The question I get most often in 2026 from B2B SMB founders is some version of: "we know we are getting AI traffic, why does it not show up anywhere we can see." Part of the answer changed on May 13, 2026: Google added a native "AI Assistant" channel to GA4's Default Channel Group. It auto-classifies sessions from ChatGPT, Gemini, and Claude with no setup required, and it applies retroactively. If you have not looked at your channel report since that date, check it before doing anything else in this guide. You may already have data you did not know was there.

The native channel is a floor, not a ceiling. It only catches the three engines Google recognizes, and only when the referrer header survives. ChatGPT's Atlas browser opens links in an internal sandbox that strips the referrer entirely, so that traffic still lands in Direct and the native channel never sees it. Perplexity, Copilot, and your own AI-citation UTMs are not covered at all. This post walks through the custom channel group I still build for client accounts on top of the native one, so the picture includes the engines and referrer-stripped traffic Google's default does not catch.

Step 1: Confirm the native AI Assistant channel, then layer a custom group on top

First, in GA4 Reports → Acquisition → Traffic acquisition, check whether "AI Assistant" already appears as a row in Session default channel group. If it does, Google is already classifying ChatGPT, Gemini, and Claude sessions for you — do not rebuild that part.

Then, in GA4, navigate to Admin → Data display → Channel groups, and create a new custom group. Do not edit the default. Name the new group "AI Traffic" or similar, and position its rule above Referral.

The rule set as of mid-2026 should include the following channels:

  • AI Traffic (custom): source matches regex chatgpt\.com|chat\.openai\.com|openai\.com|perplexity\.ai|claude\.ai|anthropic\.com|gemini\.google\.com|bard\.google\.com|copilot\.microsoft\.com|copilot\.com|bing\.com/copilot|you\.com|phind\.com|kagi\.com OR medium matches regex ai|llm|chatbot|ai-citation|ai-overview|ai-mode.
  • Google AI (Overviews + AI Mode): source is google AND medium is organic AND landing-page parameter matches known AI-mode fragments. Imperfect — see notes below.
  • Standard Organic Search: the rest of source = google / medium = organic.
  • Keep all other default channels (Direct, Paid Search, Social, etc.) intact for continuity.

Apply the new channel group to standard reports. The retroactive application is the part that surprises most operators — you will see the last 12 months of data recategorize when the new group activates. This custom group and the native AI Assistant channel are not redundant: use the custom group when you want per-engine breakdown or need to catch Perplexity/Copilot, and check the native channel as the quick, zero-setup baseline.

Step 2: Acknowledge the limits of the AI Overviews bucket

The Google AI Overviews and AI Mode portion of this channel group is the weakest link. Google has not committed to a stable referrer schema for AI-surface traffic, and the URL parameters that occasionally signal AI Mode (such as ?udm variants or AI-specific path fragments) change without notice. The bucket will undercount AI Overviews-attributed traffic.

The workaround is to triangulate. Pair the GA4 channel group with the manual citation tracking sheet covered in the AI Overviews citation tracking post. When the manual sheet shows your site cited for a high-volume query and the GA4 channel group does not show a corresponding session bump, you are looking at unattributed AI traffic — and the gap is real but bounded.

Step 3: Instrument Direct for hidden AI traffic

The bigger hidden bucket is Direct. Many AI assistants — especially mobile in-app browsers for ChatGPT, Perplexity, and Claude — strip or modify the referrer header before the user lands on your site. The traffic arrives with no referrer at all, and GA4 files it as Direct.

The fix is not instrumentation; the referrer is gone before the request reaches your server. The workaround is correlation. Track the month-over-month growth of Direct traffic against the growth of branded search impressions in Google Search Console. If Direct is growing materially faster than branded search, the excess is most likely AI-referred. For B2B SMBs with active AI citation strategies, the excess in 2026 typically runs 15 to 40% of Direct.

Step 4: Build the AI traffic explorer

Create a GA4 Exploration that pivots on the new channel group with these dimensions and metrics:

  • Dimension: Channel group (custom: AI-aware channels)
  • Dimension: Landing page
  • Metric: Sessions
  • Metric: Engaged sessions
  • Metric: Conversions (filtered to lead-form-submit or your equivalent)
  • Date range: last 90 days

This report answers the question the founder actually wants: which pages are getting AI-attributed traffic, and which of those pages are converting. The answer is usually surprising. The pages that convert best from AI traffic are not always the pages that convert best from classical organic — AI traffic tends to skew toward higher-intent, deeper-funnel queries, which makes per-session conversion rates 1.5 to 3x higher in many B2B SMB accounts.

Step 5: Report the three numbers monthly

For a non-technical stakeholder, the AI traffic report is three numbers:

  1. AI-attributed sessions from the custom channel group.
  2. Direct-share-growth differential against branded search — the proxy for hidden AI traffic.
  3. AI Overviews citation count from the manual tracking sheet.

The combination produces a credible directional read without overclaiming precision. Resist the urge to assign a single number to "AI traffic" — the surfaces, the referrer behaviour, and the citation patterns are still moving fast enough that any single number will be wrong inside of a quarter.

What to do with the data once you have it

Three actions. First, prioritize content refreshes against pages that earn AI traffic and convert — those are the most valuable pages on the site and they decay faster than classical organic. Second, identify the AI assistants that drive the most converting traffic and tune content style toward what they cite (Perplexity and Claude favour different passage structures than ChatGPT). Third, escalate the share of monthly content production toward the page templates that earn AI citations, which is usually a different mix than the page templates that earn classical SERP clicks.

When the AI traffic channel goes dark

Two scenarios break the GA4 AI channel group after you have set it up:

Scenario 1: A major AI platform changes its referrer behavior. ChatGPT changed how it passes referrer data in early 2025, causing a spike in Direct traffic that looked like a drop in AI assistant traffic. Perplexity has similarly adjusted referrer headers. The fix is to audit the Direct bucket every quarter — if Direct is up meaningfully but lead quality from Direct contacts is also up, the most likely explanation is AI referral traffic migrating into Direct again. Run the correlation test: Direct traffic volume vs branded search volume. If they move together, Direct is absorbing AI referrals.

Scenario 2: You lose a key AI citation. A citation you held for a query disappears — a competitor published a better answer, or an AI engine updated its retrieval model. This shows up as a drop in AI channel sessions before it shows up in organic sessions. If you are running the manual prompt panel monthly, you catch it when it happens. If you are not, you see it as an unexplained traffic dip three months later. The prompt panel is the early warning system.

What the monthly AI traffic report should actually contain

Three numbers, in this order:

  1. AI channel sessions this month vs last month vs 90 days ago. Trend direction is more useful than absolute numbers because the absolute numbers are directional estimates, not true counts.
  2. AI channel conversion rate vs organic conversion rate. Persistent difference (AI converts higher or lower than organic) is a signal about intent quality. AI search visitors who have been cited a specific passage typically arrive with more context about the brand and convert at higher rates than cold organic visitors on informational queries.
  3. Prompt panel citation rate this month. Of the 20 priority queries you run monthly, how many cited the brand — and on which engines? This leading indicator moves faster than traffic data and tells you where GEO optimization is working and where it is not.

That is the AI attribution report for an SMB. One page, three numbers, updated monthly by whoever owns the GA4 account. It does not require a BI tool or an enterprise analytics platform. It requires the channel group configuration, the prompt panel habit, and a founder who looks at the numbers rather than ignoring them because the referrer column seemed confusing.

AI attribution in GA4 is solvable with an afternoon of configuration

AI attribution in GA4 is solvable for SMBs willing to do the configuration work and to accept that the answer is directional rather than exact. The custom channel group recovers the bulk of explicit AI assistant traffic. The Direct-correlation proxy estimates the hidden share. The manual citation sheet completes the picture for Google AI surfaces specifically. None of this is a single tool purchase. All of it is reachable in an afternoon if the operator commits to running it monthly. The AI marketing ROI measurement post covers how to fold these numbers into the broader pipeline reporting that a founder actually cares about. For the broader GEO strategy, see the GEO for SMBs guide. For a deeper look at attribution methodologies, see the guide to marketing attribution models for B2B SMBs.