Which GA4 Reports Show Referrals From ChatGPT and Perplexity?

You can track referral traffic from ChatGPT and Perplexity in GA4 by reading the Session source and Session source/medium dimensions in the Traffic acquisition report, then building a custom channel group that collects those domains into a single row. ChatGPT referrals usually reach your site with chatgpt.com as the referrer, while Perplexity sends sessions tagged with perplexity.ai. Because neither company lets you attach UTM parameters to a citation, the referrer string is the only automatic signal GA4 receives, and it survives only when the browser passes it through. Sessions that arrive without a referrer fall into Direct or Unassigned, which is why raw referral numbers always undercount real AI traffic. The workable setup combines three layers: the default AI Assistant channel group that Google now ships, a custom channel group built on a regex rule, and log file checks that confirm which crawlers actually fetch your pages.

Which GA4 Reports Show Referrals From ChatGPT and Perplexity?

The Traffic acquisition report, Explorations, and Realtime give GA4 users three ways to isolate ChatGPT and Perplexity as session sources.

Open Reports, then Acquisition, then Traffic acquisition. Switch the primary dimension from Session default channel group to Session source/medium, and chatgpt.com / referral and perplexity.ai / referral appear as separate rows. The Session source dimension gives you the same data without the medium attached, which keeps the table short when you only track a handful of assistants.

For deeper analysis, open Explore and build a free form report. Use Session source/medium as the dimension, then add Sessions, Engaged sessions, and Key events as metrics. A filter such as Session source matches regex chatgpt\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai turns that exploration into a reusable AI referral view you can save for the whole team.

Realtime shows the last 30 minutes and confirms within seconds that a fresh citation is producing clicks. Standard GA4 properties also apply sampling to exploration queries once a query crosses 10 million events, so keep date ranges tight when you compare months side by side.

How Do You Build a Custom Channel Group for AI Assistant Traffic?

A custom channel group in GA4 groups ChatGPT, Perplexity, Copilot, Claude, and Gemini sessions under one AI Assistant label.

Channel groups are the cleanest way to keep assistant traffic in one place, because a single row survives every date range, comparison, and segment you apply later. The setup takes about two minutes and works in both standard reports and explorations.

1. Open Admin, then choose Data display, then Channel groups.
2. Select Create new channel group.
3. Name the group, for example, AI Assistants.
4. Add a condition where Session source matches the regex list of assistant domains.
5. Set the channel name to AI Assistants and save.
6. Return to the Traffic acquisition report and pick your new group from the channel group dropdown.

Two details matter here. In 2025 Google added an AI Assistant channel group to its default channel group definitions, so many properties already separate assistant sessions without any custom rule. A custom group still earns its place when you want a narrower definition, for example only the two assistants you actively optimize for, or when you want to exclude a domain that sends low quality sessions. Keep an internal note with the exact regex and the date you created it, so the definition stays readable six months from now.

Why Does ChatGPT Traffic Sometimes Look Like Direct or Unassigned Traffic?

ChatGPT referrals arrive without a referrer when the link opens in an app or a privacy browser, so GA4 files them as Direct.

GA4 assigns a session to Direct when the browser sends no referrer and the landing URL carries no campaign parameters. Several everyday situations strip that referrer, starting with assistants that open links in a native app or a built in reader view. The page loads, the visitor reads, and GA4 records the session with no source at all.

• In app browsers and native app link opens that remove the referrer header
• Assistants that render a page inside their own reader or preview view
• Consent settings that block analytics storage during the first pageview
• Link wrappers and redirects that drop the original domain along the way
• Repeat visits where someone reads on a phone and pastes the URL on a laptop later

Unassigned appears when collected data does not fit any channel rule, often because analytics storage was blocked or because an app to web handoff lost the signal. The practical response is to stop treating Direct as one bucket. Break Direct sessions down by landing page, device category, and country, then look for pages that only perform well when an answer engine cites them.

How Do You Verify AI Referral Data Outside GA4?

Raw server logs, chatbot user agents, and Search Console data confirm which AI systems crawl and cite your pages in 2026.

Log files are the ground truth for AI activity. Every request carries a user agent, and each assistant uses distinct ones. OpenAI operates GPTBot for model training and OAI-SearchBot for ChatGPT search indexing, while ChatGPT-User fetches a page when a user question requires live content. Perplexity follows the same pattern with PerplexityBot and Perplexity-User. Anthropic runs ClaudeBot, and Apple runs Applebot alongside the Applebot-Extended robots token.

robots.txt governs each of these separately. Blocking GPTBot does not block OAI-SearchBot or ChatGPT-User, and blocking a search crawler removes your pages from answers that could have produced referral sessions. Compare log dates with GA4 referral dates next. When a crawler visits on a Tuesday and ChatGPT referrals rise on a Thursday, you have a sequence you can explain rather than a coincidence. Google Search Console adds a second view for clicks that arrive from AI Overviews inside the search results page, which never show up as referral traffic.

How Do You Connect AI Referral Tracking to Content Decisions?

AI referral data tells content teams which pages earn citations, and the ChimpanSEO blog uses that loop across 80 published articles.

The ChimpanSEO blog doubles as the test bench for this workflow. The team has generated and published more than 80 articles on the company blog in a public content marketing experiment, and each article ships as an Italian and English pair produced by the same automated pipeline the product runs for customers. AI referral data matters because it answers a question a keyword report cannot: which pages are useful enough for an engine to quote?

Sort assistant sessions by landing page, then compare that list with the pages that earn the most citations in ChatGPT and Perplexity answers. Pages with strong AI visibility share three traits. They open with a question as the heading, they answer inside the first paragraph, and they name the main entity in plain language before adding detail.

• Check which landing pages collect sessions from chatgpt.com and perplexity.ai each week
• Rewrite the answer capsule on pages that get crawled often but never earn a referral
• Add schema markup so the entity, author, and date stay unambiguous for retrieval pipelines
• Keep internal links pointing at the pages you want quoted, since assistants follow them too

Content freshness signals belong in the same routine. Refresh the statistics and the publication date on the pages that already earn citations, and leave the direct traffic pages for later.

Frequently Asked Questions

These answers cover GA4 referrer tracking for ChatGPT and Perplexity, direct traffic gaps, and real AI citation measurement.

Can GA4 show which specific ChatGPT answer sent a visitor?

No. GA4 records the referrer domain, such as chatgpt.com, and the landing page on your own site. It cannot see the prompt, the conversation, or the position of your citation inside an answer. To learn which topics trigger citations, you have to test prompts yourself and check your log files for ChatGPT-User requests on the same day.

Does ChatGPT always pass a referrer to GA4?

No. Referrer data passes through only when the browser sends it. Links opened inside the ChatGPT mobile app, or through a reader view, frequently arrive with no referrer, and those sessions land in Direct. Expect your GA4 referral count for ChatGPT to sit below the real number of visits the assistant sends you.

Do I need UTM parameters to track AI traffic?

UTM parameters only work on links you control and publish yourself. You cannot add them to a citation that ChatGPT or Perplexity generates, because the assistant builds that link on its own. Rely on the referrer domain instead, and keep UTM tagging for the campaigns, newsletters, and social posts where you own the destination URL.

What is the difference between GPTBot and ChatGPT-User?

GPTBot collects pages for OpenAI model training, while ChatGPT-User fetches a page during a live user request. Blocking GPTBot in robots.txt stops training access but leaves ChatGPT free to fetch your content for answers. OAI-SearchBot sits between them and handles indexing for ChatGPT search results.

Should ChatGPT referrals count as referral traffic or organic traffic?

GA4 classifies them as referral traffic by default, because a domain in the referrer field triggers the Referral channel. Google’s AI Assistant channel group moves the same sessions into a dedicated bucket without changing the underlying source. Both views are useful: channel groups for reporting, source and medium for debugging.

How long does GA4 keep AI referral data?

GA4 event data retention defaults to two months and can be extended to a maximum of 14 months in the property settings. Assistants change their referral behaviour often, so export the AI Assistant channel group to BigQuery or Looker Studio monthly if you want a comparison that stretches past the retention window.

What Is the Next Step for AI Referral Tracking?

Your next step is to review the AI Assistant channel group in GA4 each month and annotate every spike with the page that earned it.

Export the last 90 days of Session source/medium, mark every assistant domain, and place that list next to the articles you published in the same window. The pattern becomes obvious quickly: a handful of pages carry most of your citations, and they are usually the ones with a direct answer in the opening paragraph. Improve those pages first, document the regex you used for your channel group, and check your server logs once a month to confirm which crawlers are active. That routine turns a vague question about AI visibility into a numbers driven habit you can repeat every quarter.

Related reading

• How does Perplexity decide which websites to trust?
• How Do ChatGPT, Perplexity and Gemini Handle Citations Differently in 2026?
• Come citano le fonti ChatGPT, Perplexity e Gemini in modo diverso nel 2026?
• What specific content structure does ChatGPT prefer for citations?