AI Summary
If your GA4 still shows ChatGPT and Perplexity traffic bucketed under “Direct” or “Referral”, you are missing 30 to 40 percent of the picture of how prospects find you. The good news is Google quietly shipped a native fix in late 2025. The better news is the manual setup that came before it still has a place in 2026. Here is everything you need.
The scale of the gap matters. Conductor’s November 2025 attribution study found that 89 percent of brands measured cannot properly attribute their AI referral traffic, and Trakkr.ai’s aggregated benchmark across 4,025+ GA4 properties shows the median property under-counts AI traffic by 30 to 40 percent when relying on default channel grouping. With AI-sourced sessions running at 4.4 times the conversion rate of Google organic according to Adobe, this is not a measurement nicety. It is a budget-allocation problem.
What changed in late 2025: the native AI Assistant channel
Google introduced a default channel group called “AI Assistant” inside GA4. Sessions get routed to this channel automatically when the source matches a recognized AI platform list (ChatGPT, Perplexity, Gemini, Claude, Copilot, and a growing roster). The technical signal Google uses is medium=ai-assistant and campaign=(ai-assistant), which the platform applies in the background when source recognition fires.
If you have a fresh GA4 property created in late 2025 or 2026, you may already see this channel populated without doing anything. Check your Acquisition reports today. If you see a row called “AI Assistant” with traffic numbers, the new behavior is live on your property.
Three things to know about the native channel:
- It is not retroactive. Historical traffic that landed before the platform recognized a given AI source will still sit in Direct or Referral.
- Source list is incomplete. Google’s recognition list covers the major surfaces but lags on newer entrants like phind, Pi, character.ai, and dozens of vertical-specific AI assistants.
- You still need a custom channel group for cross-platform reporting. Many properties run multiple GA4 streams, and the native channel does not back-propagate consistently.
The manual setup that still earns its keep
The custom channel group approach popularized by Dana DiTomaso and the Kickpoint team in early 2025 remains the most thorough solution. It is the right setup for any property that wants comprehensive AI source coverage today, and it gives you a documented spec you control rather than depending on Google’s recognition list.
Step-by-step: create the custom channel
- Open GA4 Admin, navigate to Data Display in the property column, then Channel Groups.
- Click “Create new channel group”. Name it something explicit like “AI-Aware Channels 2026”.
- Inside the new group, click “Add channel”. Name it “AI Traffic”.
- Set the rule: Source matches regex (case-insensitive), paste the regex below.
- Critical step: drag the new AI Traffic channel above the Referral channel. GA4 evaluates rules in order, and if Referral fires first, your AI traffic stays misclassified.
- Save the channel group. New sessions begin routing immediately. Existing sessions in historical reports will not retroactively reclassify, but new reports referencing this channel group will use the new rules.
The current 2026 AI source regex
This is the working pattern most teams are using, updated through Q1 2026:
^(?:chatgpt\.com|chat-gpt\.org|claude\.ai|quillbot\.com|openai\.com|blackbox\.ai|perplexity(?:\.ai)?|copy\.ai|jasper\.ai|copilot\.microsoft\.com|gemini\.google\.com|(?:\w+\.)?mistral\.ai|deepseek\.com|edgepilot|edgeservices|nimble\.ai|iask\.ai|aitastic\.app|bnngpt\.com|writesonic\.com|exa\.ai|cohere\.ai|huggingface\.co|anthropic\.com|poe\.com|pi\.ai|phind\.com|character\.ai|you\.com|meta\.ai|grok\.x\.ai|grok\.com)Review this list quarterly. New AI assistants launch monthly, and the regex needs maintenance to stay accurate. Trakkr.ai publishes an updated regex daily based on the sources their crawler observes across the 4,025+ properties they monitor.
The market share you need to know
Allocating attention to AI sources requires knowing which sources are sending traffic at meaningful volume. SE Ranking’s late-2025 measurement across observed AI referral traffic put the share at:
- ChatGPT: 77.97 percent
- Perplexity: 15.10 percent
- Gemini: 6.40 percent
- All other AI sources combined: under 1 percent
Trakkr.ai’s measurement of AI source share is slightly different because their sample includes business-tooling AI assistants Trakkr classifies separately. Their numbers put ChatGPT at 60 to 70 percent, Perplexity in the 12 to 18 percent range, and Gemini growing fastest in Q4 2025. Either dataset agrees on the general shape: ChatGPT dominates, Perplexity is a strong second, Gemini is a growing third, and everything else is rounding error.
The takeaway for your tracking setup: if you can only afford to track three sources well, track those three. If you want to be comprehensive, use the full regex above.
ChatGPT’s outbound link behavior
OpenAI added utm_source=chatgpt.com to outbound links from ChatGPT in 2024. This means any session arriving from a ChatGPT click should already have utm_source populated correctly in GA4 even before the channel group resolves it. If you are seeing ChatGPT traffic without utm parameters, the source is likely a different ChatGPT surface (mobile app on certain platforms, or a custom GPT) and you will need the source-match channel group to catch it.
Perplexity and Claude also pass referrer information consistently. Gemini is the trickier case: when Gemini surfaces inside Google’s own search interface (AI Overviews, AI Mode), it does not behave like an outbound click the way standalone gemini.google.com does. AI Overview clicks are classified as standard Google organic in GA4, not as AI Assistant traffic, which is one reason organic search volume can look healthier than the underlying business reality.
Why AI traffic deserves different reporting
Beyond classification, the behavior of AI-sourced sessions is different enough that you should report on it separately rather than blending it into your organic KPIs.
- Conversion rate. Adobe Analytics measured AI-sourced traffic converting at 4.4 times the Google organic rate in late 2024 and the gap held through 2025.
- Session depth. Sessions run 68 percent longer with 3 times more pages per session on average.
- First-session conversion behavior. Conductor’s November 2025 study found 73 percent of AI referral conversions happen in the first session, vs 23 percent for Google organic. AI traffic is bottom-funnel by the time it arrives.
- Revenue per visit. Adobe’s tracking showed AI-sourced revenue per visit growing from 3 percent of non-AI in July 2024 to 70 percent of non-AI by late 2025. The trajectory is steep.
- Year-over-year growth. The same Adobe data shows 500 percent plus YoY growth in AI referral volume across measured properties.
If you are reporting AI sessions blended into a single “organic” bucket, you are letting a 4.4x conversion rate hide inside an average. Splitting the channel exposes which growth channels are actually driving your funnel.
What to do with the data once you can see it
Step 1: build an AI-source revenue cohort
In GA4 Explore, create a free-form report with AI Traffic as the channel filter, then bring in conversion events, revenue, and session quality metrics. Compare month-over-month. The first month of clean data will probably look noisy. By month three you will have a stable baseline.
Step 2: identify which pages AI is sending traffic to
Most teams discover that AI sources send disproportionate traffic to a small number of pages: a definitional glossary entry, a comparison post, a pricing page, or an original research piece. Those pages are your citation magnets and they deserve disproportionate optimization attention.
Step 3: tune your content calendar around AI citation behavior
Once you can see which pages drive AI traffic, you can build more pages with the same structural properties: clear headings, explicit definitions, comparison tables, statistics with sources, and unambiguous attribution language. The intersection of AEO, GEO, and SEO best practices all reinforce the same structural patterns.
Trakkr.ai and other free benchmarks
If you want category benchmarks without paying for a tool, Trakkr.ai publishes free AI referral benchmarks across 4,025+ GA4 properties, updated daily at 00:00 UTC. The dashboard shows median AI referral share, top-growing AI sources, and citation source breakdowns segmented by industry. It is the easiest way to know whether your AI traffic numbers look reasonable for your category.
For competitive intelligence on AI citation share rather than referral traffic, Profound and Bluefish offer enterprise dashboards in the $499 to several thousand per month range. Scrunch and Adobe LLM Optimizer round out the enterprise category.
Common mistakes that break your data
- Leaving the AI Traffic channel below Referral. Order matters in GA4 channel grouping. AI Traffic must be above Referral or your data stays misclassified.
- Using case-sensitive regex. The “case-insensitive” flag matters because some browsers and intermediaries normalize URLs differently. Always use the case-insensitive matching option.
- Forgetting to update the regex. The 2024 version of this regex is missing roughly 30 percent of the sources active in 2026. Quarterly review is the minimum.
- Reporting AI traffic in aggregate with organic. The behavior is different enough that aggregate reporting hides the signal.
- Not tracking branded organic uplift. AI exposure drives downstream branded search. If you do not segment branded vs non-branded organic, you miss the secondary effect.
Where this is heading
Google’s native AI Assistant channel will keep expanding its source recognition through 2026, which means the custom regex setup becomes less necessary over time. By late 2026 we expect the native channel to cover 90 percent of relevant sources without manual intervention. Until then, run both: let the native channel handle what it handles, and use the custom group as a safety net for everything else.
The teams that get this measurement layer right in Q1 2026 will have six months of clean attribution data by the time annual planning conversations start. The teams that defer the setup will be making budget decisions in 2027 based on a 30 to 40 percent under-count of one of their highest-converting channels.
For more on how to think about the metrics this data unlocks, see the companion piece on AI search analytics that actually matter. For the budget conversation that should follow once you can see your AI traffic clearly, see our GEO vs SEO 2026 budget split framework.