SEO/AEO/GEO

Pixis Visibility

How to Track AI Referral Traffic in GA4: ChatGPT, Perplexity, and Gemini

By Shreshtha Bansal

Director of Growth

10 September, 2026

Your AI referral report shows 100 visits from ChatGPT. Is that good?

GA4 can tell you where those visitors landed, what they did, and whether they converted. It cannot tell you how often your brand appeared before those clicks happened, which prompts surfaced it, how competitors performed in the same answers, or how much visibility produced no visit at all.

That is the measurement gap this article addresses.

GA4 measures attributable traffic after the click. Pixis Visibility measures brand presence inside AI answers before the click. Neither view is complete on its own.

This guide shows how to track identifiable AI referrals in GA4, then connect that traffic with prompt-level visibility, citations, sentiment, and competitor performance in Pixis Visibility. The result is not simply an AI traffic report. It is a way to understand whether visibility in AI answers is translating into meaningful website activity.

Key takeaways

GA4 records the AI referrals it can identify, not every visit or brand exposure influenced by AI.
Pixis Visibility measures whether, where, and how a brand appears across relevant AI answers.
Use GA4 to evaluate post-click behavior and Pixis Visibility to explain the discovery environment behind it.
Start with GA4's native AI Assistant channel, then add source-level or custom reporting only when it improves the analysis.
Do not treat referral traffic as a proxy for total AI visibility. Many AI appearances produce no click, and some clicks arrive without a usable referrer.

The visibility problem behind the traffic report

AI discovery is not a conventional list of blue links. An assistant may mention a brand directly, cite the brand's website, rely on a third-party source, recommend a competitor, or answer the question without sending the user anywhere.

GA4 sees only the final case in which a detectable visit reaches your site. It does not observe the answer environment that influenced the user.

Imagine that ChatGPT referral sessions increase from 100 to 150. Several explanations are possible:

your brand began appearing for more relevant prompts;
an existing page earned more citations;
a third-party source started recommending your product;
the same visibility generated a higher click-through rate;
or referral handling changed and more visits became identifiable.

GA4 can confirm the traffic increase, but it cannot distinguish among those explanations.

Pixis Visibility supplies the missing pre-click evidence. It tracks brand appearances across ChatGPT, Gemini, Claude, and Perplexity, with Google AI Overviews treated as a separate search surface. Marketers can examine prompt-level visibility, competitor presence, brand sentiment, cited sources, and the gaps that may be limiting discovery.

This creates a clear division of work:

Pixis Visibility asks: Did the brand appear, for which prompts, against whom, in what context, and with support from which sources?
GA4 asks: Did an identifiable visitor arrive, where did they land, and what did they do next?

The rest of this guide begins with the GA4 setup, then shows how to interpret that data through the visibility layer.

Can GA4 track traffic from AI assistants?

Yes, when the visit carries enough referral information for GA4 to identify its origin.

Google added an AI Assistant channel to GA4's default channel group in May 2026. Qualifying sessions are assigned the medium ai-assistant and the campaign (ai-assistant). This gives marketers a native way to view a portion of AI-referred traffic without building a custom rule.

The important phrase is a portion.

GA4 depends on the information available when a visitor lands on your site. If an assistant passes a recognizable referrer, the visit may be attributed to that source. If the referrer is removed by an app, browser, redirect, privacy setting, or copy-and-paste journey, GA4 cannot reliably reconstruct it. The visit may then appear as Direct or under another available source.

So the right question is not, "How much traffic did AI generate in total?" GA4 alone cannot answer that with certainty. A better question is:

How much identifiable referral traffic did AI assistants send, and what did those visitors do after arriving?

That distinction keeps the analysis useful and honest.

Why AI referral traffic deserves its own view

AI referral traffic remains a small share of overall web traffic for many sites, but its strategic value is larger than its volume alone suggests.

An SE Ranking study found that AI platforms accounted for 0.32% of traffic across its dataset. That is a sample-specific result, not a universal benchmark, but it shows that measurable referral activity is already present.

Conversion behavior can also differ by site type. Microsoft Clarity found a substantial sign-up conversion advantage for AI traffic among publisher and news websites. That finding should not be applied to every industry, but it gives marketers a reason to evaluate AI visitors separately rather than burying them inside Referral or Direct.

The practical value of tracking this channel is straightforward. You can learn:

which AI assistants send identifiable visits;
which pages receive those visitors;
whether those sessions engage or convert;
which topics appear to attract qualified discovery;
and where your content strategy deserves closer investigation.

Method 1: Use GA4's native AI Assistant channel

Begin with the native report. It is the fastest and least fragile option because Google maintains the underlying classification.

In GA4:

Open Reports.
Go to Acquisition and select Traffic acquisition.
Set the primary dimension to Session default channel group.
Find AI Assistant in the table.

You can then compare sessions, engaged sessions, engagement rate, key events, revenue, and other metrics relevant to your property.

Google documents the channel under its GA4 release notes. Because classifications can evolve, Google's documentation should remain the source of truth for the native definition.

This view is the right baseline, but it groups qualifying assistants together. If you need to compare ChatGPT with Perplexity or Gemini, inspect the underlying source data.

Method 2: Compare individual AI referral sources

From the Traffic acquisition report, change the primary dimension to Session source / medium. Then search for known source domains, such as:

chatgpt.com
chat.openai.com
perplexity.ai
gemini.google.com
claude.ai
copilot.microsoft.com

This gives you a source-level view of identifiable sessions. Use it to compare landing pages and outcomes rather than traffic volume alone.

For example, a page may receive only a modest number of visits from Perplexity but produce a high rate of product-page views or demo requests. Another page may attract more visits from ChatGPT but little downstream action. That difference is more useful than simply announcing that AI traffic increased.

Add Landing page + query string as a secondary dimension when possible. This reveals where referred users entered the site. It does not prove that an AI platform cited that page in a particular answer, but it gives you a reliable destination for further investigation.

Method 3: Create a custom AI referral channel

A custom channel group is useful when:

you want to include sources not yet classified by the native channel;
you need a definition tailored to your reporting standards;
or you want one consistent AI referral category for dashboards and recurring analysis.

To create one:

Go to Admin.
Under Data display, open Channel groups.
Select Create new channel group.
Add a channel named AI Referral.
Define it using the relevant source or source-domain conditions.
Place the channel above broad rules such as Referral so qualifying sessions are classified correctly.
Save the group and use it as a reporting dimension.

GA4 standard properties support up to two custom channel groups, with up to 50 channels in each group. Plan the taxonomy before using one of those slots. Google explains the configuration and limits in its custom channel groups documentation.

Custom groups can classify historical data when selected in reports. However, if you make a custom group the property's primary channel group, its primary-group values populate from that point forward. This distinction matters when comparing reports across time.

Also note that custom channel groups are a reporting feature. Their values are not added as a field in the GA4 BigQuery export. If your organization relies on warehouse reporting, recreate the classification logic in SQL or another transformation layer.

A starter regex for AI referral sources

If your rule supports a regular expression, start with a conservative, domain-specific list:

Treat this as a starting point, not a permanent standard. Check the source values already present in your property before deciding which domains belong in the rule.

GA4 regex uses full-match behavior by default. Anchoring the expression with ^ and $ makes the intended match explicit and reduces false positives. Google's regex guidance explains this behavior.

Avoid loose patterns such as ai, chat, or openai without domain boundaries. They can match unrelated sources and inflate the channel.

Review the rule periodically, but do not add a platform merely because it is popular. Add it when you can verify the source domain in your data or authoritative documentation.

Build an Exploration for deeper analysis

Once the channel is visible, create a GA4 Exploration to analyze quality by source and landing page.

Useful dimensions include:

Session source / medium
Session default channel group
Custom channel group
Landing page + query string
Device category
Country

Useful metrics include:

Sessions
Engaged sessions
Engagement rate
Average engagement time per session
Key events
Session key event rate
Total revenue, where applicable

Apply a segment or filter for the native AI Assistant channel or your custom AI Referral channel. Then compare AI-referred sessions with Organic Search, Referral, and your site average.

Focus on questions a marketer can act on:

Which landing pages attract the most engaged AI-referred visitors?
Do AI visitors reach product or pricing pages?
Which assistants contribute to key events?
Are particular countries or devices overrepresented?
Does one content topic consistently attract higher-quality sessions?

Do not create custom dimensions for values GA4 is not receiving. A custom dimension does not automatically reveal the AI platform, answer, prompt, or citation. It only reports an event parameter that your implementation actually sends.

What GA4 cannot tell you

GA4 is useful after a visitor reaches your site. It is not a complete AI visibility system.

It cannot recover a missing referrer

If the browser or application does not pass referral information, GA4 cannot infer the original assistant with confidence. Server-side tagging may improve the reliability of data your site receives, but it cannot recreate upstream information that never arrived.

It cannot show the user's prompt

The referral normally does not include the question a user asked. GA4 can identify a source and landing page when that data is present, but not the conversation that produced the click.

It cannot prove a citation

A landing page associated with AI referral traffic is evidence that a visitor arrived there from an identifiable source. It is not, by itself, proof that the page was cited for a particular prompt or appeared consistently across answers.

It cannot measure zero-click visibility

An AI assistant can mention or recommend a brand without generating a website visit. GA4 will not record that exposure because no session occurred.

This is why referral traffic and AI visibility should be measured as related but distinct layers.

Where UTM parameters help

UTM parameters provide clean attribution when your brand controls the link, such as links placed in a custom GPT, partner experience, sponsored placement, or campaign asset.

For example:

?utm_source=chatgpt&utm_medium=ai-assistant&utm_campaign=product-discovery

UTMs do not solve attribution for links independently generated by public AI assistants. Marketers cannot append campaign parameters to citations they do not control.

Use UTMs for owned distribution and referral data for earned visits. Do not combine the two as though they represent the same acquisition mechanism.

How Pixis Visibility helps solve the AI referral tracking problem

The problem is not only that AI traffic is fragmented across GA4 channels. It is that referral analytics begin too late in the journey.

GA4 can measure a session only after someone reaches the website, and it can identify the AI source only when usable referral information survives the click. This leaves marketers with four blind spots: visits with missing referrers, AI answers that produce no click, no visibility into the prompt that shaped the answer, and no context on the competitors or sources that influenced the recommendation.

Pixis Visibility does not attempt to relabel unknown Direct traffic or manufacture attribution. Instead, it measures the discovery environment independently of the click. It monitors relevant prompts across ChatGPT, Gemini, Claude, and Perplexity, while treating Google AI Overviews as a separate search surface.

This helps marketers address the original problem in five practical ways:

Measure visibility without depending on referral data. See whether the brand appears in tracked AI answers even when no user clicks through to the website.
Recover the missing prompt context. Identify the topics and prompts where the brand appears, where competitors appear instead, and where the brand is absent.
Understand citation influence. See which domains and pages are being cited, then identify source-placement and backlink gaps that GA4 cannot expose.
Interpret traffic changes more carefully. Compare movements in GA4 referral sessions with changes in AI visibility, citation presence, competitor share, and sentiment rather than assuming one metric explains the other.
Turn measurement into an action plan. Use prompt, keyword, content, and source gaps to decide which pages to create, improve, or support through authority-building work.

The result is a more complete measurement model. GA4 shows the identifiable visits generated after an AI interaction. Pixis Visibility shows the larger field of brand exposure that may precede a visit, influence a later search, or remain entirely zero-click.

Once the GA4 report is configured, four comparisons make this combined view useful:

1. Traffic by source versus visibility by engine

GA4 may show which identifiable assistant sent the most sessions. Pixis Visibility shows whether that traffic aligns with how often the brand appears across the engines being monitored.

A platform can produce high visibility but little referral traffic because users receive an answer without clicking. Another may produce fewer brand appearances but more visits. That difference helps separate visibility from referral behavior.

2. Landing pages versus cited sources

GA4 shows which page received the visit. Pixis Visibility shows which domains and pages AI engines cite when answering relevant prompts.

Together, these views can reveal whether your own content is earning citations, whether third-party sources are shaping the answer, and whether a high-performing landing page is connected to broader citation presence.

3. Conversions versus prompt coverage

If a landing page converts AI-referred visitors well, use Pixis Visibility to find the prompts and topics where that page or brand already appears, as well as relevant gaps where it does not.

This turns a conversion result into a content question: where could the same useful page earn more qualified exposure?

4. Traffic change versus competitive visibility

A fall in referral sessions does not automatically mean demand declined. The brand may have lost appearances to a competitor, the cited-source mix may have changed, or click behavior may have shifted while visibility remained stable.

Pixis Visibility adds the competitive and answer-level context needed to investigate those possibilities.

Use the combined data as three connected measurement layers:

Pixis Visibility: Where does the brand appear in AI answers, how is it represented, and which sources support that presence?
GA4: Which identifiable AI referrals reached the site, and what did those visitors do?
Business reporting: Did that activity contribute to leads, sign-ups, sales, or revenue?

Google Search Console adds another useful layer for Google's own search surfaces. Google announced dedicated Search Generative AI performance reporting in June 2026 and completed the global rollout in August 2026. The reports cover generative AI impressions, pages, countries, devices, and dates. See Google's announcement for the current scope.

For a deeper look at the visibility layer, read Pixis's guides to getting cited by ChatGPT, interpreting AI traffic conversion data, and finding the prompt-level gaps hidden by an aggregate visibility score.

Frequently asked questions

Why does some AI traffic appear as Direct in GA4?

Direct is used when GA4 has no usable source information. This can happen when a referrer is removed by an app, browser, redirect, privacy control, or copy-and-paste journey. A custom channel group cannot recover the original source after that information is lost.

Does GA4's AI Assistant channel include every platform?

Do not assume universal coverage. Use the native channel as a maintained baseline, then inspect source / medium values in your own property. Google's definition may change as platforms and referral behavior evolve.

Can GA4 identify the prompt that sent a visitor?

No. GA4 generally receives referral and landing-page information, not the user's prompt or the content of the AI answer.

Do custom channel groups work on historical data?

Custom groups can be selected for historical reporting. If a custom group is set as the primary channel group, however, the primary values populate forward from that change. Follow Google's documentation when validating a historical comparison.

How often should the source list be reviewed?

Review it quarterly and whenever unfamiliar referral domains appear. Validate additions against your own acquisition data before changing the production rule.

Measure what GA4 can prove

Tracking AI referral traffic is not about forcing every AI-influenced visit into a neat channel. It is about building the most reliable view the available data supports.

Start with GA4's native AI Assistant channel. Inspect individual sources and landing pages. Create a custom group only when it adds reporting value. Then use Pixis Visibility to see the prompts, competitors, sentiment, and cited sources behind AI discovery.

GA4 tells you what identifiable AI-referred visitors did after the click. Pixis Visibility shows where the brand is present before the click and where it is still absent. Together, they turn an incomplete referral signal into a more useful AI search measurement strategy.