Marketing AttributionNovember 10, 202618 min read

AI Referral Traffic: How to Track Visitors From ChatGPT, Perplexity and Claude (2026)

AI assistants send fewer visitors than Google, but they convert better. How to find AI referral traffic in GA4 and Plausible, and why a one-question survey shows what analytics misses.

Your analytics says ChatGPT sent you a handful of visitors last month. Google sent fifty times more. Easy call: keep investing in Google, treat AI as a rounding error.

Except the handful from ChatGPT probably stayed longer, read more and converted better. And the real number is higher than your dashboard shows, because a good share of AI-driven visits arrive with no referrer at all and get filed under "Direct".

Note

What is AI referral traffic? AI referral traffic is visits that come to your website from an AI assistant - ChatGPT, Perplexity, Claude, Gemini, Copilot and similar tools - usually when the assistant cites or recommends your site in an answer and the user clicks through. It is different from AI crawler traffic, which is bots reading your pages, not people visiting them.

Tip

Summary: AI assistants send a small but fast-growing share of website traffic, and those visitors convert better than most channels because the assistant has already explained who you are and why you fit. Analytics undercounts them: many AI-driven visits arrive without a referrer and land in Direct. Track what you can in GA4 or Plausible, then close the gap with a one-question "How did you hear about us?" survey.

Key takeaways

  • AI visitors arrive pre-educated. The assistant already summarized your product, compared you with alternatives and pointed the user to you. They show up later in their decision than a typical search visitor.
  • They convert better. In Ruler Analytics' 2026 data, AI referrals converted at 5.8% - the highest of any channel, ahead of paid search (5.4%) and organic search (4.9%).
  • Analytics only sees part of it. Visits from AI apps, copied links and some browsers arrive without a referrer and show up as Direct. Your "AI" line is a floor, not the total.
  • You can track the visible part for free. GA4 has had a built-in AI Assistant channel since May 2026, and Plausible added one in June 2026. A custom channel group in GA4 catches what the default misses.
  • Ask, and the hidden part shows up. A one-question "How did you hear about us?" survey with an "AI assistant" answer option catches what referrers miss. Selge's How did you find us? template is built for exactly this.

Why AI referral traffic converts better

Search traffic and AI traffic look similar in a report. They are not the same visitor.

They arrive pre-educated

A Google searcher clicks a result to start their research. An AI-assistant user usually finishes a good part of that research before they click. They have asked "what's a good on-site survey tool for a small SaaS team?", read a summary, maybe asked two follow-up questions about price and setup, and only then clicked the link the assistant cited.

By the time they land on your site, they already know what you do, roughly what you cost and why you might fit. That is the same visitor you would get from a strong word-of-mouth recommendation: lower volume, higher intent, shorter path to a decision.

That is also why AI traffic is worth more attention than its volume suggests. Ten visitors who were told "this tool fits your use case" are a different audience from ten visitors who skimmed a results page.

What the data says

Ruler Analytics tracks conversions across 110M+ sessions in 13 industries, where a conversion is a qualified lead or sale. Its 2026 data puts AI referrals at the top of the channel table (Ruler Analytics):

ChannelAverage conversion rate
AI referral5.8%
Paid search5.4%
Organic search4.9%
Email4.9%
Referral4.8%
Direct4.7%
Organic social2.23%
Paid social2.11%

AI referrals edge out paid search - a channel you pay for per click - and convert at nearly three times the rate of social. Treat the exact figures with care: they come from one vendor measuring its own customers, and AI volumes are still small. But it matches what you would expect from a visitor who arrives already informed - and other datasets point the same way:

SourceFindingScope
Adobe Digital Insights, Oct 2026AI traffic converted 90% better than non-AI traffic, and grew 71% year over yearUS retail sites, 1 trillion+ visits
Adobe Digital Insights, Jan 2026AI-driven traffic to tech and software sites grew 120% over the 2025 holiday seasonUS, by industry
Ahrefs, Jun 20250.5% of traffic came from AI search, but it produced 12.1% of sign-upsOne SaaS company's own site
Similarweb, Sep 2026AI platforms sent 770.7 million referral visits a month worldwide, more than double the year beforePanel data, June 2025 - May 2026

Every row comes from a company that sells analytics or AI-visibility tools, and none of them is a SaaS-wide benchmark. Read them as direction, not as a target for your own site.

What we see on our own sites

We run Selge's site and a German SaaS directory on Plausible, which separates AI assistants into their own channel. The volumes are small, so read these as signals rather than benchmarks:

  • AI visitors stay far longer. On selge.app, visitors from AI assistants spent about 18 times longer on the site than visitors from Google organic search, and their bounce rate was 15 percentage points lower.
  • They land on data, not on the homepage. About two-thirds of AI-assistant visitors to selge.app entered on one of our benchmark pages - NPS benchmarks by industry and conversion rate benchmarks. Assistants cite pages with sourced numbers.
  • ChatGPT dominates, but not alone. On the SaaS directory, ChatGPT sends about two-thirds of all AI referrals, Perplexity about a fifth, Microsoft Copilot about a tenth, and Claude the rest. On selge.app, ChatGPT's share is closer to four in five.
  • It is no longer negligible. On the directory, AI assistants now account for roughly 3-4% of all visitors - before counting the visits that hide in Direct.

Why your analytics undercounts AI traffic

Web analytics attributes a visit by looking at the referrer - the page the visitor came from - and at any UTM parameters in the link. AI traffic breaks both more often than other channels.

Missing referrers

A visit only shows up as "from ChatGPT" if the browser passes the referrer when the user clicks. That fails in common situations:

  • App traffic. Links opened from desktop and mobile apps don't always pass a referrer, and AI assistants are increasingly used in apps rather than browser tabs.
  • Copied links. A user copies a URL from an answer and pastes it into a new tab, or shares it in Slack. The visit arrives with no referrer.
  • Remembered names. The assistant mentions your brand, the user types your domain or searches your name later. Analytics credits Direct or branded search, never the AI conversation that started it.
  • Privacy settings. Some browsers and privacy tools strip or trim referrers.

ChatGPT softens this a little. OpenAI says it "automatically includes the UTM parameter utm_source=chatgpt.com in referral URLs" (OpenAI). Because the tag is part of the link, it can survive where the referrer is lost. But it only covers clicks on links ChatGPT shows - not copied links, not brand names typed in later - and it can break. In October 2026, when a bug removed the tag from some ChatGPT links for about a week, Adobe's data showed ChatGPT visits falling while Direct visits rose by almost the same amount (Search Engine Journal). The visits hadn't disappeared. Analytics had just stopped recognizing them.

Where those visits end up

What happenedWhat analytics records
User clicks a cited link in an AI assistant in the browserAI assistant channel (GA4 and Plausible)
User clicks a ChatGPT link that has a UTM tag but no referrerChatGPT if your tool reads UTM tags, otherwise Direct or Unassigned
User clicks a link in an AI desktop or mobile appOften Direct
User copies the link from the answer into a new tabDirect
Assistant names your brand, user types your URL laterDirect
Assistant names your brand, user Googles you laterOrganic search (branded)
Colleague forwards a link they got from an assistantDirect, or the colleague's chat tool

Only the first row is reliably attributed. Every other path makes AI look smaller and makes Direct and branded search look bigger than they really are.

Warning

A Direct spike can be an AI signal. If Direct traffic grows while you haven't changed anything offline, and it lands on the pages assistants like to cite - comparisons, benchmarks, pricing - AI assistants are a likely cause. Analytics alone cannot prove it. Asking can.


How to track AI referral traffic in GA4

Since 13 May 2026, GA4's default channel group includes an AI Assistant channel. Google defines it as traffic "from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok", and says it "excludes Google's AI Overviews and AI Mode" (Google). Open Reports → Acquisition → Traffic acquisition and it's there.

Two gaps remain:

  • Untagged medium. ChatGPT's UTM tag sets only the source. A visit with source chatgpt.com and no medium doesn't match the AI Assistant rule, so GA4 can file it under Unassigned instead.
  • Unknown list. Google doesn't publish which domains it recognizes as AI assistants, so you can't check what's included.

A custom channel group based on the source domain fixes both, and Google documents this exact use case:

  1. In GA4, go to Admin → Data display → Channel groups.
  2. Click Create new channel group and name it, for example, "Channels incl. AI".
  3. Click Add new channel, name it AI assistants.
  4. Set the condition to Session source → matches regex and paste the regex below.
  5. Save, then drag the AI assistants channel above Referral in the list. GA4 applies channels top to bottom, so it must come first or Referral will claim the visits.
  6. Save the group. In Reports → Acquisition → Traffic acquisition, switch the dimension to your new channel group.
(chatgpt\.com|chat\.openai\.com|perplexity\.ai|pplx\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com|copilot\.com|deepseek\.com|grok\.com)

These domains come from the source list that Plausible maintains in its open-source code. Google's help page suggests a broader regex that starts with ^.*ai, but that also matches sources such as "email" and "gmail", so the domain list above is safer. When a new assistant starts showing up in your Referral or Unassigned report, add its domain.

A custom channel group re-classifies historical data too, so you can see how AI traffic developed over past months, not just from today.

Tip

Add a conversion view. In the Traffic acquisition report, add your key event (sign-up, demo request, trial) as a column. Comparing the conversion rate of the AI assistants channel with organic search is the fastest way to see whether AI traffic is worth investing in for your site.


How to see AI referral traffic in Plausible

Plausible does the grouping for you. Since June 2026, its Channels report has a dedicated AI Assistants channel - "a known AI assistant tool such as ChatGPT, Claude, Perplexity, Gemini, Copilot and others" (Plausible docs) - and the Sources report breaks it down by assistant. Plausible uses UTM tags as well as referrers, so tagged ChatGPT clicks are counted even when the referrer is missing.

To get more out of it:

  1. Filter by channel. Click AI Assistants in the Channels report to filter the whole dashboard. You now see which pages AI visitors land on and how long they stay.
  2. Check entry pages. The pages AI visitors enter on are the pages assistants cite. That is your shortlist of content to keep accurate and up to date.
  3. Add goals. If you track sign-ups or demo requests as goals, the filtered view shows the AI channel's conversion rate next to every other channel.
  4. Compare periods. AI traffic is still small and noisy for most sites. Compare 28-day windows, not days.

If your analytics tool has no AI channel, it still shows AI assistants as referrer domains. The domain list in the next section works for building a filter in any tool.


Which AI assistants send traffic, and from which domains

AssistantReferrer domains to look forNotes
ChatGPTchatgpt.com, chat.openai.com (older links)Also tags links with utm_source=chatgpt.com
Perplexityperplexity.ai, pplx.aiShows sources next to every answer
Microsoft Copilotcopilot.microsoft.com, copilot.com
Google Geminigemini.google.com
Claudeclaude.ai
DeepSeekdeepseek.com
Grokgrok.com

Google's AI Overviews and AI Mode are a special case: Google explicitly excludes them from GA4's AI Assistant channel, so clicks from them count as Google organic traffic. Your survey is the only way to separate them.


Close the gap: ask "How did you hear about us?"

Referrers capture the clicks. They miss the conversations. The only way to measure AI influence that leaves no referrer is to ask the people it influenced. This is called self-reported (or self-attributed) attribution, and we covered the full method in Self-Attributed Attribution: Why "How Did You Hear About Us?" Matters in the Age of AI. Here is the short version for AI traffic.

The question and the answer options

Ask one multiple-choice question:

"How did you first hear about us?"

  • Google search
  • ChatGPT, Claude or another AI assistant
  • LinkedIn
  • A colleague or friend
  • Blog article or review site
  • YouTube or podcast
  • Other

Two details matter. First, give AI assistants their own option - if it's not on the list, people pick "Google" or "Other" and the signal disappears. Second, keep "Other" with an open-text field: that's where you learn which assistant, and what the user asked.

Where and when to ask

There are two good moments:

  1. Right after sign-up. The highest-quality answers, from people who converted. Add the question to your welcome screen or onboarding.
  2. On your homepage, a few seconds into the visit. More answers, including from people who don't convert yet. Selge's How did you find us? template uses a 5-second delay and a 60-day cooldown, so nobody is asked twice.

On-site surveys work here because the question appears in the moment, without sending anyone to a form. A small survey widget that loads only when needed won't slow the page down.

Reading analytics and survey answers together

The two sources answer different questions. Use both.

Analytics (referrers)Survey ("How did you hear about us?")
What it capturesClicks with a referrerWhat the person remembers
Sees app and copied-link visitsNoYes
Sees "an assistant told me your name"NoYes
SampleEvery visitEveryone who answers
PrecisionExact, but incompleteApproximate, but complete
Best forTrends, landing pages, conversion ratesThe real channel mix

Companies that compare the two keep finding the same gap:

CompanyAI share in analyticsAI share when asked
n8n (workflow automation)About 1% of conversionsAbout 9% of conversions
Otterly.ai (SaaS)ChatGPT 7%, Claude 0.1% of sign-upsChatGPT 11%, Claude 10.6% of sign-ups

In the n8n data, organic and paid search showed similar shares in both sources - only AI was badly undercounted. Ahrefs reached the same conclusion from its own sign-up survey: "If we'd relied on our referral traffic alone, we'd have badly undersold it" (Ahrefs).

The gap between your analytics share and your survey share is your hidden AI traffic. That gap is the number to watch over time.


What to do once you know AI sends you buyers

Once the survey confirms that AI assistants send you customers, the question becomes how to show up in more answers. What works is not exotic:

  1. Publish pages with sourced numbers. Assistants cite pages that state a fact and show where it came from. On our own site, benchmark pages with named sources attract far more AI visitors than anything else.
  2. Answer the question in the first paragraph. Definitions, short summaries and FAQ sections are easy for assistants to quote.
  3. Keep facts current. Pricing, features and comparisons go stale quickly, and assistants repeat whatever is on the page. Add a visible "updated" date.
  4. Make your site easy to read for machines. Allow AI crawlers in robots.txt if you want to be cited, and consider an llms.txt file that summarizes what your site offers.
  5. Measure monthly. Track the AI channel in analytics and the AI share in your survey. If both grow, keep going.

Frequently asked questions

What is AI referral traffic?

AI referral traffic is visits to your website from AI assistants such as ChatGPT, Perplexity, Claude, Gemini and Copilot, usually after the assistant cites or recommends your site in an answer. It is different from AI crawler traffic, which is bots reading your pages.

Does AI referral traffic convert better than Google traffic?

Usually, yes. Ruler Analytics' 2026 data shows AI referrals converting at 5.8%, ahead of paid search (5.4%) and organic search (4.9%). The likely reason is that AI users arrive later in their decision: the assistant has already explained what you do and why you might fit.

How do I track ChatGPT traffic in Google Analytics 4?

Since May 2026, GA4's default channel group has an AI Assistant channel - check Reports → Acquisition → Traffic acquisition. Because ChatGPT's UTM tag sets a source but no medium, some visits can land in Unassigned instead. For full coverage, create a custom channel group in Admin → Data display → Channel groups with an "AI assistants" channel where Session source matches a regex of AI domains, and move it above Referral. Custom channel groups also apply to past data.

Does ChatGPT add UTM parameters to links?

Yes. OpenAI says ChatGPT automatically adds utm_source=chatgpt.com to referral URLs, so analytics tools that read UTM tags can identify ChatGPT clicks even without a referrer. It doesn't help with copied links or people who type your URL after reading about you.

Why does ChatGPT traffic show up as Direct?

Because many AI-driven visits arrive without a referrer: clicks from desktop and mobile apps, links copied into a new tab, and people who type your URL after an assistant mentioned you. Analytics can't attribute a visit it has no referrer for, so it files it under Direct.

How much AI traffic is hidden in Direct?

There is no reliable universal figure. Numbers like "70% of AI traffic hides in Direct" come from single vendors' datasets and measure something narrower. It depends on your audience and which assistants they use. The practical way to estimate it for your own site is to compare the AI share in analytics with the share of people who answer "AI assistant" in a "How did you hear about us?" survey.

Does Plausible track AI traffic?

Yes. Since June 2026, Plausible's Channels report includes an AI Assistants channel, and the Sources report shows individual assistants such as ChatGPT, Perplexity, Claude, Copilot and Gemini. It reads both referrers and UTM tags.

What should a "How did you hear about us?" survey include for AI?

A dedicated answer option such as "ChatGPT, Claude or another AI assistant", plus "Other" with a text field. Without a dedicated option, AI-influenced visitors pick "Google" or "Other" and the signal is lost. Selge's How did you find us? template includes it by default.


AI assistants won't replace Google as your biggest traffic source this year. They don't need to. A channel that sends fewer, better-informed visitors who convert at the top of the table deserves its own line in your reports - and a direct question to the people it sends.

Start with the question. Selge's How did you find us? template ships with the AI answer option, the trigger and the cooldown already set. Build it free and see what your analytics has been missing.

Tags:ai referral trafficchatgpt trafficllm traffictrack chatgpt traffic in ga4self-reported attributionhow did you hear about us
2-minute setup

Ready to hear what your visitors think?

Pick a template, paste one script tag, start getting real answers. No developer required.

No credit card requiredFree plan available

Free to build - pay only when you go live