AI Search Attribution: How to Measure AI Traffic in 2026

A customer in Tampines asks ChatGPT which accounting firms handle small company tax filing. Your firm gets named. She doesn't click. Two days later she googles your name and books a call from your site. GA4 records a branded organic visit. The AI answer that put you on her list gets nothing.

That gap is the whole problem with AI search attribution. Some AI traffic is easy to see now, more than it was a year ago. Most of the influence still isn't. This guide covers what GA4 and Search Console can show you in October 2026, how to set up a proper AI channel, and the cheap proxies that fill in the rest.

What is AI search attribution?

AI search attribution is the work of connecting leads and sales back to AI answers in ChatGPT, Perplexity, Gemini, Copilot and Google's AI Overviews. Part of it is tracking referral clicks in GA4. The bigger part is estimating influence that never sends a click at all, mostly from branded search and what customers tell you.

Sources: SE Ranking AI traffic study, 2026, Similarweb, June 2026, Pew Research Center, July 2025.

Read the first two numbers together. A third of one percent is small. Sixteen-fold in two years isn't. The same SE Ranking study found ChatGPT sent 74.78% of AI referrals in 2026, with Gemini at 11.56%, Perplexity at 7.23%, Copilot at 3.51% and Claude at 2.62%. If you sell to consumers or SMEs, ChatGPT and Gemini are where your measurable AI clicks will come from.

Why AI traffic is so hard to measure

Web analytics works on two signals. The browser passes a referrer that names the page the link sat on, and the link may carry UTM tags. AI products break both, in five ways.

The brand is named, not linked

This is the big one and no tool fixes it. An answer recommends you, and the person searches your name or types your URL later. There was never a referral to capture. Similarweb's June 2026 study tracked people who got a brand recommendation in ChatGPT and found they were 2.5 times more likely to visit that brand's site over the next seven days than people shown a competitor. For AI-influenced visits, search was the route for close to 56% of them, against about 40% for ordinary visits. Those people show up in your reports as branded organic traffic.

Apps drop the referrer

Every major assistant has a phone app now, and links opened from inside an app often arrive without a referrer. GA4 then calls the visit Direct. A UTM tag survives that trip, which is why OpenAI's tagging matters so much.

Copied and shared links

People copy a URL out of an answer and paste it into WhatsApp or an email to a colleague. Every one of those visits is Direct.

Cross-device and delayed visits

Someone reads an answer on the MRT and acts on it at their desk a day later. Your analytics sees the desk visit only.

Google's own AI surfaces look like organic search

Clicks from AI Overviews and AI Mode carry a google.com referrer. In GA4 they're Organic Search, same as a blue-link click. Google confirms that these appearances are counted in the normal Search Console Performance report, inside the Web search type. There is no filter to pull them out.

Put those together and every AI number in your dashboard is a floor. The real figure sits above it by an amount nobody has measured well, so label it that way in every report.

Last-click attribution misses most of AI's effect

GA4 credits the session's source, and its attribution models then share credit across the touchpoints it saw. The AI answer that started the journey is usually a touchpoint it never saw. A data-driven model can't give credit to an event that isn't in the data, so switching models won't fix this.

That cuts both ways. Don't claim every rise in branded search for AI either. Similarweb's own write-up shows an association between recommendations and later visits, not proof that each visit was caused by one. The honest position is a range: a measured floor from referrals, and an estimated upper band from proxies.

If you have the traffic to run it, an experiment beats any model. Pick a group of pages, improve them for AI answers and leave a similar group alone. Compare leads from each group over eight to twelve weeks. Most SMEs don't have the volume for a clean test, which is why the proxy approach later in this guide exists.

What GA4 and Search Console can see in 2026

Two changes this year made the picture better than most older guides describe.

GA4 now has an AI Assistant channel

Google added an AI Assistant channel to the default channel group in May 2026. Google's channel definitions describe it as traffic from "sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok". When the referrer matches Google's list, GA4 sets the medium to ai-assistant and the campaign to (ai-assistant), and the same page states that AI Overviews and AI Mode are excluded.

It's a decent start, but Google's description doesn't name Perplexity or Claude, and Search Engine Journal's testing said Perplexity visits still land in Referral. The list is Google's to change, and you can't see it. Check where Perplexity and Claude sessions sit in your own property before trusting the default.

Search Console has a generative AI report

On 3 June 2026 Google launched generative AI performance reports in Search Console for Search and Discover. The post says they were rolled out to all websites worldwide as of 31 August 2026. They count how often your pages appeared in AI Overviews and AI Mode, split by page, country, device and date.

What they don't show matters as much. There are no clicks, no click-through rate and no queries in that report. AI Overview clicks are in the main Performance report, mixed in with everything else. So you can see how often Google's AI features showed your pages, and you can't see what that exposure sent you.

For ChatGPT, the tagging is documented. OpenAI's publisher FAQ says:

"ChatGPT automatically includes the UTM parameter utm_source=chatgpt.com in referral URLs, enabling clear tracking and analysis of inbound traffic from ChatGPT search results."

OpenAI, Publishers and developers FAQ

No other AI vendor we checked publishes a tagging rule. Perplexity, Claude and the rest are tracked by referrer patterns people have spotted in their own data, and those can change without notice.

What you can and can't see, surface by surface

This is the table we use when a client asks "how much do we get from AI?". Each row says where the traffic lands today and what to watch instead when the data runs out.

Two things fall out of that table. The surface most Singapore searchers see, Google's AI Overviews, is the one you can measure least. And the one you measure best, ChatGPT, is measurable because OpenAI chose to tag its links. If you want the wider context on how many searches now show an AI Overview, our Google AI Overviews statistics roundup collects the figures.

How to set up an AI assistants channel group in GA4

The default AI Assistant channel misses sources, and you can't edit the default group. So build your own. It takes about fifteen minutes and custom channel groups apply to past data, so you get history the day you save it.

1. In GA4, go to Admin and open Channel groups under Data display.
2. Open the Default channel group and choose to copy it, so you keep every Google rule. Name the copy "Default plus AI".
3. Click Add new channel and name it "AI assistants".
4. Add a condition group: Source, matches regex, and paste the pattern below.
5. Add a second condition with OR: Medium, exactly matches, ai-assistant. This keeps whatever Google's own list catches.
6. Delete or rename Google's built-in AI Assistant channel in your copy so traffic doesn't split between two AI rows.
7. Drag AI assistants above Referral and above Organic Search. GA4 puts a session in the first channel it matches, so order decides everything.
8. Save the group, then open the Traffic acquisition report and switch the channel dropdown to your new group.

The regex, as of October 2026:

.*(chatgpt|openai\.com|perplexity|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai|deepseek\.com|grok\.com|meta\.ai|mistral\.ai).*

Why each piece is there: chatgpt catches both the chatgpt.com referrer and a bare utm_source=chatgpt tag. openai.com picks up older chat.openai.com links. The other entries are the hostnames people have reported for the remaining assistants, Mistral's Le Chat included. The leading and trailing .* mean it works whether GA4 treats the rule as a full or partial match.

Standard GA4 properties allow two custom channel groups, so don't waste one on a test. Once a quarter, open a Referral report filtered to sources you don't recognise and add any new AI hostname you see. If you already run GA4 for SEO reporting, our Google Analytics for SEO guide covers the reports this channel will sit beside.

Tie AI visits to leads, not just sessions

A channel that shows sessions is half the job. Mark your enquiry form submits and WhatsApp clicks as key events in GA4 so the AI assistants row carries a conversion count. Then compare the AI row's key event rate with organic search each month.

Expect AI visitors to behave differently. Similarweb found AI-influenced visitors looked at nearly twice as many pages and stayed about twice as long. Small samples swing wildly, though. With 40 AI sessions a month, one keen visitor can double your rate. Read quarterly totals, not weekly ones.

Check landing pages too. A lot of AI clicks land on the homepage because answers recommend brands, not articles. That makes page-level reports less useful for working out which content earned the mention, and it's one more reason the prompt checks below matter.

Measure the influence that never clicks

This is where most of AI's effect lives, and where the competitor guides go vague. Four signals, from cheapest to most work.

Branded search in Search Console

Filter the Performance report to queries containing your brand name and its common misspellings. Track impressions and clicks monthly. Say your AI visibility work started in July and branded impressions climbed from August while ad spend stayed flat. That supports the case without settling it, so log anything else that changed that month and rule it out before you report the link.

Direct traffic to deep pages

People rarely type a long URL by hand. A rise in Direct sessions landing on deep service pages, not the homepage, often means pasted links from somewhere, and AI answers are one likely source.

Prompt tracking

Write 20 to 30 prompts your buyers would ask, such as "best SEO agency for a Singapore F&B brand" or "how much does a corporate secretary cost in Singapore". Run them monthly in ChatGPT, Gemini, Perplexity and Google AI Mode. Record whether you're named and who gets named instead. That gives you a share-of-answer number you can plot against branded search.

Server logs

Your server sees machines that analytics never does. OpenAI documents its crawlers separately, with different jobs: GPTBot for training, OAI-SearchBot for ChatGPT's search index and ChatGPT-User for fetches a person triggers during a conversation. A ChatGPT-User hit means someone's question caused that exact page to be fetched. It doesn't prove a citation, but it's the closest thing to a live signal you'll get.

A small-budget setup for Singapore SMEs

You don't need a paid AI visibility platform to start. Here's the version we'd run for a business with one marketer and no analyst.

The self-reported field does the most work for the least money. Make it a dropdown with ChatGPT or another AI assistant as its own option next to Google search, social media and Other. People rarely type "ChatGPT" into a blank box, and they'll pick it from a list. Send the answer to your CRM next to the GA4 source. Where the two disagree, the form is usually telling you about a visit GA4 filed as Direct or branded search.

One local note. The answer sits on a lead record with a name and phone number, so it falls under the Personal Data Protection Act. Your form's privacy notice should already say you use enquiry details to understand how people found you. If it doesn't, add a line before you add the field.

Put the five numbers on one page each month: AI sessions and leads from GA4, self-reported AI leads, share of answer from your prompt check, and the two Search Console figures. Show the floor and the proxies side by side. If you want a fuller framework for SEO reporting, our guide to measuring SEO success covers the core metrics.

How this connects to GEO

Attribution tells you whether AI visibility is paying off. Getting the visibility is a separate job, usually called generative engine optimisation. If you're new to it, start with our explainer on GEO. Then set up the measurement in this guide before you change a single page. Without a baseline, you'll never know which change worked.

Frequently asked questions

Can I see Google AI Overviews traffic in GA4?

No. Clicks from AI Overviews and AI Mode arrive with a google.com referrer, so GA4 files them under Organic Search with every other Google click. Google's own channel definitions say the AI Assistant channel excludes them. Search Console's generative AI report shows impressions for those surfaces, but not clicks.

Does GA4 track ChatGPT traffic automatically now?

Partly. Since mid-2026 GA4's default channel group has an AI Assistant channel that catches referrals from ChatGPT, Gemini, DeepSeek, Copilot and Grok. Visits opened from apps without a referrer, or from a brand name someone typed in later, still land in Direct or Organic Search.

Why does so much AI traffic show up as direct?

Mostly because AI answers name brands far more often than they send clicks, so people type the domain or search the name later on. Links opened inside native apps often lose the referrer too, and a URL copied into WhatsApp arrives with no source.

What is a good proxy for AI visibility if I can't track clicks?

Branded search impressions in Search Console are the easiest to get. Pair them with a monthly hand-run prompt check and a 'How did you hear about us?' dropdown on your forms that includes an AI option. A single proxy can mislead you, but when all three rise in the same months you have a case worth taking to the boss.

How much traffic does AI search actually send?

Not much yet, but it's growing fast. SE Ranking's study of over 100,000 sites put AI referrals at 0.32% of all website traffic in 2026, up from 0.02% in 2024. That figure excludes AI Overviews and AI Mode, which can't be separated from Google organic.

Get your AI search attribution set up properly

We build AI channels in GA4, set up the self-reported field and run monthly prompt checks for Singapore businesses, with the floor and the proxies on one monthly page. See how our AI SEO service works, or ask our analytics team to check what your current property misses. Book a call and we'll look at your GA4 with you.