Why Measuring the Value of AI Is Much Harder Than You Think

You open your traffic report and see organic search up slightly, direct traffic ticking up, and a few brand keywords appearing more frequently than before. There is not a single line that says "ChatGPT." Yet, customers keep arriving, and orders keep coming in. What is actually going on?

That is precisely what makes measuring the value of AI one of the most frustrating puzzles in modern marketing. AI might have introduced your brand to someone three days ago, but when they type your name into Google, all the credit goes straight to... Google. And you sit there looking at the report, wondering if you are burning money on a channel that delivers nothing.

A figure worth pausing for

Similarweb research tracked people who asked ChatGPT an industry question and received a suggestion for a specific brand, then observed their behavior over the next seven days. The result: the group that received the brand recommendation was 2.5 times more likely to visit that brand's website compared to the group that received suggestions for a competitor.

What makes this credible is their sampling method: only people who had never visited the website and had never mentioned the brand name in their prompt. In other words, they measured the true ripple effect of a recommendation, rather than just recounting people who already knew the brand beforehand. The research scope covered finance, travel, and beauty.

And the most interesting part is this: the majority of people coming "from AI" do not actually come from AI, at least on paper.

55.9% of AI-influenced customers arrive via search

Similarweb recorded that 55.9% of AI-influenced traffic arrives through search channels, compared to 40.4% in the uninfluenced group. In other words, the traces of AI blend right into normal-looking organic traffic — the same color, the same label, the exact same line in the report.

In analytics, that journey shows up as a brand search. No referrer from ChatGPT. No UTM. Not a single trace to tag. You look at it and think: "Ah, returning customers." When in reality, it could be a new prospect who was just introduced to you by AI for the very first time.

The measured traffic is real. But the source of influence has vanished somewhere between discovery and conversion.

This is where many marketers misread the picture. If you only look at referral traffic from AI platforms, you will conclude that "AI doesn't drive much." Meanwhile, AI is working at the awareness layer — before the very first click happens. And the awareness layer doesn't have a column in any dashboard.

A very realistic journey, and a very blind report

Imagine this: someone is looking for management software for a small business. They ask AI which platform is worth trying. AI responds and mentions a brand name. That person does not click the link. They read a few more reviews, compare options, ask friends, and two days later open Google to type in the exact brand name.

To analytics, that is a branded search. To the user, the journey started with AI. Two perspectives, two stories, and only one of them gets recorded. Yet the unrecorded part is precisely what created the demand in the first place.

There is another noteworthy detail in the study: the AI-influenced customer group viewed nearly twice as many pages and spent almost twice as much time on site compared to the rest. They arrive more quietly in the reports, but they are significantly higher in quality. That is the kind of signal you will completely miss if you only look at raw traffic volume.

Last-click is not wrong — it is just incomplete

The last-click attribution model still works fine for short, direct journeys: see an ad, click, buy. The problem is that AI inserts an invisible step at the top of the funnel, and that step emits no trackable signals.

This directly affects how you evaluate SEO and GEO. Both are games of securing visibility before people visit your website. If you only measure the final click, you will inadvertently undermine your own value — and then cut budgets for the exact channel that is quietly nurturing demand.

Here are several signals to look at in parallel, rather than relying solely on referrals:

• Branded search demand fluctuating over time
• Direct traffic increasing without any obvious reason
• Engagement quality: pages per view, time on site, and return rates
• Brand appearances in AI tool responses
• Conversions occurring further down the line, not just on the first visit

How to get started without a perfect model

You do not need, and shouldn't wait for, a perfect AI attribution model to be invented. What you need to do is much simpler: monitor the signals sitting behind AI presence, and read them together rather than in isolation.

1. Log your brand's level of presence in AI answers on a monthly basis, using a fixed set of prompts to ensure comparability.
2. Place branded search and direct traffic charts side-by-side for the same period.
3. Highlight moments where both sides experience sharp shifts.
4. Review these alongside content campaigns, PR pushes, and website updates to rule out other external factors.
5. Formulate hypotheses and test them gradually — rather than jumping to hasty conclusions.

This method won't give you a pretty number for a report. It gives you something far more valuable: a picture with context, sufficient for making decisions without guessing in the dark.

Caution: correlation is not causation

One thing must be made clear. Similarweb's research shows a correlation between AI recommendations and subsequent traffic, but it does not prove that every subsequent visit was generated by that recommendation alone.

Rising branded search could stem from a PR campaign, a viral video, seasonality, or simply better content creation than last year. Attributing everything to AI is just another form of self-delusion — merely flipping the coin from completely denying AI. Both are narratives that feel much more comforting than the truth.

The healthiest attitude lies in the middle: treat AI as part of the customer journey, not an isolated traffic switch you can turn on and off. Measure discovery, consideration, and conversion simultaneously, rather than focusing solely on the final touchpoint.

Interestingly enough, the greater AI's marketing value becomes, the harder it is to prove with standard reports. It lives within the buyer's memory, in the name they suddenly recall when a need arises, in a keystroke made days later — long after you have forgotten where you first appeared.

If you run a small brand, do not rush to dismiss AI just because you "don't see the traffic." Try observing your branded search trends over the next three months, paired with tracking whether your brand appears in AI answers. You might just spot a curve you never noticed before.

What about you — do you believe a portion of your branded search traffic is coming from AI, or do you think it is just wishful thinking by SEOs trying to secure bigger budgets? I genuinely want to hear the counterperspective, because this is a gray area where no one has a definitive answer yet. Leave a comment, and if you found this article helpful, drop by SilverZ.Net often — I am still writing more about topics this hard to measure.