The short version: Google fixed part of this in May 2026. GA4 now has a native AI Assistant channel that catches chatbot clicks carrying a referrer. It does not catch the ones that arrive with no referrer at all, it does not cover AI Overviews, and it does not backfill your history. So the gap between what your dashboard reports and what your buyers tell your sales team is narrower than it was a year ago, but it has not closed, and the part that remains is the part that is structurally hardest to fix.
If you are reallocating budget off channel reports right now, that residual gap is where the bad decisions come from.
What actually changed on 13 May 2026
For most of 2025, tracking AI traffic in GA4 was a manual job. Google published guidance in August 2025 on building custom channel groups with regex to capture assistant traffic, naming ChatGPT, Gemini, Microsoft Copilot, Claude and Perplexity as the platforms worth watching. The workaround was fiddly. Regex needed maintaining as domains changed, you needed editor access, and GA4's two-custom-channel-group limit meant burning one of two slots on it.
Then, as Search Engine Journal reported on 14 May 2026, Google added an AI Assistant entry to the Default Channel Group. When GA4 sees a referrer matching a recognised assistant, it sets the medium to ai-assistant, files the session under the AI Assistant channel, and stamps the campaign (ai-assistant). No configuration.
Two things about that update deserve more scepticism than they've been getting.
Google named ChatGPT, Gemini and Claude as examples and has not published the full recognised-referrer list. Do not assume Perplexity or Copilot are covered until you have confirmed it in your own property. Keep your custom channel group running alongside the native one; it costs nothing and it catches what Google's list misses.
And the channel counts forward only. Your pre-May-2026 AI traffic stays filed wherever it originally landed, which makes native year-over-year comparison for this channel impossible for now. If you present a "look how AI traffic grew" chart built on the default grouping, you are presenting a rollout artefact.
The three ways AI traffic still lands in Direct
No referrer header. Clicks originating in native desktop and mobile apps frequently pass no referrer. GA4 has nothing to classify, so the session defaults to Direct. This is an operating-system and app-architecture behaviour, not something Google or OpenAI can toggle off, which is why it will persist after the referrer-based improvements keep landing.
Copy-paste and cross-device. Someone reads an AI answer on their phone, mails themselves the link, opens it on a laptop. Two sessions, one journey, zero referrers. Common in B2B, where research and purchasing happen on different machines.
Consent rejection. On sites running compliant EU banners, a meaningful share of visitors decline tracking, and those sessions never reach GA4 in the first place. This compounds with the referrer problem rather than overlapping with it.
Every published estimate of the referrer-less share comes from a vendor selling AI-traffic detection, measured on its own customer base, so treat the specific percentages as directional at best. The honest framing for a board deck is that your measured AI number is a floor, not a total, and you don't know the multiplier.
AI Overviews are the bigger hole, and nobody's fixing it
Clicks from Google AI Overviews and AI Mode carry the same referrer as an ordinary Google organic click. They land in Organic Search with no separate label. GA4 cannot distinguish a click on a blue link from a click on a citation inside an AI-generated summary, and neither can Search Console, which blends AI Mode data into existing performance totals.
This matters more than the chatbot gap for most sites, because AI Overviews reach far more queries than any standalone assistant does. The channel that your CFO is most likely to underfund is the one your analytics is structurally incapable of isolating.
What the wrong numbers actually cost you
Three failure modes, in rough order of how often they show up:
You cut the content that's working. A page loses sessions to zero-click behaviour while quietly becoming a citation source that feeds high-intent visits arriving as Direct. On a traffic-weighted content audit, it looks like decay. It gets pruned or deprioritised, and the citation disappears with it.
You over-credit brand. Rising Direct traffic reads as brand strength, which justifies more brand spend. Some of that lift may be assistant clicks with no referrer. You end up funding the wrong thing with evidence borrowed from a channel you never measured.
You underprice a channel using a benchmark that doesn't apply to you. This one is subtler and worth spending a paragraph on, because the conversion-premium numbers circulating are being repeated far past what they support.
Ahrefs published data in June 2025 showing AI search traffic at 0.5% of its visits but 12.1% of its signups, roughly a 23x conversion advantage. That figure gets quoted constantly. It is real, and it is a single site: Ahrefs measuring Ahrefs, an SEO tool whose audience is unusually likely to be researching via AI assistants. Semrush's 2025 study of more than 500 digital marketing and SEO topics put the average AI search visitor at 4.4x the value of a traditional organic visitor, which is a much broader sample but still confined to one vertical. And Ahrefs' own analysis of roughly 35,000 websites found LLMs sending about 0.1% of total referral traffic overall, with Google sending some 345x more than the three main assistants combined.
Set those side by side and the range is enormous: somewhere between a marginal lift and a 23x premium, depending entirely on your category and audience. Anyone quoting a single conversion multiplier for AI traffic without naming the vertical it came from is selling you something. Your own number is the only one worth budgeting against, which is exactly why closing the measurement gap comes before the business case, not after it.
A four-step reclaim framework
1. Establish the floor. Confirm the native AI Assistant channel is populating in your property, and keep a custom channel group running in parallel with the regex patterns from Google's August 2025 guidance. Compare the two weekly. Where the custom group finds sessions the native channel misses, you've found a platform Google isn't recognising yet.
2. Fingerprint the dark traffic. Referrer-less AI sessions are not random. They cluster on deep informational pages rather than the homepage, they skew toward URLs an assistant would plausibly cite, and their engagement profile looks nothing like someone typing your domain from memory. Segment Direct traffic by landing page depth. Direct sessions landing on a 2,000-word comparison post are almost certainly not brand recall.
3. Ask, but ask properly. Add a discovery field to your high-intent forms with a named ChatGPT or AI assistant option, and a short free-text follow-up asking what they searched for. That second field is where the actual intelligence lives. Handle this data carefully: Ruler Analytics found in 2022 that 47% of leads selecting the first option in a dropdown were inaccurately attributed, because people pick the top item to move on. Randomise the option order, and treat self-reported attribution as a directional cross-check on your analytics, not a replacement for it.
4. Correlate rather than attribute. You will never get session-level attribution for an AI Overview click. What you can do is track your citation rate across engines over time and compare it against branded search volume in Search Console and Direct traffic volume. When citations rise and branded search rises with them, you have a defensible correlation. It isn't attribution. It is enough to defend a budget line, and it is more honest than a number invented to look like attribution.
FAQ
Does GA4 now track AI traffic automatically?
Partially. Since 13 May 2026, GA4 files traffic from recognised AI assistants under an AI Assistant channel with no setup required. Sessions arriving without a referrer still land in Direct, and AI Overviews clicks are still counted as Organic Search.
Can I see my historical AI traffic in the new channel?
No. The channel applies going forward from the date it reached your property. Earlier AI traffic stays in Referral or Direct.
Is Perplexity included?
Google named ChatGPT, Gemini and Claude as examples and has not published the full list. Verify against your own data before assuming coverage, and keep a custom channel group as a backstop.
How much of my Direct traffic is actually AI?
No credible cross-industry benchmark exists. Every published figure comes from a vendor measuring its own customers. Segment your own Direct traffic by landing page instead of applying someone else's percentage.
Do AI visitors really convert better?
Directionally yes, across every study that has looked, but the size of the advantage varies from marginal to 23x depending on vertical and audience. Measure your own.
RankSage joins GA4, Search Console, first-party behavioural data and AI citation data at the page level, then ranks the actions worth taking. The reclaim framework above is the manual version of what the platform automates. We're pre-launch and building in the open, so there are no customer results to show you yet, only the method. Join the waitlist if you want early access when it opens.
Top comments (0)