How to Find Out How Much of Your "Direct" Traffic Is Actually Coming from ChatGPT

You cannot get a perfect number, and anyone selling you one is guessing. What you can do is build a defensible floor and a defensible trend: count the AI sessions your analytics does record, tag every link you control, then isolate the slice of "direct" traffic that lands deep in your site on pages AI engines are known to cite, and measure what that slice converts at. Run that for a few weeks and you will have a channel you can size, defend in a board meeting, and watch move.

Here is how to do it in order.

Why does ChatGPT traffic show up as direct traffic in the first place?

A referrer is a header the browser passes along saying where a click came from. AI assistants frequently do not pass it. The ChatGPT desktop and mobile apps behave like native apps and send nothing, the same way an email client does. Chat interfaces that open links in an embedded webview often suppress the header too, and some set a no-referrer policy outright. Your analytics receives a session with no origin, so it files it under direct.

Perplexity is the exception. It passes referrers more consistently, which is why many teams see a tidy perplexity.ai row and almost nothing for ChatGPT, then wrongly conclude Perplexity is their big AI channel. It is usually just the visible one. We wrote more on the mechanics of how AI assistants pass referral data.

Step 1: Build an AI channel out of what you can already see

Before you estimate anything, count what is measurable. In your analytics, create a custom channel grouping that captures sessions whose source matches the engine domains that matter for B2B: chatgpt.com and chat.openai.com, plus perplexity.ai, claude.ai, and gemini.google.com.

Then check your campaign parameters separately. ChatGPT appends utm_source=chatgpt.com to some outbound links, and those sessions may be sitting in a report you never open. Add them in.

That total is your floor. It is wrong, it is low, and it is real. Write it down with a date.

Step 2: Tag every URL you control before an assistant repeats it

Assistants quote URLs they find in content. If a URL carries a source tag when it is crawled, it carries that tag when a buyer clicks it, referrer or no referrer. So tag the links you place in guest posts, partner pages, documentation, review listings, and release notes. A consistent convention beats a clever one: pick something like utm_medium=ai-assistant and never deviate.

This is the one lever that converts future dark traffic into attributed traffic, and it only works forward. Doing it today is worth more than analyzing it for a month.

Step 3: Segment "direct" by landing page, not by volume

Here is the tell. People who type your domain into a browser bar land on your homepage. People sent by an AI assistant land on the specific page that answered their question: a comparison post, a pricing page, an integration doc.

So split direct sessions by landing page and pull out every session that landed somewhere other than your homepage or a page a returning customer would bookmark. Deep-landing direct traffic with no prior session is the strongest available signal of an AI referral.

Step 4: Cross-reference those landing pages against pages engines actually cite

Segmentation alone is inference. It becomes evidence when you can show that the pages absorbing your deep direct traffic are the same pages AI engines cite when your buyers ask their questions.

Get that list by asking the engines directly. Run your real buyer questions through ChatGPT, Claude, Gemini, and Perplexity, and record the source links in each answer. Clarity's prompt monitoring does this daily and keeps the full answers and their source URLs as receipts, which is exactly the list you need here. When page X shows up in the answers and page X is also eating unexplained direct traffic, you are no longer guessing.

Step 5: Judge the segment by conversion rate, not sessions

Sessions are the least interesting part of this. Compare the conversion rate of your probable-AI segment against the rest of your direct traffic and against organic. If it converts noticeably higher, that is consistent with what the channel does: Semrush's 2025 analysis of AI search traffic put the value of an LLM-referred visitor at 4.4x a traditional organic search visitor, because somebody arriving on a warm recommendation is already most of the way through their evaluation.

Run the arithmetic once and the stakes get obvious. Say direct shows 4,000 sessions a month. Strip out homepage landings and anyone with a prior session and you are left with 600 deep-landing first-touch sessions. Cross-reference step 4 and 380 of those hit pages that appeared as sources in AI answers. If that 380 converts at 3 percent while the rest of direct converts at 0.7 percent, you are looking at roughly 11 leads a month your reports currently credit to nobody. Multiply by your close rate and average deal size and you have the one sentence that ends the argument with finance.

Then attach money for real. Tie the segment to your lead and purchase events so the output is leads and revenue, not a traffic chart. A number with a dollar sign survives the board meeting. A sessions line does not.

Step 6: Ask the buyer, and reconcile

Add a free-text "how did you hear about us" field to your demo form. It is the cheapest ground truth you will ever buy. When self-reported ChatGPT mentions run far ahead of your measured floor, the gap is your dark traffic, and it is a number you earned instead of imported from someone else's blog post.

What we do, and what we deliberately do not claim

Clarity's attribution layer reads the sessions your analytics records from ChatGPT, Claude, Gemini, and Perplexity, ties them to lead and purchase events, and reports down to the dollar per engine and per page across your whole site, not only the posts Clark wrote. Pages Clark published carry a flag, so you can see which content earned its keep.

What it does not do is invent the invisible part. Traffic that arrives with no matching source is not detected, and we would rather hand you a conservative number you can defend than a confident one you cannot. Everyone shows you rank. We show you revenue, and only the revenue we can actually trace.

If you want the measured baseline before you touch your analytics config, the free AI visibility check shows you which buyer questions return your competitors and which return you. That tells you which pages to go hunting for in your direct traffic.