Tracking AI referrals in GA4. The working method, the limits, and the server-log layer that fills the gaps.

Why this is worth tracking

The short version: AI referral traffic is small, but it is unusually high-intent. Users who reach your site from ChatGPT, Perplexity, Gemini or Copilot have usually asked a specific question, read an AI-synthesised answer, and clicked through to verify or buy. They are deep in research mode. Conversion rates on AI-referred sessions often run above the site average.

If you cannot see the channel, you cannot manage it. The GA4 default reporting buries AI referrals under Referral or Direct, which makes it impossible to tell whether your AI-search optimisation work is paying off. The fix is to surface the channel cleanly so it can be measured the way every other channel is measured.

A second reason. AI search behaviour changes more abruptly than traditional search rankings. A site that was being cited last month may not be this month, and only a measurement loop catches it. The custom-channel-plus-server-logs approach is the cheapest way to build that loop. For the wider measurement context, see the SEO Measurement pillar (placeholder during Wave 3) and the log file analysis chapter.

Why the GA4 default is broken

Two reasons the default reporting fails on AI traffic.

Reason 1: No AI Search channel in the default grouping

GA4's Default Channel Group includes Organic Search, Paid Search, Direct, Referral, Email, Social, Display, Affiliates, and a few others. There is no AI Search channel. Traffic from chatgpt.com lands in Referral, which puts it in the same bucket as a backlink from a random forum post. Traffic from gemini.google.com sometimes lands in Referral and sometimes (depending on the referrer policy) gets attributed elsewhere. The channel does not surface as its own line item.

Reason 2: Stripped referrers land as Direct

Some AI clients strip referrer headers before the click reaches your site. The exact behaviour depends on the client, the platform, the user's browser settings, and the link type the AI rendered. When the referrer is stripped, GA4 has nothing to work with and the session lands as Direct traffic. That hit is genuinely AI-referred; GA4 just cannot tell.

The implication. Any AI referral number you produce will be a lower bound. The true volume is somewhere between what your custom channel captures and that number plus a slice of Direct traffic. We are honest with clients about this: the AI traffic picture is approximate, not exact.

Setting up the custom channel in GA4

The setup takes about ten minutes. Five steps.

Step 1: Open Channel Groups in GA4 Admin

In GA4, go to Admin (the gear icon, bottom left). Under Property settings, click Data display, then Channel groups. Click "Create new channel group". Name it something obvious: "Channels with AI Search".

Step 2: Clone the default group as the starting point

Start by cloning the Default Channel Group so you keep all the existing channels (Organic, Paid, Direct, etc.). The custom group will be identical to the default until you add the AI Search channel on top.

Step 3: Add the AI Search channel

Inside the new group, click "Add new channel". Name it "AI Search". Set the matching condition to: Session source matches regex containing the AI referrer domains (see the list in the next section). Leave the medium open or set to referral.

Step 4: Reorder so AI Search sits above Organic and Direct

Channel order matters in GA4. Drag the AI Search channel above Organic Search and Direct. If you leave it below, traffic that matches both an AI referrer and an Organic Search pattern will get bucketed into Organic and never reach the AI Search channel.

Step 5: Apply the custom group to your reports

In Reports, click the pencil icon at the top right of any Acquisition report and change the channel grouping to your new "Channels with AI Search" group. AI Search will now appear as its own line item with sessions, engagement, conversion data, and the rest of the GA4 reporting machinery applied.

The referrer list and how to maintain it

The list we use across client GA4 properties as of mid-2026. Seven domains cover the bulk of AI referral traffic for Australian businesses.

• chatgpt.com, the current ChatGPT URL.
• chat.openai.com, the legacy ChatGPT URL, still occasionally appears as referrer.
• perplexity.ai, Perplexity citations.
• gemini.google.com, Google Gemini.
• copilot.microsoft.com, Microsoft Copilot (the Bing AI client).
• claude.ai, Anthropic's Claude. Small volume currently but worth including.
• you.com, the smaller You.com AI assistant.

Maintenance cadence

Plan to review the list every six months. The AI search space is moving quickly. New AI clients launch regularly; existing clients change URL patterns. Watch the Referral report in GA4 for emerging high-frequency referrers that look like AI tools you have not added yet.

Domains we explicitly exclude

bard.google.com is largely retired, replaced by gemini.google.com. Bing.com is not classified as AI Search because the regular Bing SERP is mostly traditional results, even though Copilot answers can appear inside it. The classification is judgement, not science.

The server-log layer

GA4 only sees humans on your site. Server logs see everyone, including the AI crawlers. That makes server-log analysis the second pillar of AI-search measurement.

What to look for

Filter your access logs for the following user agents. Monthly review is enough for most sites.

• GPTBot, OpenAI's training-data crawler.
• OAI-SearchBot, OpenAI's live-search fetch agent for ChatGPT Search.
• PerplexityBot and Perplexity-User, Perplexity's indexing and live-fetch agents.
• Google-Extended, Google's AI-training and Gemini user agent.
• ClaudeBot, Anthropic's crawler.
• Bingbot, Bing's crawler, used to power Copilot retrieval.

What server logs tell you that GA4 cannot

• Whether AI clients are reading your content. If GPTBot has never visited your site, ChatGPT cannot cite it. The log is the cheapest way to confirm crawl activity.
• Which pages the AI clients are reading. The crawl pattern across pages tells you which content is being assessed for AI use, even when no referral traffic results.
• The crawl volume trend. Increasing or decreasing AI crawler activity is a leading indicator of changing AI-search visibility.

For the practical setup, the log file analysis chapter covers the tool options (Screaming Frog log analyser, Botify, and open-source options) and the recommended cadence.

Manual citation checks

The third layer. GA4 tells you whether AI traffic arrives; server logs tell you whether AI clients crawl. Neither tells you whether you are actually being cited inside the AI answer. For that, you need to ask the AI clients directly.

The monthly check rhythm

1. Pick ten priority queries. The top queries you care about for your business. Mix of informational, commercial and brand queries.
2. Run each through Google. Note whether an AI Overview appears and which sources are cited.
3. Run each through ChatGPT Search. Note the cited sources.
4. Run each through Perplexity. Note the cited sources (Perplexity's visible source list makes this the easiest of the four).
5. Run each through Gemini. Note the cited sources.
6. Record in a tracking sheet. Date, query, AI client, cited sources, your site's presence (cited, not cited, mentioned without link).

The whole check takes about 90 minutes per month for ten queries across four clients. The output is a citation-trend view that no automated tool currently produces reliably for Australian queries.

Common mistakes

Perth and WA context

Two patterns we see across Perth and WA client GA4 properties in 2026.

Trade and local services see very small AI referrer volume. Most of the AI search activity for these businesses happens in Google AI Overviews on informational queries adjacent to the core service. Direct AI client traffic (ChatGPT, Perplexity, Gemini) is usually under 0.5 percent of sessions. The channel still matters because the few sessions that arrive convert well, but it is not where the visibility battle is being fought. See Local SEO Perth and trades SEO.

Professional services and B2B see meaningful AI referrer volume. A Perth law firm or B2B services business often sees AI referrer traffic at 1 to 4 percent of organic-equivalent sessions, with conversion rates above the site average. The custom-channel setup pays off here because the channel is large enough to be worth optimising for and small enough that it would otherwise stay invisible. See legal SEO and the Content Strategy pillar.

For the wider context, the AI Search pillar covers the full visibility stack, the log file analysis chapter covers the server-log layer in detail, and the AI Overviews chapter covers the manual citation tracking for Google's AI surface specifically. For an entry-level diagnostic of your existing analytics setup, the free SEO audit includes the basic checks, and the website audit service walks through the channel setup as part of the engagement.