Your Customer Asks AI First: How Do You See That Visitor?

If your organic clicks fell but your sales didn't, or someone asked “are we getting customers from ChatGPT?”, the answer isn't in the click report. How to see the customer who arrives from AI, and how to measure what they're worth.

Alp SuleimanGrowth PartnerArticle 01 / 0211 min read

Published 14 October 2026

Two sentences arrive back to back in a Monday meeting. Someone says, “Organic clicks are falling.” Someone else adds, “But sales didn't fall this quarter.” Then a third person, usually from the board, asks: “Are we getting customers from ChatGPT?”

Nobody in the room can answer with a number. There is no such line in the reports.

This piece is about building that line. But let me be clear about one thing first: I'm not going to tell you how to become visible in AI answers. The subject is channel and measurement. Where does the customer find you, how do you see it, and what do you use to measure that customer's value?

The short answer: the visible metric is the click. The real metric is the customer's decision, and whether that decision is still standing three months later.

Two sources, two different pictures

Let's put the contradiction on the table first, because almost everything you read on this falls into one of two camps.

On August 6, 2025, Google's Liz Reid wrote that total organic click volume from Google Search to websites has been “relatively stable” year over year. The same post makes two more claims: click quality has gone up, and third-party reports showing sharp drops in overall traffic are often based on flawed methodology, isolated examples, or changes that happened before AI features rolled out. One detail matters here: the post gives no number for total clicks. What you are reading is a statement.

Pew Research Center published a different picture on July 22, 2025. They tracked the browsing behaviour of 900 US adults in March 2025. On visits where the search results had an AI summary at the top (what Google calls an “AI Overview”, the box that summarises the answer above the results), people clicked a traditional result on 8 percent of visits. Without a summary, that figure was 15 percent. Clicks on a link inside the summary itself happened on just 1 percent of visits.

Pew's study is observational. Nobody was randomly assigned to a group; searches that trigger a summary may simply be a different kind of search. Still, the direction is clear.

Ahrefs looked at the same question with different data. In an analysis published on April 17, 2025, it compared Google Search Console data for 300,000 keywords between March 2024 and March 2025. The result: the presence of an AI Overview was correlated with a 34.5 percent lower average click-through rate (CTR, the share of people who see a result and click it) for the top-ranking page.

Pay attention to the wording. The article's URL says “reduce”, but what was measured is a relationship: “correlated with”. The top position's CTR also fell over the same period for informational keywords without a summary; where a summary appeared, the drop was steeper (from 0.073 to 0.026, against 0.056 to 0.031 in the other group).

A Bain & Company study dated February 19, 2025 also gives a number: an estimate that zero-click searches reduce organic web traffic by 15 to 25 percent. It is an estimate published alongside a Bain–Dynata survey (December 2024, n=1,117), not a measurement.

Why both can be true

Google says the total is stable. Pew observed that on visits with a summary, clicks were roughly half of what they were without one; Ahrefs found the presence of a summary associated with a markedly lower click-through rate. Both can be true at the same time.

The total is a reservoir. Its level can stay the same while the channel the water flows through changes. Clicks can fall on informational searches that show a summary and rise on other kinds of searches, or on large platforms. Google's sentence describes the reservoir; Pew's describes one tap. Neither tells you which tap your site is connected to.

So “Google says there's no problem” shouldn't reassure you. “Pew says clicks halve when a summary appears” shouldn't scare you either. Your answer is in your own data.

For Türkiye there is also a timing question. When Google opened AI Mode (its conversational search mode) in 180 countries in August 2025, Türkiye was not on the list. On February 18, 2026, Google Türkiye announced that AI Mode and AI Overviews would roll out gradually in the country. The figures above are measurements and estimates from March 2024 to March 2025, and Pew's sample is American. In Türkiye the same shift is happening this year, and I don't know of an independent Türkiye measurement. Read the US data as a leading indicator; it won't tell you what happens in your sector.

Why the visitor from AI looks so small in your report

Say someone asked ChatGPT “which accounting software for small teams?” and your name came up in the answer. That person can reach you in three ways.

First, they click the link in the answer. If the browser passes along where the visit came from (the referrer), you see a session from chatgpt.com in GA4. This is the good scenario.

Second, they click the link inside a mobile app, or copy the address and paste it into a new tab. In these cases the referrer may not travel with them. When GA4 can't tell the source, it records the visit as “Direct”, as if they had typed the address themselves. Some assistants may add a source tag to the link; if so, those clicks show up under their own name and the hidden share is smaller. Check your own data to see which applies. A visitor who lands “directly” on a pricing page or a deep guide rather than the home page is unlikely to have typed that long address from memory. But a link in an email, a bookmark, an address shared on WhatsApp or Slack, an untagged campaign all land in Direct too. AI is only one of the possible sources.

Third, they never click. They remember your name and later search for your brand on Google. In the report this person shows up as organic search; only Search Console shows you that the query was branded. Yet the first place they heard of you was an AI answer.

This is an explanation of a mechanism; I don't know of a reliable number for the share of each path. But the consequence is clear: the “AI” line in your report shows a floor for the customers coming from that channel. It could be one explanation for why your sales held while organic clicks fell. Don't call it the only explanation until your own data shows it.

Why the measurements don't agree

You'll often hear that “visitors from AI convert better”. When you read a number like that, check four things.

The definition of conversion. One report counts purchases, another counts free sign-ups, a third counts newsletter subscriptions. The same channel can look very different depending on the definition.

Self-selection. Someone who asked an AI three questions and narrowed their options has made half the decision before reaching your site. A high conversion rate may describe the person more than the channel. Bring that person in through another channel and they might well have converted too.

Measurement leakage. As described above, some of these visits slip into “Direct” or branded search. Every visit that can't be assigned to the channel shrinks the denominator or gets credited to another channel, so the rate can inflate or deflate for that reason alone.

Time. The channel is new. Early users and the audience that comes later may not have the same profile. Today's rate may not hold once the channel grows.

The common lesson of these four questions: first conversion is a weak signal for measuring a channel's value. The stronger signal is whether that customer stays.

The measure of value: month three

My suggestion: judge the customer who comes from AI by their retention (whether a customer keeps using or paying for the product over time). Concretely, split the customers you won in the same month into cohorts (groups of customers who started in the same period) and look at how many are still active in month three.

I'm suggesting this as a test you run on your own data; there's no study behind it. The logic is simple. If the AI described you to the right person for the right reason, the customer's expectations match the product and they're more likely to stay. If it described you wrongly, the chance of converting in month one and leaving by month three goes up. First conversion can't tell these two cases apart; month three can.

Let me show it with numbers (example). In one quarter you won 300 new customers from organic search and 40 from the AI channel. In month three, 165 of the organic customers are active: 55 percent. From AI, 26 are active: 65 percent. A ten-point gap looks impressive. But in a group of 40, a single customer is 2.5 points. Four more leave and the gap closes. What this table says is: worth watching, too early to decide.

I walk through how to make this split step by step in the Cohort Analysis guide. If you want to see the value in revenue as well as retention, it helps to build the channel as its own branch in the Metric Tree; you can then see which channel feeds which top-level metric.

The counterargument: maybe it doesn't matter at all

To be honest, in many companies this channel may be small today. You might do the setup, wait three months and find 12 sessions and 1 customer on the AI line (example). In that case the right decision is not to put budget or people on the channel.

Value without volume isn't a decision. A 67 percent retention rate that means 2 customers out of 3 (example) doesn't even count as a signal. Remember this every time you see a headline saying “AI traffic is worth this much more”: on a small base, percentages grow fast and fade fast. Before you turn a claim like that into an investment decision, question its base and its definition with the Growth Claim Check.

A threshold to suggest (rule of thumb, not a source): if AI-sourced sessions are below 1 percent of the total, what you have is a signal; watch it but don't commit resources. Between 1 and 5 percent, track it regularly and follow its cohort. Above 5 percent, manage it like a funnel of its own.

Still, starting to measure costs an hour. Not measuring costs you the history you won't have when the channel grows. That trade favours doing the setup today.

There is also a more basic question. Ask it at the meeting table, without looking at the click report:

If Google sent you no clicks at all tomorrow, where would your customers hear about you?

If your answer is “I don't know”, the real problem is dependence on a single channel; AI only makes it visible. If your answer is “our customers tell each other about us” or “our name comes up in AI answers”, start collecting the data that proves it now. In a world where the customer asks an AI first, how your product is understood changes too; I cover that separately under Agent-Ready Product. Where AI really changes growth, and where it only seems to, I've gathered on the AI page and in The Real Growth of AI.

Try it today

Set aside an hour. At the end you'll have a channel definition, a list and a form field.

Build a separate channel in GA4 (20 min). Create a custom channel group in the admin settings and add a channel called “AI assistants”. Condition: the source partially matches this regular expression (regex, a pattern that catches several addresses at once; choose “partially matches regex”, not “matches regex”, since a full match can miss subdomains): chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai. The list may be incomplete; look at the source list in your own report and add any new addresses you see. Move this channel above Referral in the list. GA4 evaluates channels from top to bottom; if Referral stays above it, Referral catches these visits first.
Look inside “Direct” traffic (15 min). List the last 90 days of Direct sessions by landing page. Remove the home page. Write the remaining deep pages (pricing page, comparisons, long guides) and their session counts into a table. This list shows where the visitor with an invisible source lands; it doesn't tell you who sent them. Email, bookmarks, messaging shares and untagged campaigns land here too.
Add a question to your form (15 min). Put a free-text “How did you hear about us?” field on your quote or demo form. Don't give a list of options; let people write “ChatGPT recommended you” or “a friend” in their own words. This field sees what the report can't.
Put a note in your calendar for 90 days out (10 min). In three months, put the customers from the AI channel and from organic search side by side in the Cohort Matrix using segment comparison. The one row you'll look at: month-three retention. Write down the size of both groups first, and read the smaller group's percentage together with that size.

Take care.