GA4 gives you three channels. Your traffic uses nine.

Default GA4 channel groupings were designed around Google's own advertising products. That is not a criticism - it is just the reality of how a product built by an ad platform tends to categorise traffic.

What it means in practice is that channels which matter to your site may not appear as distinct categories in your reports.

Where GA4's default groupings fall short

GA4's default channel groups include Organic Search, Paid Search, Organic Social, Paid Social, Email, Affiliates, Referral, Direct, and a few others. That covers the main bases for most advertising-led programmes.

What it does not handle well:

AI Referrer traffic. ChatGPT, Perplexity, Claude, Gemini, Copilot, Grok, and a growing list of others now send measurable referral traffic to publisher and product sites. In GA4's default configuration, this traffic lands in Referral or Direct depending on whether a referrer header was sent. There is no AI Referrer channel. You cannot separate AI-sourced traffic from a link on a random blog without custom configuration.

Paid Display vs Paid Search. GA4 separates these, but only if your campaigns are tagged correctly. Without precise UTM medium values, display traffic collapses into the wrong bucket.

Campaign traffic. GA4 lumps non-paid UTM-tagged traffic into a "Cross-network" or "Unassigned" group unless you build custom channel definitions. Email campaigns, referral partnerships, and owned-media campaigns can become difficult to distinguish.

The taxonomy matters more than the tool

The right response to this is not to switch tools. It is to define your channel taxonomy deliberately, based on the traffic types that actually matter to your business, and then make sure your attribution tool maps to it.

For most content and product sites in 2025, a useful taxonomy includes:

1. Organic Search 2. AI Referrer (separate from Referral) 3. Social Media 4. Email 5. Paid Search 6. Paid Display 7. Campaign (UTM-tagged, non-paid) 8. Referral (external links, no UTM) 9. Direct / Unknown

Nine channels. Not three, not thirty. Nine gives you enough granularity to act on the data without enough noise to drown the signal.

Why AI Referrer deserves its own channel

AI tool traffic behaves differently from standard referral traffic. Visitors arriving from ChatGPT are typically mid-research - they have read a summary, seen your site cited as a source, and clicked through to verify or go deeper. That intent profile is distinct from a visitor who followed a link in a forum post.

If you are not separating AI Referrer from general Referral traffic, you cannot see that behaviour pattern. You cannot measure whether AI-sourced visitors convert at a different rate, browse more pages, or engage with different content. That is a genuinely useful signal to have, and it is currently going unmeasured for most sites.

A note on what server-side tools can and cannot see

Server-side attribution reads the referrer header sent by the browser. AI tool platforms vary in what they send. Some always send a referrer. Some send one in some contexts but not others. Some rarely send one at all.

The honest figure: depending on the platform and the user's path, somewhere between 30% and 65% of AI tool referrals arrive with no referrer header. A tool that classifies AI traffic by referrer domain - which is the only reliable server-side signal available - captures the fraction that carries a referrer and correctly attributes it. The fraction with no referrer remains in Direct.

This is a limitation of the referrer header as a signal, not a specific tool failure. Client-side tools face the same constraint. The AI traffic that carries a referrer is real, measurable, and worth separating out. The rest is not currently measurable by any attribution method that relies on the referrer.

*CQI Referrer Attribution classifies traffic into nine channels server-side, with a built-in AI Referrer Taxonomy covering 19 platforms including ChatGPT, Perplexity, Claude, Gemini, Grok, and Copilot. The taxonomy is editable and exportable.*