Measurement

How to Measure Whether AI Is Sending You Customers

AI assistants do send customers, but most of that traffic never shows up in your reports. Links opened in apps often lose their referrer and land in Direct, and much of the impact involves no click at all. Measure it with a custom Google Analytics 4 (GA4) channel group, cited-page reports, server logs and branded search.

Author Darius Lukosius, Google Ads strategist Published September 5, 2026 Updated September 5, 2026 11 min read

The short version, if you're in a hurry

In apps, a link is often opened without a referring source, so the visit lands in direct traffic rather than a recognizable channel.
When only a source parameter is passed without a medium, the session can fall into an unassigned group and vanish from view.
GA4 has its own AI assistants channel, but it doesn't cover everything. Without your own channel group, you see only part of the picture.
The most valuable report is not the visit count but which pages get cited. It shows what content is worth making more of.
Most of the impact is zero-click: you were mentioned, the person decided and came directly or searched your name. That is why managing by visit count alone is a mistake.

1. Why it barely shows up in your reports

Analytics systems were built for a world where a person clicks a link in a browser and carries along information about where they came from. In conversational search, that mechanism breaks in several places at once.

What happens	Where the visit ends up	Can it be fixed?
An app opens the link without a referring source	Direct traffic	Partly, only if a parameter is passed
A source is passed without a medium	Unassigned group	Yes, with your own channel group
A person copies the address by hand	Direct traffic	No
The answer is received with no click at all	Nowhere	Not directly, only through indirect indicators

The practical takeaway: the number you see is a floor, not the truth. Deciding that "this channel doesn't matter" based on it alone is like judging radio advertising by how many people called directly from the broadcast.

2. A channel group in ten minutes

The first step is to gather in one place whatever can be gathered. GA4 has its own AI assistants channel, but it doesn't cover every platform, so it's worth building your own.

In the Admin section, create a new channel group, add a channel and set the condition so that the session source matches this pattern:

chatgpt\.com|chat\.openai\.com|openai\.com|perplexity\.ai|
claude\.ai|gemini\.google\.com|copilot\.microsoft\.com|
you\.com|poe\.com|grok\.com|deepseek\.com|meta\.ai

Three things are worth doing right away:

Place the new channel above the general referral channel, otherwise sessions will fall into that one.
Add a second condition based on the source parameter, since some visits arrive with it but without a medium.
Create the group before you need it: channel groups work going forward and only to a limited extent backward.

3. How to check that the number is right

Don't trust a newly built channel right away. The check is simple: compare the channel total with the traffic acquisition report broken down by session source.

If the numbers don't match, there are two possible reasons. Either there is an error in the pattern, in which case you simply fix it. Or part of the traffic sits in the direct channel, which is the more interesting case: it means your real channel size is larger than any report shows.

An indirect way to estimate the hidden portion. Note the level of direct traffic before and after you start being cited more often. If direct traffic grows with no campaign running, and the share of new visitors within it rises, that is the strongest evidence you have that some mentions turn into visits with no recognizable source.

4. The most important report: which pages get cited

Once the channel is working, the most useful cut is not the overall number but landing pages by source. This report answers the one question that truly changes decisions: what content is worth making more of.

What you'll see in it:

Which pages are cited most often. Almost always these are specific, question-style texts, not the home page.
Which important pages are never cited. These are candidates for a rewrite.
How this traffic behaves: it is usually smaller, but engagement and conversion tend to be above average, because the person arrives already decided.

If you want a single metric for management, don't take visits. Take the number of conversions from this channel and compare its conversion rate with other channels. The gap is often so striking that the debate about the channel's importance ends on its own.

Need measurement you can rely on?

We set up channel groups, conversion structure and reports so you see not only visits but what they deliver. We start with a tracking check.

Fix my measurement

5. Server logs: what only they can see

Analytics sees people. Server logs see bots too, and bot visits are the earliest signal that your content is entering circulation at all.

What to watch in the logs:

Whether the search bot visits at all. If not, there is nothing to discuss on visibility. You need to check access.
Which pages it visits most often. This shows what the system considers your most valuable content.
Whether visits increase after you publish new content. This is the fastest feedback loop you can get.
Whether error codes appear. A bot that gets an error returns less often.

On most hosting services, logs are available with no extra tools. Half an hour of work once a quarter delivers more than most paid monitoring systems.

6. The zero-click share

The hardest idea in this article: most of the benefit may happen with no visit at all. A person asked a question, got an answer that mentioned your business, and two days later contacted you directly or found you by searching your name.

Analytics doesn't reflect this in any way. But it is visible indirectly:

Search for your name Google Search Console shows how many people search for you by name. Growth without ad campaigns is one of the most reliable indirect indicators.
The makeup of direct traffic Not the overall number, but the share of new visitors within it and which pages they land on. If people arrive "directly" on a deep service page, they got a link somewhere.
A question on the inquiry form A single "how did you find us" field with a few options delivers more than any analytics. Add an option for an AI assistant and watch for a few months.
Regular question testing A list of questions asked once a month, and a record of whether you were mentioned. It's the only way to measure visibility itself rather than its consequences.
Observations from the sales team If customers start arriving with exact phrasing you never wrote anywhere, they read it somewhere. It's worth asking where.

7. What to measure instead of visits

Put a few indicators into one table and review it once a month. There is no single number that answers everything, and trying to find one is the most common mistake.

Metric	Where from	What it shows
Share of mentions	Question testing	Visibility itself, independent of clicks
Cited pages	Analytics and testing	What content to make more of
Conversions from the AI channel	Analytics	Real value, not volume
Branded search	Search Console	Indirect impact without clicks
Bot visits	Server logs	Whether content enters circulation at all
"How did you find us"	Inquiry form	What no system can see

8. Mistakes that lead to bad decisions

Judging the channel by visits. It is the least informative metric of all those available.
Ignoring direct traffic. That is exactly where a large part of the answer hides.
Comparing periods in which measurement changed. You built a new channel group, and the "growth" is simply better visibility.
Trusting a single tool. Monitoring platforms show their own sample, not reality, and their numbers differ from one another.
Reacting to one week's fluctuation. Answers are unstable by nature. Only the trend means anything.
Not measuring visibility itself. Without question testing you are left with only the consequences, and those come with noise.

9. Frequently asked questions

Why does traffic from AI assistants land in the direct channel?

Most often because in mobile apps the link opens without a referring source. In that case analytics assigns the session to direct traffic, because it has nothing to go on.

Is the GA4 AI assistants channel enough?

It helps, but it doesn't cover every platform or every case. A custom channel group with a source pattern gathers a broader picture, especially when a source is passed without a medium.

Can I add tracking parameters to links myself?

For links you distribute yourself, yes. For links an assistant generates, no, because you don't control them. That is why you can't rely on parameters alone.

How much traffic should I expect?

For most Lithuanian businesses it is still a small share of total traffic, but with higher engagement and conversion. Judge it not by volume but by what that volume brings.

Is it worth buying an AI visibility monitoring tool?

Only once you have your own list of questions and understand what you want to measure. Otherwise you will get a beautiful report on a metric whose meaning you don't know. Manual testing is a cheaper and more precise way to start.

How do I talk about this with management?

Show two things: the share of mentions across your tested questions and the conversion rate from this channel. The first shows visibility, the second shows quality. In this conversation, visit count is the most useless number.

Let's fix measurement while it's still cheap

Channel groups, conversions, log monitoring and a question-testing method, all in one plan tailored to your business.

Get a proposal

Darius Lukosius

Google Ads strategist who has managed search and shopping campaigns in the Lithuanian and Baltic markets since 2016. The method is drawn from work in client analytics accounts. Platforms change quickly, so the patterns are worth reviewing every quarter.