How to Measure Traffic From ChatGPT and AI

The Short Answer: You cannot measure AI search traffic from one report, because it arrives through three doors. Clicks from ChatGPT, Gemini, Copilot, Deepseek and Grok land in the GA4 AI Assistants channel. Clicks from AI Overviews and AI Mode land in Organic Search. The rest arrives with no referrer at all. Track two numbers, not one, and add a “how did you hear about us” question to catch the gap.

Every month or so a Malaysian business owner forwards us a screenshot of a ChatGPT answer naming their company, then asks the obvious follow-up: how much business did that actually bring in? It is a fair question, and it is where most reporting falls apart — because the honest answer in 2026 is that no single dashboard can tell you.

That is not a reason to give up on measurement. It is a reason to decide upfront which numbers you will treat as real, which are merely directional, and which you will stop chasing. This guide walks through that decision, and sits under our AI SEO, GEO and AEO work — where the same principle applies: decide what you are measuring before you optimise for it.

Before we get into the reports, here is a short walkthrough of the setup itself.

PART 1 · DIAGNOSE

Where AI Traffic Actually Lands in Google Analytics

IN BRIEFAI visits split across two GA4 channels by design. The AI Assistants channel catches ChatGPT, Gemini, Deepseek, Copilot and Grok. Clicks from AI Overviews and AI Mode are filed under Organic Search instead. Anyone quoting one number to measure AI search traffic is quoting half of it.

Google Analytics added an AI Assistants channel to its default channel group, and most write-ups stopped there. Read the definition properly and the interesting part is the exclusion. Google’s default channel group documentation describes AI Assistants as traffic from sources like ChatGPT, Gemini, Deepseek, Copilot or Grok — and states plainly that it excludes AI Overviews and AI Mode. Those clicks stay in Organic Search.

So the same behaviour (a person asking a machine a question, then clicking a link in the answer) gets recorded two different ways depending on which machine they asked. Three consequences follow:

• Two numbers, not one. Your AI reporting needs an AI Assistants figure and an AI-influenced-organic figure. They are not interchangeable.
• Growth in Organic Search may not be search. If AI Mode usage rises, part of your “organic” line is now an AI answer surface behaving very differently from a blue link.
• Two well-known assistants are missing. Perplexity and Claude are not named in Google’s list, so their clicks default to Referral until you build a custom channel group.

The rule behind the channel explains the gaps: GA4 sets the medium to “ai-assistant” only when the referrer matches its internal list. No referrer, no match, no channel.

PART 2 · DIAGNOSE

Why Some AI Clicks Never Show Up At All

IN BRIEFAnalytics only sees an AI visit when the browser passes a referrer header. Native mobile apps often send none, and a copied-and-pasted URL sends none by definition. Those visits land in Direct, indistinguishable from someone typing your address. That gap is structural, not a setup mistake.

Cloudflare, which sees a large share of global web traffic, built its crawl-to-refer metric on exactly this mechanism. Its published methodology counts referrals only where the Referer header carries a hostname belonging to an AI platform. The same write-up notes that traffic referred by Claude’s native app carries no Referer header, and that other native apps most likely behave the same way.

The scale of the imbalance is worth sitting with. Over one measured week in June 2025, Cloudflare reported crawl-to-refer ratios stretching from roughly 70,900 crawled pages per referral at the top end down to 0.1 at the bottom. AI systems read far more than they send back, which is the structural reason AI search traffic looks smaller in your reports than it feels in your inbox.

For a Malaysian business the practical effect is a quiet inflation of Direct traffic. Three visit types cause almost all of it:

• Native app taps. Someone using the ChatGPT app on a phone taps your link and arrives with no referrer.
• Copy-paste journeys. The assistant names your business, the person searches or types the address themselves, and the trail ends.
• Delayed visits. A conversation at 11pm becomes a WhatsApp enquiry two days later, with no digital breadcrumb between the two.

The same referrer mechanics govern Microsoft’s assistant surfaces, which is why we treat it as a separate question in our take on whether Bing and Copilot are worth the effort in Malaysia.

BENCHMARK BRIEFING 1 OF 4

What Can Each Report Actually See in 2026?

IN BRIEFFive surfaces report on AI search traffic and none covers the whole picture. Search Console shows AI impressions but no clicks. GA4 shows clicks but not impressions. Server logs show robots, not buyers. The table maps what each gives you before you promise a client a number.

We compiled this from the platform documents that define the behaviour, rather than from summaries of them.

What each AI reporting surface captures and misses, 2026.

Surface	What it counts	Main blind spot	Setup effort
GA4 AI Assistants channel	Sessions from ChatGPT, Gemini, Deepseek, Copilot, Grok	Excludes AI Overviews and AI Mode	None — automatic
GA4 custom channel group	Any AI source you name, including Perplexity and Claude	Referrer-less sessions still land in Direct	Low — one rule set
Search Console generative AI report	Impressions in AI Overviews and AI Mode, by page, country, device, date	No clicks, CTR, position or query data	None — rollout dependent
Server or CDN logs	Crawler hits by user agent; referred requests by hostname	Says nothing about buying intent	Medium — needs access
Self-reported attribution	Enquiries that name an AI tool as the source	Only covers people who convert	Low — one form field

Source: compiled by IZILI from Google Analytics, Search Console and Cloudflare documentation, 2026. Licence.

The Search Console row deserves a second look. Google’s generative AI performance report documentation confirms it covers AI Overviews and AI Mode, groups data by page, country, device and date, and is still rolling out to a subset of properties. Impressions only — and an empty report may simply mean your property has not received it yet.

PART 3 · DESIGN

Decide Which Number You Will Actually Report

IN BRIEFThree ways to measure AI search traffic are realistic for a Malaysian SME, and they differ in effort more than in accuracy. Choose by what decision the number will drive, the same way you would scope any SEO audit engagement.

The trap is buying the most advanced option before knowing what you would do with the output. If nobody will move a budget on the strength of the number, the lighter approach is the correct one.

BENCHMARK BRIEFING 2 OF 4

How Visible Is an AI Click, Depending on How It Arrives?

IN BRIEFThe same assistant produces very different reporting depending on whether the visitor used a browser, an app, or copied your address. This model maps that difference across the two axes that decide it: which tool sent the visit, and how the visitor arrived.

Read the grid as a visibility map, not a forecast: it shows where a click is likely to be recorded, and where it quietly becomes Direct.

Where AI clicks are recorded in GA4, by tool and entry route, illustrative.

AI tool	Browser click	Native app tap	Copy-paste or later search
ChatGPT	AI Assistants	Often Direct	Direct or Organic
Gemini, Copilot, Grok	AI Assistants	Often Direct	Direct or Organic
Perplexity, Claude	Referral until rule added	Direct	Direct or Organic
AI Overviews, AI Mode	Organic Search	Organic Search	Direct

Illustrative model by IZILI Digital Marketing, based on Google and Cloudflare documentation, 2026. Licence.

Only the top-left corner of that grid is comfortably measurable. Everything shaded darker is a visit you will feel in enquiries before you see it in a report.

PART 4 · DESIGN

How to Set Up AI Traffic Tracking in GA4

IN BRIEFFour steps measure AI search traffic well enough to report on it, and they take an afternoon: check the default channel, add a custom channel group for the tools Google misses, isolate AI Overview impressions in Search Console, then record the enquiry-level question. No paid tool required.

How to set up AI traffic tracking in Google Analytics 4

These four steps produce the two-number baseline described earlier, plus the qualitative check that catches what neither number can.

1. Check what the default channel already gives you. Open Reports, then Acquisition, then Traffic acquisition, and set the primary dimension to Session default channel group. If an AI Assistants row exists, that is your first number — and it only counts forward from the date the channel began reporting.
2. Add a custom channel group for the rest. In Admin, under Data display, create a channel group with a rule matching session source against the AI hostnames you care about, Perplexity and Claude included. Place it above Referral in priority order.
3. Separate AI impressions in Search Console. Open the generative AI performance report if your property has it, and record impressions by page monthly. Treat it as a visibility trend, never as a traffic figure.
4. Add one question to your enquiry form. Ask “How did you hear about us?” as an open field with an AI-assistant option, and log the answer against every enquiry. This is the only source that captures influence rather than clicks.

BENCHMARK BRIEFING 3 OF 4

Which Industries Are Seeing AI Traffic First?

IN BRIEFGrowth in AI-referred traffic is running far ahead of its absolute share, and it is arriving unevenly. Retail leads by a wide margin, media and entertainment trails, and the gap tells you how urgently your sector needs a baseline.

Adobe Analytics tracks AI-referred visits across more than a trillion visits to United States retail sites, which makes it the largest published view of this behaviour. The figures are directional for Malaysia rather than local, but the ordering of sectors travels well.

Year-on-year growth in generative AI referral traffic by industry, 2025 holiday season.

Industry	Relative growth	YoY growth
Retail		693%
Travel		539%
Financial services		266%
Tech and software		120%
Media and entertainment		92%

Source: Adobe Analytics, 2025 holiday season, year-on-year.

The behaviour behind those numbers matters more than the growth rate. Adobe reports that visitors arriving from generative AI assistants were 33% less likely to bounce, spent 45% more time on site, and converted 31% better than other traffic during the same period. Small volume, unusually good quality — which is exactly the profile that gets missed when it hides inside Direct.

PART 5 · DEPLOY

The One Question Analytics Cannot Answer For You

IN BRIEFSelf-reported attribution is the cheapest fix for the referrer gap. One open field on your enquiry form, logged for a quarter, tells you more about AI’s real contribution than any dashboard. Grade the quality of those leads as well, or the count misleads you.

Malaysian buyer behaviour makes this more useful here than in many markets. Telenor Asia’s Digital Lives Decoded survey of 1,000 Malaysian internet users found that 89% use AI daily, up from 75% in 2024, with 51% using it at work. A population that comfortable with assistants will happily tell you they used one.

Three rules make the answers usable rather than decorative:

• Keep the field open, not a dropdown. A dropdown teaches people to pick “Google”. An open box surfaces “ChatGPT suggested you” in the customer’s own words.
• Log it against the enquiry, not the session. The value is in linking it to a closed sale later, which means it belongs in your CRM or enquiry sheet.
• Read it monthly, decide quarterly. One month of answers is noise. A quarter is a signal you can act on.

BENCHMARK BRIEFING 4 OF 4

What Does a 12-Month AI Baseline Look Like?

IN BRIEFThe point of a baseline is to make small numbers readable. This model tracks a mid-sized Malaysian service business across a year, showing measured sessions, the portion likely hidden in Direct, and the enquiries that name an assistant.

The figures are modelled, not measured. What matters is the shape: identified sessions rise slowly, hidden sessions rise alongside them, and enquiry mentions appear well before the analytics line looks impressive.

Modelled monthly AI measurement figures for a Malaysian service business over 12 months.

Metric	Month 1	Month 4	Month 8	Month 12	Month 18*
Identified AI sessions	34	96	205	310	430
Estimated sessions hidden in Direct	20	58	125	190	265
Generative AI impressions	1,100	3,400	7,900	12,600	18,000
Enquiries naming an AI tool	1	3	7	11	16

Illustrative model by IZILI Digital Marketing, based on published AI referral growth rates, 2024–2026. Licence. *Projection.

PART 6 · DRIVE

How to Tell Whether AI Traffic Is Worth Anything

IN BRIEFJudge AI traffic on quality, not volume, because the volume will look disappointing for a while. Compare conversion rate and enquiry value against your other channels — the same test we apply when weighing up any new market or channel.

Decide the review questions before the data arrives, so you are not tempted to reinterpret a small number afterwards. Four are enough:

• Does it convert better than organic? If yes, the channel deserves attention even at low volume.
• Which pages are cited? Comparison and pricing-context pages usually outperform brand pages here.
• Is the enquiry quality holding? Higher conversion with worse-fit enquiries is not a win.
• What would make you stop? Name the evidence in advance — a quarter of flat impressions, for instance.

Set the review cadence at quarterly. Monthly reviews of a channel this small produce reactions to noise, and that is how sensible budgets get moved for bad reasons.

FAQ

Frequently Asked Questions

1. How do I see ChatGPT traffic in Google Analytics 4?

Open Reports, then Acquisition, then Traffic acquisition, and set the primary dimension to Session default channel group or Session source. ChatGPT clicks appear under the AI Assistants channel where a referrer was passed. Whether you see everything depends on how your visitors arrive — app taps and pasted links have no referrer and land in Direct instead.

2. Does the GA4 AI Assistants channel include AI Overviews?

No — Google’s documentation states the channel excludes AI Overviews and AI Mode. That depends on nothing you can configure; those clicks are classified as Organic Search by design. It means your AI reporting needs a second figure from Search Console alongside the AI Assistants number, or you will consistently understate AI’s role.

3. Why is my Search Console generative AI report empty?

Most likely your property has not received the report yet, as Google is rolling it out to a subset of sites. It also depends on volume — a site with few impressions in AI features will see nothing. Check that your pages are not excluded from generative AI features before assuming the report is broken.

4. Can I measure AI traffic without paid tools?

Yes, for most Malaysian SMEs. A custom channel group in GA4, the Search Console report, and one open question on your enquiry form cover the ground that matters. Paid visibility tools become worthwhile when you need to track which prompts cite you across several assistants, which is a different question from measuring traffic.

5. Should I change my content strategy because of AI traffic?

Not on volume alone, at least not yet. It depends on conversion quality — if AI-referred visitors convert better than your other channels, that justifies attention long before the session count does. Build the baseline first, watch one quarter, then decide with evidence rather than urgency.

THE VERDICT

Where This Leaves Your Decision

You will not get one clean number for AI search traffic in 2026, and any tool promising otherwise is selling certainty it cannot deliver. What you can have is a defensible baseline: an AI Assistants figure, an AI impressions figure, and a running count of enquiries that name an assistant.

Build that in an afternoon, leave it alone for a quarter, and judge the channel on how its visitors behave rather than how many arrive. If conversion quality holds up, invest in being the source AI engines cite — the work covered across our AI SEO, GEO and AEO services. If it does not, you have lost an afternoon and gained a number you can trust. That is a reasonable trade either way, and it is the same discipline we apply when advising on how to choose a digital marketing partner in KL.