GA4 can spot ChatGPT visits now, but that still won't tell you how AI is driving demand

Google Analytics 4 can finally do a decent job of showing visits from ChatGPT, Gemini, Perplexity, and Copilot. That solves one problem and creates another: small teams are about to over-trust a number that only counts the clicks.

If you run a small SaaS, AI discovery usually starts before the session GA4 sees. Someone asks ChatGPT about your category, reads a summary, sees your brand or your page cited, and comes back later through branded search or a direct visit. GA4 will happily label that as organic, direct, or branded. It will not tell you AI did the first part of the work.

How do I measure AI search traffic beyond GA4?

Use GA4 for assistant referrals, then pair it with visibility and demand signals GA4 cannot see. Vynce Digital lays out the practical GA4 setup: create a custom channel group and map known referrers from assistants like ChatGPT, Gemini, Perplexity, Claude, and Copilot. That gives you a clean view of sessions that actually clicked through.

That matters. By default, GA4 tends to bury this traffic inside broader buckets, and Vynce Digital notes Google's native AI Assistant channel only changed in May 2026. Even now, a custom group is the safer move if you want a fuller picture.

But GA4 should be one input, not the scoreboard. The minute you only track assistant sessions, you bias your content decisions toward pages that get clicks today and away from pages that build trust before the click.

AI discovery often happens before the visit GA4 can attribute

Referral traffic from AI tools is real, but it is not the whole story. Growth Unhinged argues traffic is the wrong main KPI because AI influence often happens off-click. A model can cite your page, summarize your product, or mention your brand without sending a visit in that moment.

That matches how people actually buy software. They ask broad questions first. They compare options inside the assistant. Then they come back later and type your brand into Google, click a saved tab, or visit direct. GA4 credits the final hop. It usually misses the earlier AI touch unless the user clicked straight from the assistant.

Growth Unhinged makes the same broader point from the market side: AI search sends little referral traffic, but it still influences discovery. The piece cites 1.5+ billion real user prompts behind Profound's index and says AI search is the #1 channel readers are investing more into in 2026. You do not need to buy an enterprise platform to act on that. You do need to stop assuming low referral volume means low impact.

If AI introduces your product and branded search closes the visit, GA4 will under-credit the page that did the introduction.

Search Console can catch zero-click visibility that GA4 misses

Google Search Console is useful here because it shows impression and query movement even when no assistant referral shows up in GA4. When a page starts appearing more often for problem-aware searches or your branded queries begin to rise, that can be a clue your content is getting surfaced upstream and users are coming back another way.

This is the gap in a GA4-only workflow. Vynce Digital focuses on finding assistant referrals and seeing which pages AI cites. That is useful. Growth Unhinged pushes the next step: measure influence, not just clicks.

For a small team, that usually means watching a few simple signals together:

• AI assistant referral sessions in GA4
• which pages get cited or referenced by assistants
• Search Console impressions and query shifts
• branded search lift over time
• assisted conversions on cited pages, not just last-click conversions

None of these alone is enough. Together, they tell you whether a page is doing discovery work even when the visit gets attributed later to direct or branded search.

Pages that look weak in GA4 may be doing the real selling work

Pages that introduce a category, explain a workflow, or answer comparison questions often look underwhelming in last-click reports. They may not close the session. They may not even get the click from the assistant. But they can still be the reason a buyer trusts you enough to search your name later.

That is the practical mistake I expect small SaaS teams to make next. Once GA4 can label ChatGPT traffic more cleanly, teams will trim anything that does not show obvious assistant sessions. That is backwards. The pages worth keeping are often the ones that earn citations, shape the summary, and create memory before the buyer ever lands on your site.

Start simpler than the big teams do. Set up the GA4 custom channel group from Vynce Digital. Then check whether cited pages line up with rising branded queries and assisted conversions, using the measurement logic from Growth Unhinged.

If a page influences demand upstream, do not kill it because GA4 gave the credit to direct.