Agent Analytics
AI assistants and agents already reach your site. Most of it is hiding in traffic you already report on.
Agent analytics identifies measurable activity from AI assistants and agents, separates it from the rest, and shows how it behaves and what it contributes to commercial outcomes.
How do AI assistants show up in my analytics today?
Some as recognisable referrals. Much of it as direct traffic.
It depends on the information available when someone arrives. Some referrals from AI assistants can already be identified. Others appear as direct traffic, because the information needed to identify the source is missing or isn't passed in a form your analytics setup recognises.
We assess the available signals, identify traffic from AI assistants where the data supports it, and establish what remains unknown. If your setup isn't capturing what's needed, that becomes a data collection question.
And agents that browse the site themselves?
They run a real browser, fire your tags, and are counted as people.
A browsing agent executes your JavaScript like any visitor, so it lands in your client-side analytics. A filter that only knows "human" or "bot" has no place for it. In a 2026 study by TU Munich, standard two-class models classified agents as human in 30 to 39 percent of sessions. That traffic then sits in the denominator of your conversion rate and your A/B tests.
Usually the fix is small. Agents behave differently from people, for example in how they move and click, and a few extra signals in your data layer are often enough to separate them.
A conversion rate with unknown agents in it is a number nobody can defend. Separate them first, then decide what they're worth.
How much traffic are we talking about?
There is no useful universal percentage. It depends on where you measure, and what you call non-human.
1.7%
Named AI user agents, counted per request, across all web traffic *DataDome, June 2026
5.3%
AI agents on 32,000 European B2B sites, classified on IP behaviour *Leadinfo, Q3 2026
>50%
All non-human traffic on the internet, counted at network level *Cloudflare, July 2026
None of these is wrong, and none of them is your number. Traffic volume also says little about value: a small source can account for a very different share of transactions or revenue.
What can you actually measure?
Where the data supports it: who sent the visit, where it went, and what it led to.
Identifiable AI traffic can be separated from broader categories and analysed alongside other sources: which assistants send traffic, which pages those visits reach, whether they behave differently, and whether they lead to conversions or revenue. The identified activity can then become part of your wider attribution analysis.
Where the evidence identifies an AI source, we report it. Where it doesn't, we don't assign one based on an assumption.
Where our work stops.
Our role is the measurement layer.
Agent analytics starts when activity from an assistant or agent becomes observable on your site or app. Visibility within AI-generated answers is content and search work, which specialist teams within LevelUp Group handle.
Frequently asked questions
-
Do we need a new analytics platform to measure this?
Not necessarily. The first step is to assess what your existing data collection captures, how traffic is currently classified, and what additional measurement is needed.
-
Can you always identify which AI assistant sent the traffic?
No. Sources can be identified where the data contains enough information. Where it doesn't, we don't assign one based on an assumption.
-
Do AI agents show up as bots in GA4?
Often not. Agents that browse with a real browser run your tags like a visitor and are usually counted as people. Separating them takes additional signals in your data collection.
Find out what is already happening
Some of it may be sitting in traffic you already report on. We'll assess your data, identify what can be measured and set a baseline with a date on it.
Contact details
Cloud Nine Digital
Johann Siegerstraat 24
1096 BH
Amsterdam, Netherlands
T: +31 (0)20 787 1323
E: info@cloudninedigital.nl
Need something else?
Please do reach out. We are happy to help!