Here’s the conversation that tends to happen in most organizations as soon as AI research comes up in a strategy meeting: someone asks what its actual value is, someone points to AI referral sessions in GA4, and then someone, usually in finance, wonders why that number is so small.
The answer is that observed AI referral traffic is the floor of AI’s business contribution, not the ceiling. And if you’re only reporting that number, you’re systematically underestimating the impact.
This post is about how to build a more complete and credible picture, one that explains the impact of AI you can see and the impact you can’t, without making claims you can’t support.
The four layers of trust
When you’re reporting what AI research actually does for your company, don’t try to squeeze everything into one number. Instead, think of it as four separate groups, each with a different level of reliability:
What you can see live: This is the traffic that arrives at your site from the AI platform and is recorded in your analytics. Someone clicks a link in ChatGPT, lands on your site, and you can see it. Powerful data, but it only tells part of the story, as many people see your brand in the AI answer and never click on it, but go and search for you later.
Evidence from your own data: This is where you look for signs that AI is affecting people even when you can’t directly prove it. Things like: Are there more people than usual searching for Koozai by name? Is traffic to certain pages increasing without any campaign driving people there? Are more people tagging “AI Assistant” when you ask how they found you in your contact form? None of these prove conclusively that AI drove them, but together they paint a picture.
What external tools tell you: Tools like Sameweb can estimate how much AI traffic you and your competitors are getting across the web. It’s not your own data, so it’s less reliable, but it’s the only way to know how you’re doing compared to others, and understand what types of questions are sending people to your pages in the first place.
Your best educated guess: This is where you take everything above and put together a rough trading number for planning purposes. Something like this: “We believe AI impacted approximately 30% of our brand search growth this quarter. Based on how that traffic typically converts, that’s likely to be worth between £X and £10 in the pipeline.” Be honest that this is an estimate with assumptions behind it; Never present it as hard income or it will fall apart the minute someone asks you how you got there.
Each layer answers a different question. Keep them separate in your reports, label the trust, and never provide a model estimate as evidence.
Set up the GA4 to capture what it can
First, check if you already have an AI-powered channel in Google Analytics4
Go to Reports, then Acquisition, then Traffic Acquisition, and set the primary dimension to the session default channel group. If you’ve had AI traffic since May 2026, you should see your AI Assistant row appear automatically. If there is, that’s the starting point.
But don’t rely on it alone. The built-in channel misses Perplexity entirely, and a significant portion of AI traffic still lands in Direct or Referral because referrer information is lost along the way. To get more of them, you need to set up a custom channel group as well.
Set up a custom AI channel group
Go to Admin, then Data View, then Channel Groups. Create a new group, add a channel called something like “AI Search,” and set it to capture traffic from the major platforms by their domain names: chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, copilot.microsoft.com, and any others that appear in your referral data. One thing to fix: custom channel groups are applied as of the date they were created and do not backfill historical data, so set this up as soon as possible. Also make sure that your AI channel is above the general referral channel in the list as GA4 assigns traffic to the first matching rule it finds, so if the referral comes first, your AI traffic will be absorbed into it before the AI rule takes effect.
What about Google’s own AI features?
This is still a blind spot in Analytics4. Traffic from AI-driven overviews and AI mode doesn’t show up as a separate source, and Google’s own AI decks remain invisible in Analytics4 because Google doesn’t attribute them separately. New Search Console reporting helps here, because it will show you impressions within these features even if GA4 can’t explicitly track clicks. Use both together.
Even with all of this set up correctly, you’re still only seeing part of the image. The sessions you can track live are the minimum AI contribution, and a lot of people see you in the AI answer, don’t click, and then come back later through branded or direct search. That’s why the proxy signals covered earlier in this post are just as important as the traffic data itself.
A poll question worth adding today
There is one agent signal that is disproportionately helpful compared to how easy it is to set up: a single survey question in the sign-up process or post-purchase flow.
Something like this: “Before signing up, had you come across our AI Assistant or AI search experience, such as ChatGPT, Perplexity, Claude, Gemini, or Google AI features?”
Yes / No / Not sure. My choice. One question, no follow-up.
The insight this gives you is important. Users who arrive via brand search but answer “yes” are the invisible influence of AI; They wouldn’t have searched for you without seeing you in the AI answer first. Users who arrive via live traffic and answer “yes” are the mobile-to-AI copy-and-paste group that GA4 completely ignores.
A high “yes” rate among branded and direct visitors is one of the strongest first-party signals that AI is driving real business impact beyond what your analytics can see.
How to report this to leadership
No one is asking you to prove that AI research is responsible for every new lead. The goal is simply to put together an honest monthly picture that helps you make better decisions about where to invest your time and budget.
Here’s an example of what this could look like in practice for a digital marketing agency:
What Google Analytics 4 shows us directly
450 AI platform sessions arrived this month. These visitors spent significantly more time on the site than regular organic visitors and were more likely to view our case studies and service pages. Few, but decent quality.
What our own data indicates
Searches for our agency name were up significantly compared to the same period last quarter, without any paid activity or PR campaign that would explain this. Traffic to our SEO and digital PR service pages has also increased, even though we haven’t run any campaigns specifically mentioning them. This pattern is consistent with people seeing us mentioned somewhere and then coming to check us out. AI answers are the most likely explanation.
What third-party tools show
Tools like Sameweb indicate that the actual number of AI-influenced traffic is probably higher than the 450 GA4 that is recorded, because a lot of AI-driven traffic loses its reference data before it reaches your analytics. On the competitive side, it appears that a competing agency is getting a larger share of AI-referred traffic in our category, which tells us there is ground to make up.
Estimate our planning
Taking branded search growth above our normal baseline, applying a conservative assumption that about a quarter of it is related to AI impact, and checking against the number of new inquiries that told us they found us via the AI tool, we estimate that AI search contributed somewhere in the region of £8,000 to £14,000 of impacted pipelines in the quarter. The assumptions behind this are documented separately.
This last number is a schematic number, not something that can be included in a client report as evidence. The minute you give an estimate as hard income, someone will ask you exactly how you got there, and if the answer includes assumptions, you’ve undermined everything else on the page. Clearly label it as an estimate, show how it works, and it will become a really useful tool for deciding whether you want to invest more in this area.
Three patterns to pay attention to
As you structure data across these layers, some common patterns tend to emerge:
- Scenario 1: Hidden success. AI referral sessions are steady, but brand searches are up significantly, direct traffic is growing, and the “yes” rate to your survey is on the rise. Reading: Vision works. Users watch your AI answers and come in through other channels. Do not cut investment because the observed channel appears quiet.
- Scenario 2: Traffic without valid. AI referral sessions are up, but the conversion rate from AI traffic is lower than your organic benchmark. Reading: You get cited for claims that don’t match your most relevant pages. Check which claims are driving AI traffic and update the landing pages they refer to.
- Scenario 3: Clean condition. Rising AI sessions, brand searches, visible AI conversions, rising survey signals, and growing AI competitive share. Reading: Multiple independent signals all pointing in the same direction. Scale investment, rapidly expand coverage, and maintain validation.
Pull everything together
If you’ve followed this series from the beginning, you’ll now have a framework that covers the whole picture:
- Presence tells you if and how you appear in AI research
- Preparedness tells you why your vision looks the way it does
- Business Impact tells you whether this vision creates business value
Running these three layers together turns what could be three separate audits into one connected diagnosis. You’ll know which lever to pull next; Whether that’s fixing a technical accessibility issue, boosting endorsements in key external sources, improving how you’re described on comparison pages, or simply communicating the impact you’re already making more clearly.
Measuring AI search is not about more dashboards. It’s about connecting what’s happening, why it’s happening, what your value is, and being willing to act on what that tells you.
This series is based on frameworks developed by SEO consultant Aleida Solis, including the three-layer AI search measurement model and AI Quick Search Library Guide. We recommend reading her original work for full technical details.
