How AI Search Conversion Tracking Works (and Why GA4 Misses 70%)

Your AI search traffic is probably converting better than your Google traffic. You just can't see most of it. 70.6% of AI-driven visits land in GA4 as “Direct” — the wastebasket bucket where attribution goes to die. Meanwhile, the slice you can measure is quietly outconverting non-branded organic by 31%.

So you're underinvesting in your best channel while overcrediting a worse one. Let's fix that.

How attribution is supposed to work

Before AI broke it, web attribution had one job: read the Referer header your browser sends with every request. That header says “I came from google.com,” GA4 turns it into source=google / medium=organic, and your dashboards light up correctly.

AI assistants violate this pipeline in three different ways — and each violation hides a slightly different slice of your conversion data.

The 31% conversion-rate ghost

Here's the part that makes underinvesting feel painful. Visibility Labs ran the numbers across 94 ecommerce brands — 135,000 ChatGPT sessions, 9.46 million organic sessions. ChatGPT didn't just keep up. It outconverted non-branded organic by a wide margin.

For some industries the gap is wider. Hotels and resorts convert at 7% from ChatGPT. Legal services at 5.6%. Healthcare at 4.5%. The pattern repeats across every published dataset: the people AI sends you have already been pre-qualified inside the conversation.

ChatGPT Atlas: the irony

You'd assume OpenAI would help marketers see ChatGPT-driven traffic. It does the opposite. ChatGPT Atlas, the agentic browser launched in late 2025, often strips or blocks referrer headers entirely — so links opened through Atlas land in GA4 as Direct or (not set). Perplexity's browser Comet, by contrast, passes referrer data like a normal browser and shows up cleanly as perplexity.ai / referral.

So Perplexity, the smaller player, gives you better attribution than the market leader. Welcome to AI search analytics.

The four paths an AI visit can take

When someone reaches you via AI search, their visit travels one of four routes through your analytics. Only one of them shows up as “AI traffic.”

How to fix your tracking

Nothing here recovers the full 70% — some referrer data is genuinely gone for good. But each of these moves a chunk of it from “invisible” to “countable.”

One gotcha: the rebound illusion

Once you implement the fixes above, your reported AI-search traffic will appear to explode — not because anything changed, but because you finally stopped throwing the data away. Don't put it in a press release. The traffic was already there; you just bought glasses.

And if you previously calculated organic conversion rates with all that invisible AI traffic mixed in, your “organic” numbers were artificially inflated. Splitting AI out will make organic look slightly worse than you thought. That's also the truth catching up with you.

The short version

AI search conversion tracking works the same way regular attribution does — it reads referrer headers and UTMs. The problem is that AI assistants strip, drop, or sidestep both signals for 70% of the traffic they send you. The fix isn't a single setting; it's layering several second-best signals (custom channel grouping, edge-side referrer logging, self-reported attribution, brand-search velocity, direct citation monitoring) until your visible picture stops lying to you.

Treat AI search as a real channel before your attribution catches up. The numbers will lag the reality by a year or two. The brands that move now compound a head start; the ones waiting for clean dashboards will be late.