Why does ChatGPT traffic show up as direct in Google Analytics?

Some ChatGPT clicks reach your site without a referrer, for example from apps, in-app browsers or links copied into a new tab, so Google Analytics counts them as direct. Links that carry the utm_source=chatgpt.com parameter stay attributable, because the parameter value becomes the session source. For the rest, look at direct sessions on deep landing pages people rarely type, and treat the result as an estimate, not a measurement.

Can I see AI Overviews traffic separately in Google Analytics?

Not as a separate line. Google documents that traffic from AI features such as AI Overviews and AI Mode is reported in the Search Console Performance report within the Web search type, alongside classic results. In Google Analytics those clicks arrive from Google Search and are attributed as Google organic. Any split you report between AI features and classic results is therefore an inference, not a measurement.

Does a CTR drop mean I lost visibility or just lost clicks?

Not necessarily. In Search Console, compare impressions and clicks for the same queries and pages over the same periods. If impressions hold or rise while clicks fall, your pages are still shown but earn fewer clicks. If impressions fall too, you have lost visibility. Check at page and query level before concluding, since a site-wide CTR can hide both patterns at once.

Can I see AI crawler visits in my server logs?

Yes. AI crawlers and many agents fetch pages without running the Google Analytics tag, so they appear in server logs rather than in your reports. Filter access logs by user agent for names such as GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot or ClaudeBot. Because user agents can be spoofed, check the request IPs against the ranges each vendor publishes, where one exists.

How much traffic do AI platforms usually send to a website?

There is no reliable general figure, and we could not verify a dated cross-site benchmark for this article. The share varies by industry, by site and by the type of query an answer covers. Measure your own: compare sessions in an AI channel with total sessions over the same period, then add your undercount estimate as a range rather than a single number.

How to Measure AI Referral Traffic in Google Analytics

To measure AI referral traffic in Google Analytics, create a custom channel group with an AI channel that matches session source against a regex of AI assistant domains, and place it above Referral. Then account for two limits: AI visits that arrive without a referrer land in direct or unassigned, and AI Overviews clicks are recorded as Google organic.

What Google Analytics can and cannot show about AI referrals

Here, AI referral traffic means sessions whose session source is an AI assistant domain: chatgpt.com, perplexity.ai, claude.ai, gemini.google.com or copilot.microsoft.com. The default channel group files most of them under Referral, mixed with every other site that links to you.

What Google Analytics records when someone clicks a link in an AI answer

When a user clicks a cited link and the browser passes a referrer, Google Analytics records the AI domain as the session source. Some clicks arrive with no referrer: apps and in-app browsers do not always send one, and a link copied into a new tab never does. Those sessions count as direct, or as unassigned when a tagged source matches no channel rule.

ChatGPT links often carry a utm_source=chatgpt.com parameter. When it is present, Google Analytics uses its value as the session source even without a referrer. On 8 October 2026 we could not open OpenAI's own page documenting this parameter, so treat it as a widely reported pattern to confirm in your Session source data, not as a vendor guarantee.

Why AI Overviews and AI Mode clicks land in Google organic

AI Overviews and AI Mode sit inside Google Search. Google's Search Central page on AI features, verified on 8 October 2026, says of their traffic in Search Console: "they're reported on in the Performance report, within the "Web" search type." We infer that these clicks share the Web totals with classic results, with no separate line. In Google Analytics, they arrive from Google Search and are attributed as Google organic like any other result.

Search Console still helps when CTR falls on a query. Compare impressions and clicks first. Stable or rising impressions with fewer clicks mean the page is still shown but earns fewer clicks. Falling impressions mean lost visibility. These are two problems with different fixes.

How to set up an AI channel group in Google Analytics

The source regex, and which domains to include

In Admin, open Data display, then Channel groups, and create a new channel group. It starts as a copy of the default channels. Add a channel named AI referrals with one condition, Source matches regex, using this starting pattern:

Escape the dots: an unescaped dot matches any character, so chatgpt.com written bare would also match chatgpt-com. The pattern keeps chat.openai.com, the legacy ChatGPT domain, so any session still carrying it is caught. It also catches tagged visits, because a utm_source value becomes the session source.

Place the AI channel above Referral

Order decides the result. Google's help page on custom channel groups, verified on 8 October 2026, states: "Traffic is included in the first channel whose definition it matches given the current order of channels in the group." Referral matches any session with a referral medium, so if it sits above your AI channel, AI sessions stop there. Use Reorder to move AI referrals above Referral, then save. That quote covers order only: confirm on the same page whether a new group applies to dates before you created it, before reporting a trend across that date.

Read it in Traffic acquisition and the landing page report

Open Reports, then Acquisition, then Traffic acquisition, and switch the primary dimension from the default channel group to your custom one. AI referrals gets its own row with sessions, engagement and key events. Add Landing page as a secondary dimension to see which pages AI platforms send visitors to. The channel group is one of four ways to isolate this traffic:

Method	What it catches	Setup effort	Works on historical data	Main limitation
Exploration filter	Sessions matching the regex in one exploration	Low	Yes, within data retention	Stays inside one exploration
Custom channel group	The same sessions as a channel in standard reports	Low	Check the help page	Misses visits with no referrer or tag
Comparison or segment	Matching sessions beside other traffic	Low	Yes, for the selected range	Reapplied report by report
Server-side logs	Crawler and agent requests, plus referrers	High	As far back as logs are kept	Needs log access; counts requests, not sessions

How to validate and maintain the AI regex over time

A regex is correct on the day you write it. AI platforms add domains, rename products and launch apps, so check it on a schedule.

Once a month, open an exploration with Session source/medium as the dimension. Add one filter that keeps sources matching terms such as gpt, ai, chat, copilot or perplexity, and a second that excludes sources your regex already matches. A short term like ai also matches unrelated domains, so expect noise. Look up each remaining source, confirm it belongs to an AI assistant, and add it.

Version the regex in a changelog that records the date, the domain added or removed, the reason and who made the change. ChatGPT's move from chat.openai.com to chatgpt.com is the reference case: a pattern written only for the old domain would have quietly lost ChatGPT sessions after the switch. We could not verify an OpenAI source dating that move on 8 October 2026, so your own changelog is the record that matters.

When the definition changes mid-period, do not compare a month counted under the old regex with one counted under the new one. Rerun both periods in an exploration filtered with the current regex, within your data retention window, and report that figure. Note the change date on any chart that spans it.

The same logic works outside Google Analytics. Privacy-first analytics tools and server-side or first-party tagging also record the referrer and UTM parameters. Apply the same domain list and utm_source rule so that numbers from different tools describe the same traffic.

How to estimate what the AI channel undercounts and overcounts

AI visits that arrive as direct traffic

In-app browsers, copied links and some apps strip the referrer, which pushes AI visits into direct or unassigned. Google Analytics cannot recover them, but you can estimate their size.

This heuristic is our inference, not a documented method. Filter direct sessions by landing page and look for deep pages, such as a specific guide or comparison URL, that people would rarely type or bookmark. If those pages also appear in your AI channel, part of their direct traffic is plausibly AI-driven. Write the result as a range, label it an estimate and keep it out of the AI channel total.

Bot and agent sessions that inflate the channel

Google Analytics excludes known bots automatically, and its help page, verified on 8 October 2026, states that "you cannot disable known bot traffic exclusion". The exclusion targets known bots, so agents and headless browsers that run the tag like ordinary browsers can still be recorded as sessions. Signals to watch in the AI channel: very short engagement, no key events, and spikes concentrated on a single page.

Server logs cover what Google Analytics never sees. AI crawlers fetch pages without running the analytics tag, so they do not reach your reports. Filter your access logs by user agent for strings such as GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot or ClaudeBot. User agents can be spoofed, so check requests against IP ranges where the vendor publishes them.

How to read AI referral conversions without overclaiming

Key events and attribution when AI is one touchpoint

Start with your own data. In Traffic acquisition with the custom channel group, compare the key event rate of AI referrals with Organic Search over the same period. That comparison reflects your site and your audience.

Published benchmarks come from narrow cohorts. Seer Interactive's case study, consulted on 8 October 2026, lists the conversion rate for ChatGPT referrals as "ChatGPT - 15.9%". That is a result from a specific sample, not a gain to expect on yours.

AI can be an earlier touchpoint: someone reads an answer, clicks once, and converts later through search or direct. Google's attribution help page, consulted on 8 October 2026, states: "There are 3 attribution models available in the Attribution reports in Google Analytics properties". Check which reporting attribution model your property uses in Admin, then open the Attribution reports to see where AI referrals appear in journeys, not only as the last click.

When a conversion gap is large enough to trust

Small samples swing hard. With a few dozen AI sessions, one conversion moves the rate by several points. Our rule of thumb, an inference and not a documented threshold: wait until the period holds enough sessions and conversions that one more conversion barely moves the rate, and widen the date range if it does not. Report session and conversion counts next to every rate.

To put conversions into money without a false point estimate, start with estimating the revenue impact of AI visibility as a range.

Where to start measuring AI referral traffic this week

Create the AI channel group with the escaped regex, and move AI referrals above Referral.
Save a Traffic acquisition view with the custom channel group and Landing page as a secondary dimension.
Check that a session tagged utm_source=chatgpt.com lands in AI referrals, and do not add tagged visits on top of the channel total when you sum sources by hand.
Start a dated regex changelog and put a monthly Session source audit in the calendar.
Before you report the first number to a client or to leadership, write down an undercount estimate and the result of a bot check.
Pair referral numbers with citation tracking. Clicks show only part of AI visibility, since an answer can name your brand without anyone clicking. See automating AI visibility checks across ChatGPT, Perplexity and Gemini and building a monthly AI visibility client report.

To see which AI answers name your brand before any click happens, run a free Namedrop scan next.