How to Track AI Search Referral Traffic from ChatGPT, Perplexity, and Gemini

You can track visits from ChatGPT, Perplexity and Gemini when your analytics receives enough information to identify the source. Start with GA4's built-in AI Assistant channel, check the individual sources and validate that your site records the visits.

AI referral traffic is observable but incomplete. Some discovery produces no visit, and some visits arrive without an identifiable AI source. This guide explains how to report the traffic you can see, which platform reports add context and how to combine those findings with CRM and customer-reported discovery.

Why is AI search referral traffic hard to track?

A referral source tells you where a recorded visit came from. It does not tell you everything a person saw before arriving. Three limits matter when setting up your reports.

• A visit may arrive without a usable referrer or campaign tag. Copying a URL, switching devices or opening links through some apps can leave the original discovery source unidentified.
• Not every AI answer leads to a visit. A person can read an answer mentioning your brand or citing your page without clicking through.
• Google's AI Overviews and AI Mode are part of Google Search. GA4 includes them in Organic Search, rather than separating them into AI Assistant.

A channel rule can group source information already collected. It cannot recover a source that was never recorded. Likewise, a visibility report can show your content appearing in AI answers without proving that those appearances produced particular sessions or conversions.

Which AI sources should you look for in GA4?

Starter list, checked 11 October 2026: look for the following domains in your Session source / medium data. These are places to begin checking, not a complete list or a guarantee that every visit from each platform carries that source.

• ChatGPT: chatgpt.com. Check historical rows for chat.openai.com too, but do not assume that a legacy domain is still used for current referrals.
• Perplexity: perplexity.ai and, if present in your data, www.perplexity.ai.
• Gemini app: gemini.google.com. Keep this separate from Google's AI Overviews and AI Mode.
• Microsoft Copilot: check copilot.com and copilot.microsoft.com. Verify which source your own visits carry rather than assuming one domain covers every experience.
• Claude: check for claude.ai where identifiable referrals are recorded.

The source shown in GA4 is not always a copy of the browser's referrer address. Campaign tags can also supply source information. Inspect the medium alongside the source; identifiable AI visits need not all be labeled referral.

Do not label all traffic from google.com, bing.com or x.com as AI. Those domains also carry ordinary search or social traffic, and a domain alone cannot identify which feature someone used.

Keep the list current from your own reports

Review Session source / medium periodically without the AI channel filter. Investigate unfamiliar sources, check that they belong to the relevant platform and test an outbound link where possible. Keep a short record of the source, observed medium and date checked. Add a reporting rule only when the traffic is identifiable and you need it in your AI view.

Avoid broad matches such as any source containing "ai" or "gpt". They can include unrelated websites. Google's custom channel documentation explains how source conditions work; use verified values from your property.

How to set up AI search tracking in Google Analytics 4

If GA4 already collects data from your site, begin with its existing reports. A custom channel group is optional.

Step 1: Find the built-in AI Assistant channel

1. Open GA4 and go to Reports > Acquisition > Traffic acquisition. If it is missing from your navigation, ask an Editor or Administrator to restore the report.
2. Choose Session default channel group in the table's dimension selector and look for AI Assistant. Use a report filter or comparison to include that channel.
3. Add Session source / medium as a secondary dimension to see which identifiable sources are included.

AI Assistant is a category inside GA4, not a separate tracking tool you need to install. Google uses it for identifiable traffic from supported assistants such as ChatGPT, Gemini, Copilot and Grok. Google's channel definitions exclude AI Overviews and AI Mode from it; those sit in Organic Search.

This channel only covers visits GA4 can identify. It does not count every AI mention, answer or citation, and it cannot identify a visit that arrives without usable source information. An empty row means no sessions were classified there for your selected period, not that nobody discovered you through AI.

Step 2: Check sources and useful outcomes

Switch the table to Session source / medium to inspect the source list directly, including rows outside AI Assistant. Search for the domains above and note their current channel classification.

Use Sessions, Engaged sessions, Engagement rate, Key events and Session key event rate. Select the key event that matters, such as a completed demo request, rather than treating every event as a lead. Add Total revenue only where purchase tracking is configured and relevant.

Use session dimensions for this visit-level view. First user dimensions answer a different question: how GA4 originally acquired that user. Google's Traffic acquisition guide documents the report and available metrics.

Optional: Create a custom channel group

Use this only if you find additional verified sources outside the built-in channel, or need to group traffic differently for your reports. You need Editor access or above.

1. Go to Admin > Data display > Channel groups > Create new channel group. This copies the default group. Give the copy a clear name, such as "AI referral reporting".
2. Add a channel named "Identifiable AI visits". Use one condition group where Default channel group exactly matches AI Assistant.
3. Add a separate OR condition group for an additional verified source. For example, if your data contains perplexity.ai / referral outside AI Assistant, use Source exactly matches perplexity.ai AND Medium exactly matches referral. Include only values you have checked.
4. Reorder the new channel above AI Assistant and Referral, then save the group. GA4 uses the first matching channel.
5. In Traffic acquisition, choose your custom group's session dimension from the table's dimension selector. This is separate from Session default channel group, which remains unchanged.

You do not need to make this your property's primary group. A custom group can reclassify existing source data in reports, but it cannot recreate missing referrals. Record rule changes so you can explain changes in the reported totals.

Step 3: Save a landing-page view

For a reusable page-level report, open Explore > Blank and choose a Free form exploration. Import Session source / medium, Session default channel group and Landing page + query string. If you use a custom group, import its session dimension instead.

Add Sessions, Engaged sessions and Key events. Put Landing page + query string in Rows, Session source / medium in Columns and the metrics in Values. Filter the relevant session channel dimension to AI Assistant or your custom channel. This shows where identifiable AI visits land and what happens during those sessions.

Step 4: Validate collection and classification

1. Use a fresh browser session, such as a private window where you can access the assistant. Click a real outbound link to your site from ChatGPT, Perplexity or Gemini. Record the time, destination and browser or app used. Do not reuse an already active visit on your site.
2. Check the final landing URL. Note any source tags already present and whether redirects preserve them. Do not add your own AI tags to this test: that would test your tagging rather than the platform's native referral behavior.
3. Use Realtime to check that the visit reaches GA4, or DebugView when debug mode is enabled. If needed, ask your analytics owner to inspect the collected landing URL and page referrer. These views help validate collection; they are not the final acquisition report.
4. After normal report processing, check Traffic acquisition for the expected source / medium and channel. Repeat on the browser or mobile app that matters to your audience; one test does not cover every platform experience.
5. If you created a custom group, compare its session total with the included sources over the same dates. Check that ordinary Google, Bing and social traffic remain outside the custom AI channel.

Google recommends Realtime and DebugView for collection checks, while acquisition reports provide more complete attribution processing. If a test is absent, use the troubleshooting section below before assuming there is no AI traffic.

Optional: Set up a Looker Studio dashboard

For ongoing reporting, connect your GA4 property to Looker Studio and build a compact dashboard showing:

• Identifiable AI sessions over time and by source.
• Top landing pages from those sessions.
• Engaged sessions, engagement rate and the key events relevant to your business.
• Session key event rates alongside session and event counts, using the same dates for channel comparisons.

Use the same source definition as your GA4 report. If the connector does not expose your custom group, reproduce its verified source/medium rules and check the totals against GA4. Keep platform visibility and CRM outcomes in separate, clearly labeled panels rather than combining them into a single conversion funnel.

How to check AI Overviews and AI Mode visibility in Google Search Console

Use Google's Generative AI performance report for Search. It reports impressions for AI Overviews and AI Mode: occasions when links to your site were shown in these features.

What the report shows

1. Open Search Console for your property and open the Generative AI performance report for Search.
2. Choose a date range, then inspect Pages, Countries, Devices and Dates. Use the documented search-type filter for Web: text-based or Web: multimodal where relevant.
3. Export the chart and table if you need them in a recurring report.

If the report is absent, Google's documentation says the site may not have enough impressions in these features. An absent report is not a traffic measurement.

What it does not establish

The documented report provides impressions, not a separate set of AI clicks, click-through rates or user queries. Do not use AI Overview or AI Mode search-appearance filters as a way to isolate those clicks. These impressions also appear within broader Search performance data, so do not add them to overall impressions as if they were extra exposure.

This report covers Google Search features, not referrals from the Gemini app or other assistants. It adds visibility context alongside GA4; it cannot identify which AI answer produced an individual visit.

When are server logs useful for AI referral tracking?

Server or CDN logs are an optional diagnostic tool when a test visit appears to reach your site but is missing from GA4. Ask your web or analytics team to check whether the landing-page request contains an identifiable AI referrer or source tag.

Log coverage depends on your hosting, caching and logging configuration. Requests are not sessions: a page visit can create multiple requests, and automated systems create requests too. Logs cannot recover a referrer or tag that was never supplied.

AI crawler traffic versus user referral traffic

Keep automated content access separate from people opening your site. OpenAI distinguishes OAI-SearchBot for search, GPTBot for possible training use and ChatGPT-User for user-triggered actions. Perplexity likewise documents PerplexityBot and Perplexity-User. A user-triggered fetch by an assistant is still not a human browser session on your website.

A browser-like user-agent alone does not prove a request came from a person. Use the vendor's current identification guidance when investigating automation. For more context, see how AI crawlers read your website. You do not need a crawler-analysis project to start reporting identifiable referrals.

What does AI search referral traffic look like compared to organic?

Compare your own recorded traffic rather than assuming AI visitors are more engaged or more likely to convert. Results can differ by source, landing page, market and the action you measure.

• Compare engagement and key event rates over the same period, with the same event definitions.
• Inspect landing pages and markets: a demo-page visit and a visit to an informational article serve different purposes.
• Show the underlying counts. A rate based on a handful of sessions or leads can move sharply after one additional event.

An identifiable-referral report measures the visits your setup can classify. It is not a click-through-rate report for AI answers because it does not provide a matching count of the answers people saw.

Which AI platforms send the most referral traffic?

For your site, answer this by comparing sessions by Session source / medium. Report what each identifiable source contributed during the selected period. There is no platform order that every business should expect, and the Google Search AI visits included within Organic Search cannot be ranked separately using this view.

What Bing AI Performance adds

In Bing Webmaster Tools, select your verified site and open AI Performance. Review Total Citations, Average Cited Pages, cited URLs and the sampled grounding queries, which are phrases the AI used to retrieve cited content. Microsoft's launch documentation covers Microsoft Copilot, AI-generated summaries in Bing and selected partner integrations.

Microsoft also documents preview views for Intents, Topics, Citation Share and Compare. Use them to understand citation context and changes over time where available. They do not turn citation activity into referral clicks or traffic share.

Keep Bing's citation reporting alongside identifiable Copilot visits in GA4, with separate labels.

How to read AI visibility alongside referral traffic

Referral reporting and AI visibility metrics answer different questions. Use them together, but keep clear what each report measures.

A brand can be mentioned without your website being cited, and a page can be cited without a visit. A visit can also follow a later branded search rather than a citation click. The difference between brand mentions and citations matters, but neither is a reliable substitute for sessions.

Do not divide GA4 sessions by citations in a tracked prompt sample and call the result an AI click-through rate. The people visiting your site and the answers collected for your visibility report are not the same measured population.

Superlines provides AI Search analytics for enterprise marketing teams and agencies, including visibility, citations, sources, competitors and multi-brand or multi-market analysis. Keep those findings alongside your referral report to see where you appear and which pages receive identifiable visits, without assigning those visits to particular tracked answers.

Add self-reported discovery for B2B

After a demo request or other conversion, an optional "How did you hear about us?" question can add useful evidence. Include ChatGPT, Perplexity, Gemini or other AI tools where appropriate, and allow a free-text answer. On inbound sales calls, ask the same question directly and record relevant detail in the CRM.

Keep the answer separate from the recorded analytics source. Someone may report ChatGPT discovery while GA4 records a later Google visit. Both can describe part of the journey. Self-reporting also has gaps: people may forget, simplify or skip the question. It adds evidence rather than providing complete or superior attribution.

How to connect referral reporting with business outcomes

For B2B, retain the recorded source and self-reported discovery where available, then use the CRM to follow qualified leads, pipeline and closed-won outcomes. Deduplicate by lead or opportunity: one customer appearing in both sources is still one customer.

Report pipeline separately from closed-won revenue. If you multiply leads by an assumed close rate and deal value, label the result estimated lead value or a revenue scenario. It is not revenue already earned or proof of AI's exact contribution.

For ecommerce, purchase revenue shown against a source is revenue associated with that reporting definition. It does not establish how many purchases would have occurred without AI discovery. AI-related work also has content, software and team costs even when there is no fee for a particular referral click.

For financial interpretation, use how to measure AI Search optimization ROI. For budgets and assumptions before investing, see how to calculate the AI Search business case.

What are common mistakes when tracking AI search traffic?

No AI Assistant row or unexpected channel classification

Check the date range and remove other filters. Inspect Session source / medium directly. If a verified AI source appears outside AI Assistant, investigate its tagging and consider the optional custom group. A missing-referrer visit cannot be recovered by changing a channel rule.

A test visit is missing or shows an unexpected source

Confirm the correct GA4 property and that the landing page loads the tag. Check consent choices, blocking software, redirects and any unwanted-referral settings with your analytics owner. Test in a fresh session and allow report processing. Use optional logs only if the collection check leaves an unexplained gap.

The custom group captures too much or too little

Check exact source values, medium conditions and channel order. Remove overly broad patterns and verify that paid campaigns or ordinary search traffic have not been swept into your AI channel. Compare the custom view with the underlying source rows over matching dates.

The report contains an (other) row

Google uses (other) to aggregate rows when a report has too many distinct values. It is not another name for Direct or proof that AI referrals lost their source. Simplify the report dimensions and inspect a more focused source view.

Visibility rises but visits do not

A citation does not require a click. The tracked questions may also differ from those that customers ask. Review prompt relevance, landing pages and identifiable sources, while keeping AI visibility monitoring separate from referral counts.

A weekly change looks unusually large

Check volumes, rule changes and collection issues before drawing a conclusion. Use consistent periods and annotate changes in your reporting setup. Citation patterns can change over time, but a coinciding traffic change alone does not show that the citation change caused it.

How to improve pages receiving AI referrals

Use the landing-page view to find where identifiable AI visits arrive. Check that those pages deliver the answer, evidence or product information a visitor expects. If the AI cites your statistics page, the user should be able to find the relevant statistic quickly.

Keep useful pages accurate and current, and make the next action clear where it fits the reader's task. Compare engagement and qualified outcomes after changes, while allowing for other changes in traffic. A page appearing in tracked answers and receiving AI referrals is worth inspecting, but that overlap does not reveal the exact answer each visitor saw.

For broader content work, use the GEO best practices checklist. Do not treat technical shortcuts as guaranteed referral growth; our llms.txt guide explains the limits of that proposal.

Start tracking the AI visits you can identify

Begin with Traffic acquisition, inspect AI Assistant and check the underlying sources. Validate a real visit before building more reporting. Add custom grouping only when you have a verified reason, then maintain the source list as platforms and your traffic change.

Keep referrals, platform visibility, customer-reported discovery and business outcomes clearly labeled. Together they help your team understand AI Search performance without implying that every answer, visit and sale can be connected.

Superlines adds visibility, citation, source and competitor analysis alongside your website analytics. Explore where your brand appears across AI Search and use those insights to guide marketing decisions, while GA4 and your CRM report the visits and outcomes they can record.