How to Track Traffic From ChatGPT, Gemini, and Copilot in GA4

To track traffic from ChatGPT in GA4, open the Traffic acquisition report and inspect Session source/medium for chatgpt.com. Then check the AI Assistant channel and create a source-based report that includes Gemini and Copilot. Use landing pages and meaningful key events to see what these visitors do after arriving.

A mention in an AI answer can introduce your business to someone who has never heard of you. The next question is whether that person visits your website and takes a useful next step.

For a blogger, that might be a newsletter signup. For a service business, it could be an enquiry. For an online store, it might be a purchase.

I recommend building your report around those outcomes. A traffic total becomes more useful when you know which pages attracted visitors and whether those pages helped them.

This guide explains how to track traffic from ChatGPT, Gemini and Copilot in GA4, including the current AI Assistant channel, a precise source filter and the attribution gaps you should expect.

Start with the current GA4 AI Assistant channel

Google introduced AI Assistant traffic measurement on May 13, 2026. Its release notes explain that recognized AI referrers receive an ai-assistant medium and appear in the AI Assistant channel, with an associated (ai-assistant) campaign value.

That changes the starting point. You do not always need a custom channel just to find AI traffic.

However, checking only Referral can miss visits classified under the new channel. The channel is also not retroactive: it classifies traffic going forward, so historical AI visits remain under whatever channel they were originally recorded in. Manually tagged links and unrecognized sources also make a source-level check valuable.

Google’s default channel definitions name ChatGPT, Gemini and Claude as its published examples of recognized AI Assistant referrers; Google has not published the complete list. Copilot is not confirmed among the named examples, which is the reason this guide still builds an explicit Copilot check into the source-level and custom-channel steps below rather than relying on the automatic channel alone. Perplexity is a similar case: several independent trackers of this rollout report it still lands in Referral rather than the native AI Assistant channel. Build your own verification into Step 2 rather than assuming the automatic channel already covers every platform this guide discusses.

Google’s default channel definitions also explicitly separate Google’s AI Overviews and AI Mode, which fall under Organic Search. Keep that distinction in your report. A visit from the Gemini app and a visit from a Google AI Overview are different acquisition situations.

What you need before starting

You need a GA4 property collecting data from the website, access to its reports and a date range that contains recorded visits. For a small website, I suggest starting with the last 28 complete days and comparing them with the previous 28 days.

Check that your important enquiry, signup or purchase events work before using them to evaluate traffic quality. A missing event can make a useful source look ineffective.

The steps below use standard GA4 report names. Navigation can differ when a property’s report collection has been customized. If Traffic acquisition is not visible, search for that report or ask the property administrator to make it available.

Step 1: Find ChatGPT traffic in Traffic acquisition

Open Reports and locate Acquisition, then Traffic acquisition. Select your date range and change the primary table dimension to Session source/medium.

Search the table for chatgpt.com. If you are reviewing older data, also look for chat.openai.com. Record the source and medium exactly as GA4 displays them.

Next, switch to Session default channel group and inspect the AI Assistant row. Add Session source/medium as a secondary dimension where available to see the contributing sources.

The Traffic acquisition report is designed to explain where sessions come from. Use it for this visit-focused question.

Do not require the medium to equal referral. A source-based check lets you inspect ai-assistant, referral and other recorded medium values without excluding them prematurely.

Why Session source matters

GA4 distinguishes user, session and event scopes. Google’s explanation of traffic-source scopes describes First user source as the source that originally acquired the user, while session dimensions describe session acquisition.

Imagine someone first visits through a social post and returns later through a ChatGPT link. First user source answers how you originally acquired that person. Session source answers the question relevant to the later visit, subject to GA4’s attribution processing.

Use the session dimension consistently when comparing traffic from ChatGPT, Gemini and Copilot. Mixing it with First user source creates a report that answers two different questions.

Step 2: Identify Gemini and Copilot sources carefully

Inspect the source values below in your own property. Treat them as a starting list to verify, rather than a promise that every platform interaction sends the same information.

A source value identifies acquisition information available to analytics. It does not establish which prompt someone entered or prove that your brand was recommended.

Avoid adding the whole openai.com domain automatically. A visit from a company blog or documentation page is not necessarily a ChatGPT search visit.

Likewise, bing / organic may describe ordinary search traffic. Without additional evidence, assigning all of it to Copilot would inflate your AI traffic estimate.

For your first report, keep the scope narrow enough that another person can understand and review every rule.

Step 3: Create a reusable GA4 exploration

A saved exploration lets you return to the same source and landing-page view without rebuilding it each month. Google’s free-form exploration guide explains the dimensions, metrics and filters used here.

Open Explore and create a free-form exploration. Give it a clear name, such as ChatGPT Gemini Copilot traffic.

Import these dimensions:

• Session source.
• Session medium.
• Landing page + query string.

Import Sessions, Engaged sessions, Engagement rate and Key events as metrics. Add Total revenue only when ecommerce revenue is implemented and relevant.

Put Session source and Session medium in Rows, and Sessions and your chosen outcome metrics in Values. Under Filters, select Session source, choose matches regex and paste this expression as one line:

^(chatgpt\.com|chat\.openai\.com|gemini\.google\.com|copilot\.microsoft\.com)$

Apply the filter. Then duplicate the tab and use Landing page + query string as the main row dimension to inspect entry pages. Keep the same source filter.

Why this filter is intentionally specific

The pattern matches the four complete source values listed above. It excludes unrelated domains that merely contain words such as chat, AI or GPT.

Google’s regex documentation explains that regex matching is full and case-sensitive by default. Escaped dots represent literal dots; vertical bars separate alternatives. Custom channel definitions have their own case-handling rules, so do not assume every GA4 surface behaves identically.

This expression is for Session source. It is not a filter for Session source/medium, a full page URL or Page referrer. Those fields contain different strings.

If your report contains a verified additional source variant, add that exact value deliberately. For example, the current expression will not match a value beginning with www. unless you add it. This is a transparent coverage limit, and it prevents silently accepting every possible subdomain.

Before relying on the filtered total, compare it with the unfiltered source rows. The matched session counts should agree for the same dates and dimensions.

Step 4: Add a custom channel only if you need it

A custom channel is useful when you want your chosen source definition alongside other acquisition channels, especially for historical comparisons.

Follow Google’s custom channel group instructions:

• With Editor access or above, open Admin, then Data display, then Channel groups.
• Create a new group from the default group and name it AI source reporting.
• Add a channel named Selected AI sources.
• Set Source to match regex and use the expression above.
• Place it before the existing AI Assistant and Referral channels so the specific rule is evaluated first.
• Save the group and select its session-scoped dimension in Traffic acquisition.

GA4 assigns traffic to the first matching channel in the group. Custom channel groups can classify available historical data retrospectively; they cannot recover information that was never collected.

Keep this separate from changing the property’s primary channel group, whose new criteria apply going forward. For this guide, selecting the custom dimension in the report is sufficient.

Step 5: Understand ChatGPT UTM tracking

OpenAI’s publisher documentation says ChatGPT includes utm_source=chatgpt.com in referral URLs from ChatGPT search results.

For example, a destination link might look like this: https://example.com/guide/?utm_source=chatgpt.com

That parameter is useful acquisition evidence, but it is not a measurement of every ChatGPT interaction. Redirects, copied links and browser or app behaviour can affect what ultimately reaches your website.

Check that your redirect chain preserves legitimate tracking parameters until analytics can read the landing URL. If your website strips them immediately, ask your developer to review the implementation.

Do not add utm_source=chatgpt.com to ordinary internal links or unrelated promotional links. That creates misleading campaign evidence. You also cannot force every independently generated AI citation to use a tracking URL you choose.

Keep a record of any links your own team deliberately tags for assistant campaigns. Report those separately when you need to distinguish controlled distribution from independently earned discovery.

Step 6: Measure useful actions and landing pages

Once you can identify visits, ask which pages help those visitors move forward.

For an educational article, review newsletter signups and relevant onward journeys. For a service page, review completed enquiry forms. For a store, examine purchases and revenue alongside engagement.

Use key events that represent a real outcome. A form-button click does not establish that the form submitted successfully. Check the event against a completed submission before trusting it as a lead count.

Your first reporting table can stay small:

Here is a hypothetical example. ChatGPT sends 120 recorded sessions, including six sessions with a successful enquiry event. That gives a session enquiry rate of 5%: six divided by 120.

Use sessions containing the enquiry event for that calculation. Dividing six event occurrences by 120 sessions would answer a different question if one visitor submitted repeatedly.

Suppose Gemini sends 30 sessions and one enquiry session. Its observed rate is about 3.3%. The sample is too small to confidently declare one platform the better source. Report the counts alongside the percentages and review a longer period.

I would also compare landing-page intent. Visitors reading an introductory guide should not be judged against visitors arriving on a page designed to request a quotation.

Step 7: Validate the report before using it for decisions

Check both the website measurement and the report rules.

Open a genuine link from an assistant to your website and record the time, destination and platform. A test from a fresh browser session reduces interference from an existing site session, but does not guarantee a particular attribution outcome.

Inspect whether the destination retains relevant parameters. Use your normal tag-testing workflow to confirm that the page view and a test outcome event reach the intended property. Mark or exclude internal tests according to your existing measurement process.

Then review the processed acquisition data after allowing time for reporting. Real-time activity can confirm collection without proving the final source classification.

Google explains in its campaign and traffic-source documentation that encountering new campaign information during an existing session does not automatically start another session. This is one reason repeatedly clicking test links can produce confusing acquisition results.

Finally, compare the source-filtered exploration with Traffic acquisition for identical dates. If totals differ, inspect filters, scopes and data-quality notices before assuming traffic has disappeared.

Why ChatGPT, Gemini, or Copilot traffic may not appear

An empty report needs investigation. It does not prove that nobody has discovered your business through AI.

No identifiable visit was recorded

Someone may read an answer without opening its sources. A brand mention, crawler request and human website visit are separate events. GA4 cannot record a website session that never occurred.

For the content side of discovery, see my guide to optimizing for AI search. Keep discovery observations separate from website traffic totals.

The report excludes the actual medium or source

A referral-only filter can omit AI-assistant traffic. A Session source regex applied to Session source/medium can also return no matches. Remove filters, inspect the actual values, and rebuild the rule from that evidence.

The journey loses attribution information

Copied URLs, redirects and some app-to-browser journeys may not preserve the information needed to identify an assistant. Consent choices and blocking can also limit collection.

Do not assign unexplained Direct traffic to ChatGPT. A missing source is uncertainty, not proof of AI discovery.

The date range or data configuration differs

Standard reports and explorations can have different data availability and processing characteristics. Check the property’s retention settings, report dates and data-quality indicators when historical comparisons do not line up.

Avoid using today’s partial data to judge a trend. Compare complete periods and note any tracking or classification changes.

What GA4 cannot tell you about AI search?

GA4 can help measure identifiable website visits and actions. It does not provide a complete record of:

• Every AI answer that mentioned your business.
• The exact private prompt that led a visitor to your page.
• Every citation that received no click.
• The reason an assistant selected your content.
• The source of all visits that lost referral or campaign information.

Also avoid treating a landing-page report as a full citation report. It shows recorded entry pages, which may differ from the originally linked URL because of redirects. Unclicked citations are absent.

Keep Google’s AI search experiences separate too. A google / organic row alone cannot distinguish an AI Overview click from a conventional search click. My Google AI Overviews SEO guide covers the broader visibility strategy.

My monthly reporting approach

I would review this report monthly and record the selected sources, date range, rule version, and any analytics changes beside the results.

Then answer three business questions in plain language. Which sources brought identifiable visits? Which pages helped those visitors? What should change next?

If a guide attracts visits but few readers continue, improve its relevant internal links and next step. If a service page attracts enquiries that do not fit your offer, check whether the page explains your audience, scope and limitations clearly.

At Leelija Web Solutions, my company provides website, app and software development alongside SEO and social media marketing. My practical view is that acquisition reporting should help connect those activities with a better customer experience.

Start with a report your team can explain. Expand it when the additional detail helps you make a decision.

Frequently asked questions