Your clients come from ChatGPT. Google Analytics says "Direct".
When an AI recommends your business, the person often types your name instead of clicking. The visit then lands in "Direct" in Google Analytics and the link to the AI disappears. Visits from AI can be under-attributed by up to ten times as a result. The most reliable measure is still to ask your clients how they found you.
Why your numbers no longer tell the truth
Since this summer, I have been getting one to two inquiries a week from people who found my name by asking ChatGPT a question. I know because they tell me on the phone.
My Google Analytics does not know. As far as it is concerned, these visits come from nowhere: they are classified as "Direct", as if each person had typed my address from memory.
If I trusted my reports, I would conclude that AI brings me no clients. That would be wrong, and it is exactly the conclusion an SMB reaches when it reads its reports at face value.
Why AI ends up in "Direct"
Google Analytics attributes a visit by looking at where the click came from. A person clicks an ad, a Google result or a LinkedIn post, and the website gets a label: the source.
The problem is that an AI recommendation does not always involve a click.
Here is the typical path:
- Someone asks ChatGPT "which mortgage broker should I pick in Laval?".
- The AI names two or three businesses, often without a clickable link.
- The person reads the name, opens a new tab and types the address, or searches the name on Google.
At step 3, the label is lost. The website sees someone arrive who "typed the address": a Direct visitor, by Google's own definition. When the person goes through Google, it is even more misleading: the visit gets credited to Google, not to the AI that did the real work of recommending you.
When the AI provides a link and the person clicks it, some of those visits keep their label. The path without a click leaves none.
How many clients your reports are missing
According to a Graphite analysis cited by SEO analyst Kevin Indig in his Growth Memo of September 28, 2026, visits coming from AI can be under-attributed by up to ten times in classic analytics tools.
That is an observed maximum, not an average. But the order of magnitude is enough: the "ChatGPT" line in your reports, when it exists, is the tip of the iceberg.
The tools are not broken. They measure what they can see, and they see less and less. AI answers get consumed without a click. Some visitors also decline cookies, which is their right.
Three signs AI is already bringing you clients
You do not need a tool to check. Three clues, visible in what you already have:
1. Your "Direct" traffic is rising for no reason. You launched nothing that explains it, no ads, no newsletter. Yet the Direct line has been climbing for a few months. For an SMB, those are rarely people who know your address by heart.
2. Your new clients describe you in an AI's words. They arrive already knowing what you do, for whom, and sometimes your prices. They are not comparing you, they have already chosen you. An AI summarizes, and people arrive with the summary.
3. Searches for your name are going up. In Google Search Console, look at the queries that contain your business name. When an AI names a business without a link, people go search for it on Google. A rise in searches for your name, with no campaign to explain it, is often that.
How to measure where your clients really come from
No tool solves the problem completely. What works is crossing three measures that each have a different blind spot.
1. Ask.Add an optional question to your contact form: "How did you find us?", with "ChatGPT or another AI" among the choices. It is the simplest and most reliable measure, because it does not depend on any click. One tip: do not put AI first in the list, or you will steer the answer.
2. Send that answer to your analytics tool. If your form talks to Google Analytics, the answer can travel with the conversion. You can then compare what the person declares with the source the tool thought it saw. The gap between the two is your under-attribution rate, measured on your own clients.
3. Test the AI yourself, every month. In a private window, ask ChatGPT, Perplexity and Google the question your ideal client would ask, and note whether you come up. Always the same question, always the same conditions. That gives you your share of voice in the answers, month after month.
When all three move together (more presence in the answers, more leads declaring AI, more contracts), you have your answer. When they diverge, you know where to dig.
What I did on my own website
I added the question "How did you find me?" to my form in September, with the answer flowing into Google Analytics. I run no advertising for my business, and for months my reports told me my inquiries came from "nowhere".
The inquiries stayed the same. What changed is that I can now prove where they come from instead of guessing. In a few months I will have a real number instead of an impression, and I will publish it here.
Frequently asked questions
Can Google Analytics see ChatGPT?
Partly. When the AI provides a link and the person clicks it, the visit can show up under a source like chatgpt.com. But a large share of visits arrive without a click (the person types the name or searches it on Google), and those land in Direct or get credited to Google.
Why is my Direct traffic going up?
If you launched nothing that explains it (ads, newsletter, press), one of the most common causes in 2026 is a recommendation by an AI. The person reads your name in an answer and types your address. Also check searches for your business name in Search Console.
Does a paid tool fix the problem?
There are tools that track your presence in AI answers, and they are useful for share of voice. But no tool can see what happens between the AI's answer and the arrival on your website. The question in your form remains the most direct measure.
Does this replace my conversion tracking?
No, it adds to it. Click-based tracking remains the right way to measure paid advertising. It is for clients coming from AI and referrals that you need a declared measure on top.
Want to know where your clients come from?
I set up conversion tracking for service SMBs in Quebec, including the measurement of clients who came from AI. We start by looking at what your reports see and what they miss.
Measure my real sourcesSource: Kevin Indig, Growth Memo, "The usefulness of attribution for AI Search", September 28, 2026, citing a Graphite analysis.