Get a free AI visibility report
This guide covers how GA4 AI traffic reports through the AI Assistant channel Google shipped in May 2026, when a manual regex still earns its place, and the visibility layer GA4 was never built to reach.
For most of the past two years, measuring AI traffic in Google Analytics meant building something yourself. You wrote a regex, hoped no platform had changed its referring domain, and accepted that the number was directional at best. That era is over. Google now classifies recognized AI assistant referrals into their own channel by default.
The convenience is real, and so are the limits underneath it. The native channel only sees visits carrying a referrer Google recognizes, which leaves whole categories of AI-influenced traffic sitting quietly in other buckets. This article covers the new setup, where a manual filter still pays for itself, and the line between what click-based analytics can measure and what it structurally cannot.
On 13 May 2026, Google added a dedicated way to measure traffic from AI assistants inside Google Analytics. When a visit arrives with a referrer that matches a recognized AI assistant, three dimensions are populated automatically.
ai-assistant(ai-assistant)Nothing needs to be configured. As Search Engine Land reported at launch, the point of the update is to let teams track AI clicks, see which AI sources are trending, and compare that traffic against traditional channels without building workarounds first.
This is the part to handle carefully in any internal report, because Google documents the channel in two places and the two do not match. The What's new announcement names ChatGPT, Gemini, and Claude as examples. The Default Channel Group definitions page names a different set, including ChatGPT, Gemini, DeepSeek, Copilot, and Grok, and it explicitly excludes Google's own AI Overviews and AI Mode. Neither page publishes a complete enumerated list of recognized referrers.
Perplexity is the notable absence. It does not appear in the published definition, so Perplexity sessions continue to classify as Referral rather than AI Assistant, despite being one of the higher-intent AI sources. Confirm the behaviour in your own property before reporting it either way.
Two practical consequences follow. Coverage for platforms outside the named examples cannot be assumed, and any platform list you publish internally should carry the date you verified it.
Sessions that occurred before the channel reached your property remain in Referral. That creates a specific trap in reporting: a year-over-year or quarter-over-quarter comparison that spans the rollout date will show AI Assistant traffic appearing from nothing, and Referral traffic dropping by a similar amount, purely as an artifact of reclassification.
Rollout also arrived at different times for different properties. If a property shows little or no AI Assistant data for a stretch after the announcement, treat that as rollout timing before concluding the traffic is absent. Annotate the date your property started populating the channel and hold trend claims to the period after it.
The fastest read takes four clicks and no configuration at all.
That row is your headline number. Sort the table by sessions to see where AI sits relative to organic search, direct, referral, and paid, then add engagement rate and key events as secondary metrics so the conversation is about quality rather than volume alone.
The channel total answers "how much," not "from where." To split it, change the primary dimension to Session source / medium and look for the sources paired with the ai-assistant medium. Each recognized platform reports under its own source value, so you get a per-engine breakdown without writing anything.
To see which pages the traffic hits, add Landing page + query string as a secondary dimension while the AI Assistant filter or dimension is applied. This is where the reporting turns into strategy, because the pages that pull steady AI sessions are the pages answer engines already treat as a source worth citing. Those are the ones to deepen, update, and interlink first.
The native channel does not make manual filtering obsolete. It narrows the job to three cases.
Platforms the channel does not claim. Anything outside Google's recognized referrer list continues to arrive as an ordinary referral. A filter on session source is the only way to pull those visits out of the generic Referral bucket.
Historical continuity. Because the channel counts forward only, pre-rollout AI sessions are still sitting in Referral. A filter or custom channel group is evaluated at query time rather than at collection, so it does reach back across historical data and rebuild a comparable series.
Auditing the channel. Running a filter over your full source list and comparing the result against the AI Assistant row tells you how much AI referral traffic the native channel is leaving behind. That delta is worth knowing before you present the channel figure as complete.
.*(chatgpt\.com|openai\.com|perplexity\.ai|claude\.ai|anthropic\.com|gemini\.google\.com|copilot\.microsoft\.com|edgeservices|deepseek\.com|grok\.com|x\.ai|meta\.ai|mistral\.ai|you\.com).*
Apply it through Add filter in the Traffic acquisition report, with Session source / medium as the dimension and Matches regex as the match type. GA4 regex matching is case sensitive and the source values these platforms record are lowercase, so keep the pattern lowercase. Before trusting it, build an unfiltered source/medium table, find the exact values GA4 actually recorded for your property, and extend the pattern to match what you see rather than what you expect.
Default channel groups cannot be edited. You cannot add a platform to the native AI Assistant channel yourself, which means a custom channel group remains the only route to a permanent, shared channel that covers a platform list you control.
There is a second reason to keep one. Custom channel groups are applied at query time, which means yours works retroactively across your full history, unlike the native channel. The cost is worth stating plainly too: standard properties allow only two custom channel groups, with five on 360, so dedicating one to AI is a real tradeoff, and the regex inside it needs maintaining as platforms change domains. Editor or Administrator access is required to create it.
The build is short. Open Admin, then Channel groups under Data display. Copy the default group so you keep the standard channels, add a channel named Artificial intelligence, set the condition to Source matches regex, paste the pattern above, and order the new channel above Referral so AI sessions get claimed first. Once it processes, select your custom group from the dimension dropdown in any Acquisition report.
Clicks from Google's own AI features are the significant exception to everything above, and Google's channel definition excludes them explicitly rather than by omission. They do not appear in the AI Assistant channel. They arrive as ordinary organic google sessions, with no referrer detail separating an AI Overview click from a standard blue-link click, which means no regex will ever catch them.
Google has closed part of that gap from a different direction. On 3 June 2026 it launched Search Generative AI performance reports in Search Console, a dedicated view of impressions within generative AI features on Search, including AI Overviews and AI Mode, plus generative AI features in Discover. The reports break impressions down by page, country, device, and date, and they rolled out to a subset of sites first rather than to everyone at once.
Read the scope precisely before building a report on it. The view shows impressions, not clicks, click-through rate, or query data. That makes it a visibility signal rather than a traffic signal: useful for seeing which pages and markets surface inside AI answers, silent on what those appearances are worth. Pair it with your GA4 organic trend rather than treating either number as the whole story.
Everything above is worth doing, and it is still worth being honest about the ceiling. GA4 measures one thing: human clicks that arrived with attribution intact.
A large share of AI users never click a live link at all. They copy a URL out of the answer and paste it into a fresh tab, which carries no referrer. Mobile AI apps and privacy-focused browsers strip referrer headers for their own reasons. When the referrer is missing, GA4 has nothing to attribute the session to, so it defaults to Direct, and the native channel misses it exactly as the old regex did. A useful diagnostic is to watch for deep, specific pages suddenly earning direct sessions with high engagement.
The deeper limit is conceptual rather than technical. Referral traffic is the click stage of AI visibility, equivalent to clicks in traditional search. It is always smaller than the mention stage, equivalent to impressions, because most AI answers describe a brand without attaching a clickable link at all. No analytics tool can see how often an answer engine mentions you, cites you, or crawls your pages to consider you as a source, and that layer is where AI visibility is actually won or lost. Mapping it starts with a dedicated approach to tracking brand mentions across AI platforms and to measuring share of voice inside AI answers.
GA4 reports on sessions. HubSpot reports on contacts, deals, and revenue, which is where AI traffic finally proves its worth.
HubSpot includes a dedicated AI Referrals traffic source. Open Reports, then Marketing, then Traffic Analytics, and select the Sources tab. AI Referrals appears as its own row, covering visits from ChatGPT, Claude, Perplexity, Gemini, Grok, and other answer engines. Click the sessions number to drill into which platform sent each visit, then click a platform to see any associated campaign, which HubSpot reads from the utm_campaign parameter. Regular Google and Bing search stay under Organic search, and only genuine AI domains roll up into AI Referrals.
The CRM side is where attribution gets valuable. HubSpot stamps every contact with an Original Traffic Source and a Latest Traffic Source property, and an AI referral sets these to AI Referrals with drill-down properties recording the specific platform. Associated companies and deals mirror those values, so a deal inherits the AI source of its contact. From there you can build an active list or a custom report filtered on either property equal to AI Referrals, then layer deal stage or amount on top to measure what the traffic is worth. GA4 cannot reach this step, because it stops at the session and never touches the deal.
HubSpot has the same referrer problem. Visitors arriving through an in-app browser often land in direct traffic instead of AI Referrals, and ad blockers distort the count further. There is also a methodology gap: HubSpot credits the first-touch source while GA4 credits the last non-direct click, which is why the two tools rarely agree. To close it, add a short "how did you hear about us" dropdown to your key forms with an explicit AI option, and report on it beside the AI Referrals source. Self-reported answers catch the buyers who saw an AI mention, searched your brand, then converted with no trackable referrer at all.
Cognizo's Autopilot tier exists because measurement on its own does not move the number. Its agents run the full loop, from market research and prompt planning through content production, publishing, and lead attribution, so the gap a GA4 report exposes gets acted on rather than logged. That end-to-end execution is the primary reason teams adopt the platform rather than adding another dashboard.
Underneath it, the AI Traffic Analytics module captures the layer no client-side tool can reach: the server-side activity of AI crawlers on your site, broken down by intent and by provider. The module tracks visits from bots such as GPTBot, ClaudeBot, and OAI-SearchBot, then sorts them into AI Citations (an engine fetching a page to reference in a live answer), Training Visits, and Indexing Visits, each with its trend against the prior period. Every visit is attributed to a provider, a Top Providers list ranks them by volume, a time-series chart plots activity day by day, and a real-time log shows which pages the engines are pulling. Alongside that, a Humans Referred metric counts the people AI systems send you, by volume and provider.
The division of labor stays clean. HubSpot shows which AI-sourced contacts became deals, GA4 shows the session trend, Search Console shows impressions inside Google's AI features, and Cognizo shows the crawl and citation activity upstream that decides whether any engine can cite you in the first place. Upstream matters because it moves first: an engine has to fetch and cite your pages before a human referral, lead, or deal can follow. Hat Club used Cognizo to track its AI visibility, and although only about 1 in 50 of its visitors came from AI referral traffic, that traffic drove 20x growth in AI-driven revenue, a reminder that a low click count can sit on top of an outsized business effect.
Once the channel is measurable, the goal shifts from counting it to increasing it. Because answer engines lean heavily on off-site reputation and cleanly structured content, the levers are familiar but reweighted: earn credible third-party mentions, make pages easy for crawlers to reach and parse, and cover the questions buyers actually put to AI. Watch the mention and citation layer rather than referral clicks alone, since mentions move first and clicks follow. A structured playbook for that work lives in our guide on how to improve AI search visibility. Pair it with the reporting setup above, and a change in how often engines cite you can be lined up against the referral traffic that follows in GA4 and the pipeline your CRM records downstream.
Not necessarily on the same date. Google announced the channel on 13 May 2026 and rolled it out to properties gradually rather than all at once, so different properties began populating the AI Assistant row at different times. If a property shows an empty or very thin AI Assistant channel shortly after launch, check whether the rollout had reached it before concluding that AI is sending no traffic. The practical step is to note the first date your property recorded an ai-assistant session and treat that as the start of reliable data.
No. Sessions recorded before the channel reached your property remain classified as Referral, and Google has not announced any backfill. This has a direct reporting consequence: any comparison spanning the rollout date will show AI Assistant traffic appearing suddenly while Referral drops, which is reclassification rather than a behavior change. For a continuous series across the boundary, apply a source regex over the full date range instead of relying on the channel, and annotate the transition date in any dashboard stakeholders read.
No. Default channel groups in GA4 are fixed and cannot be edited, so the AI Assistant channel covers exactly the referrers Google chooses to recognize and nothing else. If a platform matters to your reporting and is not being claimed by the channel, your options are a temporary filter on session source for a one-off read, or a custom channel group with your own regex condition for a permanent one. Verify the behavior in your own data first, since the recognized list has changed since launch.
Treat it as reserved. The ai-assistant medium and the (ai-assistant) campaign name are populated automatically by GA4 when it matches a referrer, and manually assigning the same values through UTM tags on links you control would mix self-tagged traffic into a channel that is supposed to represent detected AI referrals. If you want to tag your own placements inside AI-adjacent contexts, use a distinct medium value so the two data sets stay separable in reporting.
Only where the app passes a referrer. Web clicks from chatgpt.com carry a referrer GA4 can match, so they classify cleanly. Traffic from native mobile apps and in-app browsers frequently arrives with the referrer stripped, in which case GA4 has nothing to match and the session lands in Direct rather than under any AI source. The result is that your AI Assistant row skews toward desktop and browser usage, and the app share of real AI traffic is systematically under-represented.
Not immediately. Keep it running in parallel for at least a full reporting cycle and compare its total against the native AI Assistant row. The difference tells you how much AI referral traffic the native channel is not claiming for your property, which is the only way to know whether the built-in figure is complete enough to report on its own. Given that GA4 allows only two custom channel groups, decide deliberately whether AI keeps one of those slots after you have seen the delta.
Review whenever new platforms appear in your data, and at minimum alongside each reporting cycle rather than on a fixed annual schedule. New answer engines launch regularly, existing ones change referring domains, and Google's recognized list has already been revised once. The habit that catches drift is simple: build an unfiltered session source/medium table, scan it for anything that looks like an AI platform, and add whatever is new to your regex and your custom channel group condition before it accumulates unattributed sessions.
Mostly because the two tools attribute sessions differently. HubSpot credits the first known source that brought a contact to your site, while GA4 credits the last non-direct click, so a visitor who arrived first from Google and later from ChatGPT counts as Google in HubSpot and ChatGPT in GA4. GA4 also detects some referrers that HubSpot's in-app handling misses. Neither total is wrong. Use HubSpot for contact and revenue attribution, and GA4 for overall session trends, rather than expecting the two counts to reconcile.
.svg)
.svg)
