ChatGPT traffic in Google Analytics: data from 235 sites

In almost every marketing meeting, someone asks about ChatGPT traffic in Google Analytics: "so how much is ChatGPT actually sending us?". Usually the answer is a short silence while someone opens Analytics on their phone and nobody can find the row.

So we went and looked. We reviewed GA4 for 235 company websites in the US and Latin America, from July to September 2026. The median share of traffic coming from AI assistants (ChatGPT, Gemini, Perplexity, Claude and Copilot) was 0.26% of visits. Google organic search sent about 36 times more (from here on I'll just call it organic).

Up to that point, I'd file it as an after-lunch chat for SEO teams. But on 2 out of 3 sites, those visits completed an important action (a form, a purchase, a sign-up) more often than organic visits did. My read: I wouldn't move budget because of the volume, and I would check which pages those visits land on, which GA4 now separates on its own.

Numbers first, then how to measure the traffic in your own account.

How much ChatGPT traffic does a site get?

Not much. At the median, traffic from assistants was 0.26% of the total between July and September. If your site gets 100,000 visits in a quarter, about 260 come from there (an illustrative example, just for scale).

Only 15% of sites go above 1%, and 20 of the 235 didn't get a single visit from an assistant in three months. By region it barely changes: 0.23% in the US, 0.26% in Latin America and 0.42% in other markets.

The other way to see it is the ratio. Across the 215 sites with at least some visits from assistants, organic sent 36 visits for every one from an assistant (median). In the lowest quartile it was 17 to 1, and in the highest, 85 to 1.

One detail before moving on: that 0.26% is a floor. Some of the traffic coming from AI apps arrives without saying where it came from, and GA4 counts it as direct. I explain why further down.

Which assistants send that traffic?

ChatGPT, by a mile. Across the 168 sites with at least 50 visits from assistants, ChatGPT was 88% of AI traffic (median). Gemini came second with 5.8%, Perplexity had 1.4%, Claude 1.3% and Copilot 0.8%.

In other words, when someone in the meeting says "AI traffic", in practice they're talking about ChatGPT plus a bit of everything else. And its share rose 3.8 points between the second and third quarters, so the mix is concentrating even more on a single assistant.

Does AI traffic convert better than Google organic?

On most sites, yes. To compare it properly we only used the 125 sites with key events set up in GA4 and at least 50 visits from each source, and we compared each site with itself (because every company defines differently what counts as a conversion).

The rate of visits that completed at least one key event was 3.6% from assistants and 1.75% from organic, at the median. Site by site, AI won in 67% of cases, with a median ratio of 1.43 (it converts 1.4 times as often as organic).

It holds by region. Among the 67 US sites with conversion data, AI won in 69%, with a ratio of 1.61. Among the 20 in Latin America it won in 80%, with a ratio of 1.42. Keep in mind that 20 sites is not many: I'd read it as "it also happens in Latam".

Where there's almost no difference is engagement: 60% of visits from AI and 59% of organic visits were engaged. So they stick around about the same, but more often end up doing something.

My read is that whoever arrives from ChatGPT has already done half the work with the assistant (compared, asked, ruled things out) and comes to your site to close. I can't prove that with this data, because it's a correlation: the data shows those people convert more, but it doesn't prove AI is the cause. For deciding where to put your attention, that doesn't matter. For promising your manager that "if we double AI traffic we'll double sales", it matters a lot.

To check whether the result depended on sites with very little volume and few conversions, we reran it with a threshold of 200 or more AI visits (98 sites): the ratio was 1.47 and AI won in 70%. In the previous quarter (120 sites), 1.52 and 73%.

Is traffic from AI assistants growing?

Not in this period. Between April-June and July-September 2026, the median change in traffic from assistants was 0%, across 172 sites. Half went up and half went down (the bottom quartile fell 28% and the top one grew 45%).

Google organic, meanwhile, had a median change of -11% over the same period. That third quarter includes mid-year holidays in several markets, so I wouldn't read either number as a long-term trend. It does help deflate the idea that AI traffic is doubling every quarter everywhere.

How do you see ChatGPT traffic in Google Analytics 4 (GA4)?

Google Analytics 4 now has its own channel in the default channel group, called AI Assistant. According to the official documentation, when the referrer matches an assistant the tool recognizes, GA4 sets the medium to "ai-assistant" and the campaign to "(ai-assistant)". The documentation gives ChatGPT, Gemini, Deepseek, Copilot and Grok as examples, and leaves out AI Overviews and AI Mode, which stay inside organic search.

To measure ChatGPT traffic in your account:

1. Go to Reports > Acquisition > Traffic acquisition.
2. Look at the table with the Session primary channel group (Default channel group) dimension and find the AI assistants row.
3. To identify which assistant sends each visit, switch the dimension to Session source. ChatGPT shows up as chatgpt.com.
4. Add the key events metric (or the session key event rate) to compare it with organic in the same report.

Two warnings for this step. Google doesn't publish the full list of assistants it recognizes, and it doesn't say whether it reclassified data from before the change. If you're going to compare with the first half of 2026, filter by source (chatgpt.com, perplexity.ai and friends) rather than by channel. That's what we did in the study.

Why did ChatGPT traffic get lost in Referral?

Before the native channel, GA4 classified visits from an assistant like any other traffic coming from another site. A click on a link inside ChatGPT arrived as chatgpt.com / referral and got mixed into Referral with any blog that linked to you.

Two other cases made it messier:

• ChatGPT adds utm_source=chatgpt.com to many of the links it shows. If there's no medium, the visit gets medium (not set) and can end up in Unassigned.
• When the link opens from a desktop or mobile app that doesn't send the referrer, GA4 has no way to know where it came from and counts it as Direct.

The native channel fixes the mix-up with Referral when the assistant sends the referrer. Google doesn't say what happens with links that only carry utm_source, so I'd also check the chatgpt.com / (not set) row. And apps without a referrer stay the same: if the app doesn't send the data, no channel can recover it. That's why any AI traffic number you see in Google Analytics is a minimum.

How do you create an AI channel group with regex?

If you want to separate assistants the native channel doesn't recognize, or have your own stable definition for reporting, it's still worth creating a custom channel group. Analytics help includes an example on its custom channels page. Step by step:

1. Go to Admin > Data display > Channel groups.
2. On the default group, choose Copy to create new. That way you keep every channel and can see what share of the total is AI.
3. Add a new channel, for example "AI assistants", with the condition Source matches regex. A regex with full domains works better than loose words: ^(.*\.)?(chatgpt\.com|chat\.openai\.com|gemini\.google\.com|perplexity\.ai|claude\.ai|copilot\.microsoft\.com)$.
4. Drag the channel above Referral. GA4 assigns each visit to the first channel that matches, so if it sits below, Referral takes everything and your channel shows zero.
5. Save and pick the new group in the Traffic acquisition report.

Custom groups also apply to historical data, so you see previous months the same day. A standard property allows two custom groups, on top of the default one. And I'd avoid putting "chat", "ai" or "gpt" on their own in the regex: they catch support widgets and any .ai domain that has nothing to do with it.

Which pages get that traffic?

This is the part I'd look at first. In a free-form exploration, cross the source or channel with the Landing page dimension. That lets you identify which pages the visits from assistants land on, and the list is almost always short: a guide, a service page, a comparison.

Those pages are the ones the assistant considered good for answering someone. For an SEO team, it's the list of content that's already working in AI. If they convert worse than the rest, you have a concrete and cheap piece of work to do.

What doesn't Google Analytics see about AI traffic?

Quite a lot, and it's worth knowing before you present the number:

• AI Overviews and AI Mode: clicks from there arrive as google / organic and can't be separated in GA4. To measure AI Overviews and AI Mode there's the Search Console AI report, and I explain how AI Mode works in another article.
• Copilot inside Bing arrives as Bing traffic and gets mixed with the search engine.
• Apps that don't send a referrer end up in Direct.
• Answers that mention your brand without generating a click don't exist for GA4. They're most of what happens inside an assistant, and they're measured with other tools: I wrote about how to measure your brand's visibility in ChatGPT, Gemini and Perplexity.

Study methodology

This is how the study was built:

• Data source: Google Analytics 4 (Data API) for 456 properties we have access to through consulting work, from company websites in the US and Latin America.

• Main period: July 2 to September 29, 2026 (90 days). Comparison: April 3 to July 1, 2026 (90 days).

• Criteria set before looking at results: properties with 500 or more sessions in the main period (251). We excluded 16 properties whose names indicate testing, duplication or disuse ("test", "old", "staging", "dev", "roll-up" and similar) and 2 without API access. That left 235.

• Region: the country with 50% or more of the period's visits (US 130, Latam 42, other 63). It's classified by where the traffic comes from, even if the company is headquartered elsewhere.

• AI assistants: sessionSource equal to chatgpt.com, chatgpt, chat.openai.com or openai; gemini.google.com or gemini; perplexity or perplexity.ai; claude.ai or claude.com; copilot, copilot.com or copilot.microsoft.com. All paid media was excluded (cpc, ppc, paid, ads, display), as was openai-ad, which corresponds to OpenAI's ads.

• Organic: google / organic, which includes AI Overviews and AI Mode.

• Conversion: rate of sessions with at least one key event (sessionKeyEventRate), compared between AI and organic within each site. Only sites with key events set up and 50 or more visits from each source (125).

• Growth: sites with 500 or more sessions and 30 or more AI sessions in both quarters (172).

• Assistant mix: sites with 50 or more AI visits (168).

• Medians and 25th and 75th percentiles were calculated per site. No raw data was stored, only the aggregates, and no number in this article belongs to a single site.

Limits of this data

• It's a correlation and doesn't show causation: whoever arrives from an assistant may be further along in their decision, and that could explain part of the difference.
• Each site defines its own key events. That's why we only compare within the same site and never rates across sites.
• The 0.26% is a floor. Some traffic from AI apps arrives without a referrer and GA4 counts it as direct.
• AI Overviews and AI Mode are inside google / organic and aren't separated.
• The quarter-over-quarter comparison includes seasonality (mid-year holidays).
• It's a convenience sample: these are the properties we have access to, not a representative sample of the market.

What should a marketing team do with this data?

If I had to boil it down to a list for Monday, it would be this, and the first step takes ten minutes:

1. Confirm that GA4 already shows you the AI assistants channel and note the date it appeared. If you need more assistants or want to compare with earlier months, create your channel group with the regex and put it above Referral.
2. Pull the list of landing pages for that traffic and check that each piece of content has a clear next step: the form, the pricing, the booking button. These are few visits, and they arrive with the decision pretty much made.
3. Compare the AI key event rate with organic's within your site, and don't report it as a finding until you have a reasonable number of conversions. With 10 conversions, any difference could be noise.
4. Don't move SEO or media budget into "optimizing for AI" because of the volume. At 0.26% of visits, the channel doesn't yet pay for a restructure. What does pay off is having the SEO team look after the pages already getting those visits, and measuring your visibility in the answers with tools built for that, because that's where it's decided whether you get recommended.

And if someone in the next meeting asks how much ChatGPT is sending, you can now answer with the row open in GA4 and the median from 235 sites next to it.