Contentsquare separates humans from AI agents: what it changes for your A/B tests and KPIs

Section 1 · The news

Contentsquare relaunches its platform to track humans and agents

Contentsquare is one of the most widely used customer experience analytics tools in e-commerce. It records what visitors do on a website or app: where they click, how far they scroll, where they get stuck. On October 13, 2026, the company announced a reimagined platform, presented as built to understand and optimize “human and agent journeys”.

The new feature that matters here is called human, LLM and agent interactions analytics. According to Contentsquare, it separates traffic coming from LLMs and AI agents and analyzes it alongside human sessions. The stated goal: optimize journeys for visits sent by AI assistants and for agentic commerce, on top of human visitors.

• Automated WebMCP monitoring: it captures the activity of autonomous agents using this new standard, activity that “most tools can't see at all”, according to Contentsquare.
• Multi-touch attribution: it spreads the credit for a sale across every touchpoint, including LLM-referred traffic, instead of giving it all to the last click.
• Conversation analytics: chatbots, voice and text conversations are linked to friction points on the site and in the app.
• Sense, the platform's AI layer, can now form hypotheses, launch tests and validate their impact, with a human approving.

Contentsquare also announced deeper integrations with testing and personalization tools, including Dynamic Yield, Optimizely and Kameleoon. The platform is in beta for selected customers. Broader access will follow in phases, with no specific date.

Section 2 · The context

Billions of AI agent requests, with e-commerce on the front line

Why now? Because AI agents are already here, in volume. DataDome, a company that specializes in bot detection, published a report on this traffic in March 2026. Its network recorded 7.9 billion AI agent requests in January and February 2026, 5% more than in the last quarter of 2025.

The report looks closely at agentic browsers: browsers driven by an AI that open pages the way a person would. Their traffic is concentrated in the industries where transaction data is most valuable. E-commerce and retail come first.

DataDome also cites one of its customers, where agentic traffic made up 9.75% of total traffic over 30 days. It is a single case, but it gives an order of magnitude: on a highly exposed site, nearly one visit in ten can come from an agent.

Bot, LLM, agent: what are we talking about? A crawler browses the web to collect pages, for example to train or feed an AI model. An AI agent acts on behalf of a person: it looks for a product, compares prices, sometimes fills a cart. An agentic browser is a browser that the AI drives itself. That last case is the trickiest for your statistics, because it loads the page like a human and can therefore trigger your measurement tools.

Section 3 · The problem

An agent does not look at your promo banner

A human lands on a product page, sees the promo banner, scrolls through the photos, reads a few reviews, hesitates, comes back the next day. An AI agent does something else: it reads the page content, picks up the price, availability and specs, compares with other sites and leaves. It does not scroll and does not react to your visuals.

“An agent reads the product page, compares the price and leaves. In your numbers, it looks like a visitor who did not buy.”

As long as these visits stay mixed with human visits, they distort the metrics you look at every week. The most common effects:

Section 4 · A/B tests

In an A/B test, every agent counts as a visitor who does not buy

An A/B test compares two versions of a page on two randomly drawn groups of visitors. If some of those visitors are agents, two things can happen.

SRM, a first warning sign SRM (sample ratio mismatch) is an abnormal imbalance between the groups of a test: you planned 50/50, you get 52/48. It often means one type of traffic is counted differently on one side. Bot and agent traffic is one of the first causes to check.

This is where separating humans and agents becomes useful. If the analytics tool can recognize an agent session, you can remove it from the results, or study it on its own. Contentsquare integrates with testing tools such as Kameleoon, Optimizely and Dynamic Yield, which makes it possible to cross test results with the observed behavior of visitors.

Section 5 · The questions

Five questions to ask your team this week

For an e-commerce leadership team, the point is to know whether its numbers account for this traffic. The questions to ask:

1. What share of our traffic is not human? Per month, per page, and especially on product pages.
2. How do we spot it? Declared bots, behavior, bot-protection tool, analytics tool: each source gives a different number, so you need to know which one you are looking at.
3. Is it excluded from our A/B tests? And since when. A test launched before that filter deserves a second look.
4. Are our KPIs calculated on humans only? Conversion rate, bounce rate, average order value: you want the human version and the total version, side by side.
5. What do agents do on our site? Which pages they read, which prices they pick up. That is what will be compared with competitors when a customer asks their AI assistant to choose.

The last question matters as much as the others. Keeping agents out of KPIs makes your numbers more accurate. Studying them separately tells you what AI assistants take away from your offer, and so what they will tell your customers.

Section 6 · What's next

Three things to watch in the coming months

• The detection method. Contentsquare's announcement does not explain how an agent is recognized. It is the first question to ask during a demo.
• The access timeline. The platform is in beta for selected customers. Check when your contract will give you access, and to which features.
• The WebMCP standard. If it spreads, agents will increasingly go through functions the site exposes to them, rather than reading pages. Measuring agent traffic will change again.