How to find your first users hiding in your direct traffic

You are still sending notes and links by hand for the first hundred. The users you cannot see may already be inside Direct: Aimerce's Shopify store data showed ChatGPT's pageview share increasing 1.97x over the year to September 2026. The harder part is that the visits often arrive without the usual referrer: Aimerce's data showed that the proportion of ChatGPT visits with no referrer rose from 68.1% in September 2025 to 81.8% in September 2026.

Shopify's Q2 2026 merchant data showed AI-referred sessions rising 197% year over year. That is a signal, not a channel plan. You do not need a new stack. You need a small change to the links you already send and a monthly look at the Direct bucket. For a founder before distribution, the useful question is whether assistant visits are already hiding in your Direct bucket, not whether they can be scaled into a campaign.

Assistant visits often arrive as Direct

Assistant links can drop the referrer that analytics normally use to name the source. Aimerce calculated that a referrer-only rule would fail to identify 81.8% of ChatGPT pageviews. Clickport's 30-day snapshot showed that 35.7% of recognized AI Search sessions lacked a recorded referrer. A small check can stop you from mistaking a tracking gap for an unknown source, rather than proving that assistants are a growth engine. When your first hundred is still coming from hand-holding, that missing referrer can make assistant visits look like ordinary direct traffic.

GA4's new channel is not enough for the first hundred

GA4 gained a native AI Assistant channel on May 13, 2026. That helps later, but it does not fix the click path that is already confusing your current reports. Wislr's device testing found that ChatGPT, Gemini, and Claude mobile apps can disrupt UTM and referrer tracking, causing GA4 to miss 2.5x to 5x of actual LLM traffic. For the first hundred, the channel label matters less than the click path. For a new product, the missing slice may be the only slice you need to see.

Run the five-step check while you are still hand-holding signups

The check should be boring. It should answer one question: did the Direct bucket shrink because you labeled the assistant path, or did it stay flat because the path is still untracked? Keep it small enough to finish in one sitting.

• Label assistant links with utm_source values that identify the assistant. Done is a test click from a shared assistant link landing in a source you can filter, not in the Direct bucket. Make the source visible before you wonder where the visit came from, whether the link comes from a note, a doc, or a chat.
• Match first-pageview referrers to chatgpt.com, chat.openai.com, gemini.google.com, and claude.ai. You know it works when a small report shows which assistant domains appear before the first pageview. You are not trying to count every assistant. You are trying to catch the ones that already show up.
• Compare the Direct bucket before and after the labels. The output is a month-over-month view where the Direct share changes in a way you can explain. A fall in Direct after you add labels may mean you found the hidden slice. A flat Direct share may mean the gap is in the click path.
• Treat a low share as an early warning, not a channel. The habit is a note in your review, not a new growth tactic. A tiny share can tell you that assistant visits are starting to appear. It cannot tell you how to scale them.
• Review it monthly instead of building a strategy on it. The result is a quick check that tells you whether assistant visits are becoming visible. You are looking for a change in shape, not a number big enough to justify a plan.

When the number stays small, your job is still to talk to the people who signed up and ask how they arrived. A user who says they found you through an assistant gives you a record even if the analytics never show it. The answer often comes from the person, not the report.