traffic as the whole of AI influence. It counts clicks, not mentions or recommendations.
  • Placing the AI channel below Referral, so the custom group shows almost nothing and the team concludes AI search does not matter.
  • Copying a regex from a blog post without checking it against your own source list, which can miss assistants or catch unrelated sites.
  • Expecting GA4 to separate AI Overviews and AI Mode clicks. They sit inside Google organic traffic, and Search Console is the place to look.
  • Judging GEO work after a few weeks. We hold a baseline of at least three months, because low volumes swing from month to month.
  • Reporting sessions without key events. A visit count with no conversion data invites the wrong debate.
  • Leaving the hostname list unchanged for a year while new assistants and new domains appear.

Questions to ask your team or agency

Our AI SEO and GEO team sets this reporting up as the baseline for its work. Whoever does it, these questions show whether it is sound.

  • Which channel does our ChatGPT traffic sit in today, and has anyone checked the AI Assistant channel GA4 added in May 2026?
  • Do we have a custom channel group for AI assistants, and is it ordered above Referral?
  • Which landing pages receive AI referral traffic, and how do they convert compared with organic search?
  • How are we tracking AI Overviews and AI Mode, given GA4 counts those clicks as organic?
  • What are we telling the board this number does not include?

Frequently asked questions

What is LLM traffic?

LLM traffic is visits to your website from AI assistants built on large language models, such as ChatGPT, Perplexity, Gemini, Copilot and Claude, usually when someone clicks a link in an answer. In GA4 it appears in the AI Assistant channel or as referral traffic from those domains. It measures clicks only, not every time an assistant mentions or recommends your brand.

Does GA4 track ChatGPT traffic?

Yes. Since May 2026 the GA4 default channel group has included an AI Assistant channel that classifies visits from recognised assistants, including ChatGPT. OpenAI also adds utm_source=chatgpt.com to links in ChatGPT answers, so those visits show chatgpt.com as the source. Some AI visits still arrive without a referrer and are counted as Direct, so treat the figure as a minimum.

Can you see AI Overviews traffic in Google Analytics?

Not separately. Clicks from Google's AI Overviews and AI Mode arrive as Google organic search traffic, and Google's definition of the GA4 AI Assistant channel excludes them. Search Console's generative AI performance report, available to all sites since 31 August 2026, shows impressions in those features by page, country, device and date, but it does not show clicks or queries.

Sources

Let’s create your future growth plan

Get My Growth Plan
Visibility Playbooks

Related Playbooks

All Playbooks
More Insights

More articles we
think you’ll like

Explore More Insights