Key highlights
- Visitors sent by AI assistants such as ChatGPT, Perplexity, Gemini, Copilot and Claude are usually split across the Referral, Direct and Unassigned channels in GA4, so most brands undercount them.
- A GA4 custom channel group that matches session source against a regex of AI domains, placed above Referral, gives you one consistent AI line that also applies to historical data.
- Analytics only sees clicks, so pair GA4 with first-touch source capture in your CRM and a “how did you hear about us” question that lists AI assistants as an option.
- On abovethefold.in, visitors from chatgpt.com recorded a 4.98% key event rate against 1.24% for direct traffic, which is roughly four times higher.
If ChatGPT names you when someone asks for a social media agency in Mumbai or the best cookware brand in India, some of those people will click through. Most Indian brands have no idea how many, because GA4 does not show AI visitors in one clean line. Here is how to find them, group them, and catch the AI influence that never shows up as a referral at all.
Where does ChatGPT traffic show up in GA4?
ChatGPT traffic usually appears in GA4 with the session source chatgpt.com, but it is scattered across the Referral, Direct, Unassigned and, on newer properties, AI Assistant channels. Unless you group it yourself, you will undercount it and misread its trend.
- Referral by default. When someone clicks a link in a ChatGPT, Perplexity or Copilot answer inside a browser, the browser passes the referring domain. GA4 files it as chatgpt.com / referral, alongside every directory that links to you.
- Direct when the referrer is stripped. Links opened from mobile apps and in-app browsers often arrive with no referrer, so the visit lands in Direct. If your audience mostly uses the ChatGPT app on a phone, this is likely your larger leak.
- Unassigned when the tagging is half there. ChatGPT adds utm_source=chatgpt.com to many of the links it shows. When the source survives but the medium does not, GA4 cannot match a default channel rule and the session sits in Unassigned.
- The built-in AI Assistant channel is partial. According to Search Engine Journal, Google added an AI Assistant channel to the GA4 default channel group in May 2026. It helps, but its list of recognised assistants has changed since launch, it did not include Perplexity when SEJ reviewed it, and it only counts forward from the date it switched on for your property.
Clicks from Google’s AI Overviews and AI Mode arrive as ordinary google / organic traffic, which GA4 cannot split out. For those, see step 5.
Step 1: Build a custom channel group for AI sources
The cleanest fix is a custom channel group with an “AI assistants” channel that matches on session source only, placed above Referral, Organic Search and the built-in AI Assistant channel. Channel groups apply to historical data, so you get a consistent trend line back to the start of your property.
- In GA4, go to Admin, then Data display, then Channel groups.
- Click Create new channel group. GA4 starts you with a copy of the default group. Name it something like “Default plus AI”.
- Click Add new channel, name it “AI assistants”, and add a single condition: Source, matches regex, using the pattern below.
- Click Reorder and move “AI assistants” to the top, above AI Assistant, Referral and Organic Search. GA4 assigns each session to the first rule it matches, so the order decides everything.
- Save the group. Then open Reports, Acquisition, Traffic acquisition, switch the primary dimension to your new channel group, and add Session source as a secondary dimension to see each assistant separately.
A starting regex for the source condition:
^(.*.)?(chatgpt.com|chat.openai.com|perplexity.ai|gemini.google.com|copilot.microsoft.com|claude.ai|deepseek.com|chat.deepseek.com|grok.com|meta.ai|chat.mistral.ai)$
The (.*.)? at the start catches subdomains such as www.perplexity.ai, and anchoring both ends stops it catching unrelated domains that contain “gpt” or “google”. Never add google.com or bing.com, which would sweep ordinary search into your AI line. Check your Session source report each quarter for new AI domains, because this list changes quickly.
For analysis without changing the channel setup, build a free-form Exploration: Session source and Landing page as rows, a filter where Session source matches the same regex, and Sessions, Average engagement time, Key events and Session key event rate as metrics. Use the channel group for reporting and the exploration for digging into which pages AI engines favour.
Step 2: Track the actions that matter for Indian buyers
AI traffic is only worth measuring if you can see what it does, and for most Indian businesses that means WhatsApp clicks, phone calls and form submissions rather than page views. Mark each of these as a key event in GA4 so every channel, AI included, is judged on enquiries.
| Action | How to capture it | Suggested event name |
|---|---|---|
| WhatsApp click | Clicks on any link containing wa.me or api.whatsapp.com, through Google Tag Manager or a GA4 “create event” rule on the outbound click event | whatsapp_click |
| Call click | Clicks on any tel: link, through Google Tag Manager | call_click |
| Email click | Clicks on any mailto: link, through Google Tag Manager | email_click |
| Form submit | Fire only on a confirmed submission (thank-you state or success response), never on the button click alone | generate_lead |
| Purchase (ecommerce) | Native Shopify or WooCommerce GA4 integration, checked against real orders every month | purchase |
Mark each one as a key event under Admin, Events. Session key event rate then becomes the most useful number for comparing AI visitors with everyone else. For the WhatsApp side of the funnel in more depth, see our WhatsApp marketing guide for Indian brands.
Step 3: Capture first-touch source in your CRM
GA4 tells you about sessions; your CRM tells you about customers. Storing the first-touch source on every lead record is the only reliable way to connect a ChatGPT visit in March to a signed client in May.
A developer can set this up in a day:
- On a visitor’s first page view, store the referrer, any UTM parameters and the landing page in first-party storage, such as a cookie or localStorage.
- On later visits, leave that first-touch record untouched and store a separate last-touch record alongside it.
- Pass both into hidden fields on every form, so they arrive in the CRM attached to the lead.
- For WhatsApp, log each click server-side with the stored source, or add a short reference code to the prefilled message, so the chat can be matched to its source when it reaches your team.
We set this up on abovethefold.in in late September 2026. In its first month it recorded WhatsApp enquiries whose first touch was ChatGPT, each landing on our social media marketing service page. GA4 would show those as a few referral sessions; the lead tracker shows named enquiries we can follow to an outcome.
Step 4: Ask people how they found you
Asking “how did you hear about us?” catches the AI influence that analytics cannot, because many AI-driven buyers never click at all. They read the answer, then search your brand name, call you, or message you on WhatsApp.
Add it to your enquiry form as a dropdown, with “ChatGPT or another AI assistant” as its own option, separate from “Google search”. Train whoever answers calls and WhatsApp to ask it in the first conversation and log the answer. Self-reported data is messy, but it is the only source that sees this part of the journey.
Step 5: Use Search Console’s generative AI report for Google’s AI features
For AI Overviews and AI Mode, the right tool is Search Console, not GA4. Google announced Search Generative AI performance reports in Search Console on 3 June 2026, showing impressions, pages, countries, devices and dates for appearances in its AI features.
According to Google’s announcement and coverage by PPC Land, the reports rolled out to a subset of websites first and show impressions but not clicks, so treat them as a visibility measure. If your property does not have the report yet, watch the regular Performance report for long, conversational queries with impressions and almost no clicks. These are often AI fan-out queries, the searches an assistant runs while assembling an answer.
What good looks like: a worked example from our own site
A healthy setup gives you one consistent AI line in GA4, a key event rate you can compare across channels, and CRM records showing which enquiries started with an AI answer. Here is our own GA4 data for abovethefold.in, January 2025 to October 2026.
| Traffic source | Key event rate |
|---|---|
| chatgpt.com | 4.98% |
| 2% | |
| Bing | 1.82% |
| Direct | 1.24% |
Our key events are call clicks, WhatsApp clicks, email clicks and form submissions. ChatGPT visitors converted at roughly four times the rate of direct visitors. They also spent 1m 23s on the site on average against 21s for direct, and triggered 7.78 events per session against 4.91.
The volume was small: 201 sessions from chatgpt.com over the whole period, producing 20 key events. That is the honest shape of AI referral traffic for most Indian service businesses: low volume, high intent, and easy to miss inside Referral.
The other lesson is volatility: monthly ChatGPT sessions peaked at 31 in May 2026 and fell to 2 in July, so read the trend by quarter, not month. We have written up the full story, including what we think caused the drop and what we changed, in our AI visibility case study.
Common mistakes when measuring AI traffic
Most errors come from judging a small, high-intent channel by the rules of a large one.
- Judging AI traffic on volume. A few dozen sessions a month can still carry more enquiries than channels ten times the size. Compare key event rates, not sessions.
- Relying on the built-in AI Assistant channel alone. It misses some assistants, misses referrer-stripped visits, and has no history before its rollout date on your property.
- Ignoring branded search. Rising clicks on your brand name in Search Console are often the downstream trace of AI recommendations.
- Measuring without acting on it. Getting recommended more often is a content and third-party presence job, which our guide on how to get recommended by ChatGPT and AI Overviews covers.
Does ChatGPT traffic show up in Google Analytics 4?
Yes. Visits from ChatGPT appear in GA4 with the session source chatgpt.com, but they are usually split across the Referral, Unassigned, Direct and AI Assistant channels. Visits from the ChatGPT mobile app often arrive without a referrer and land in Direct. Building a custom channel group that matches AI domains on session source, placed above Referral, brings the visible share together in one line.
Why does my ChatGPT traffic show as direct in GA4?
Traffic shows as Direct when the browser passes no referrer. This happens often with links opened from mobile apps, in-app browsers and some privacy settings. GA4 then has nothing to attribute the visit to. You cannot recover these sessions in GA4, which is why first-touch capture in your CRM and a “how did you hear about us” question are needed alongside analytics.
What regex should I use to track AI referrals in GA4?
Use an anchored pattern that matches known AI domains on session source, for example one covering chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, claude.ai, deepseek.com and grok.com, with an optional subdomain prefix to catch www variants. Avoid broad terms such as “gpt” or “google”. Review your source report every quarter, because new assistants and referrer domains appear regularly.
Can I see clicks from Google AI Overviews in GA4?
No. Clicks from AI Overviews and AI Mode arrive in GA4 as ordinary google organic traffic and cannot be separated. Google’s Search Generative AI performance reports in Search Console, announced in June 2026, show impressions, pages, countries and devices for AI feature appearances, but not clicks, and were rolled out to a subset of sites first.
Is AI referral traffic worth tracking if the numbers are small?
Usually, yes. On abovethefold.in, ChatGPT sent only 201 sessions between January 2025 and October 2026, yet those visitors converted at 4.98% against 1.24% for direct traffic. Small volume with high intent can matter more than a large channel that rarely enquires. Tracking also shows which pages AI engines favour, which tells you where to invest.
If you want this measurement set up properly, or want to find out whether AI engines are recommending your brand at all, talk to our team. Our SEO and GEO service covers both sides, and our GEO vs SEO explainer sets out how the two fit together.