Contentsquare Experience Platform review: agent visits finally get their own line
Agents and AI referrals are already in your traffic, mixed in with people. Contentsquare's new platform promises to pull them apart. It's in beta, priced on request, and worth a close read before anyone signs.
Verdict: promising for mid-size and large stores that already feel AI referrals and agent visits muddying their numbers, and that have someone to own experiments. As of October 16, 2026 it is a selected-customer beta with custom pricing for the Pro and Enterprise tiers, and its most interesting claims (WebMCP agent monitoring, Sense running tests end to end) are untested in public. Ask for a pilot on your own traffic.
Who should read this
If you run an online store, some of your traffic this year isn't people. Some visitors arrive from an AI answer. Some are agents browsing, comparing and occasionally adding to cart for someone. Most analytics setups lump all of it together, and then a team argues about why conversion moved.
Contentsquare announced a rebuilt Experience Platform on October 13, 2026 aimed squarely at that problem. It's available in beta to selected customers, with broader access promised in phases. This review is for the person deciding whether to ask for a seat in that beta, or whether to wait.
What Contentsquare says it shipped
From the launch announcement, the new pieces are:
- Human, LLM and agent interaction analytics, which separates LLM and agent traffic from human sessions.
- Conversation intelligence and chatbot analytics, which pulls voice, text and chatbot conversations into the journey as events and links them to friction on the site or in the app.
- Multi-session journey analysis across sessions, channels and devices, plus automated WebMCP monitoring for agent activity the company says most tools can't see.
- Multi-touch attribution that credits conversions across touchpoints, including AI-referred traffic.
- Sense, an AI layer that flags friction, estimates financial impact, and writes emails, tickets and notifications into your team's tools. Sense can now form a hypothesis, design and code a test, launch it and validate the result. In the company's words, all you need to do is approve.
- Benchmarks built on ten years of zone-level data, and MCP connections that bring the data into Claude, Microsoft Copilot, ChatGPT, Figma, Dust, Cursor, VS Code Copilot and Amazon QuickSight.
It also announced deeper links with Dynamic Yield, Optimizely and Kameleoon for teams already testing with them.
How the work moves
Questions to settle before the beta
- Split
Can we see our own agent and LLM visits apart from human sessions?
- Coverage
What do we see for agents that never use WebMCP?
- Cost
How do agent events count against the 3.6M server-side events and the 150-per-session line?
- Approval
Who signs a Sense experiment that changes checkout or pricing display?
A conversion rate that mixes people and agents is two numbers pretending to be one.
Green flags
It treats agents as their own audience. A conversion rate that mixes people and agents is two numbers pretending to be one. Splitting them is the first honest step to knowing whether your product pages work for either. If agent visits are rising and human conversion looks soft, you need this split before anyone redesigns anything.
Attribution includes AI sources. If an AI answer sent someone who bought three days later on their phone, last-click reports give that sale to whatever happened last. Multi-touch attribution that names AI-referred traffic at least lets you see the path.
There's an approval step in the experiment loop. Sense running a test end to end is a lot of autonomy. The announcement keeps a human approval in it. Keep that, and make sure the approver can see exactly what the variant changes.
The data can leave the platform. MCP connections into the tools your team already uses mean the analyst doesn't have to live in one more dashboard. The public pricing page lists MCP even on the Free plan (up to 300 tool calls a month).
Pricing for the old product is public. The Free plan is $0 for up to 200,000 monthly sessions, and Growth starts at $39 a month billed yearly. That's a low-risk way to see whether your team actually uses this kind of tool before a bigger contract.
Red flags
Beta, for selected customers. The press release says beta today and phased broader access. Contentsquare hasn't said when each capability will be generally available. Plan as if the feature list will move.
The agent features sit on custom-priced tiers. On the pricing page, AI agent site readiness and interaction analysis appears on Pro, which is priced on request, and Enterprise is custom too. You won't know the cost of the part you care about until you're in a sales conversation.
Read the event terms. The beta service schedule (version 2026.1) includes 3.6 million server-side events a year. Above that, each 30 server-side events are charged as one session. In months where your average tops 150 events per session, every extra 150 events is charged as another session. Agents can be chatty. Ask how agent traffic counts against those limits before you turn everything on.
WebMCP monitoring is a claim to test. The release says automated WebMCP monitoring captures agent activity most tools can't see. It doesn't say what share of real agent traffic uses WebMCP today, or what you see for agents that never touch it. Ask the vendor to show you your own agent visits, split the way they promise.
Sense writing and shipping tests touches your storefront. A variant that changes checkout, pricing display or shipping copy needs the same review as any other code change. If your developer isn't in that approval, it isn't really an approval.
The proof is vendor-chosen. The launch quotes a Crocs executive who likes the platform. That's a testimonial, not an independent test, and there were no public independent reviews of the beta as of today.
Decision guide
Green flags and red flags at a glance
| Consideration | Green flag | Red flag |
|---|---|---|
| Traffic | Agents and LLM visits split from people. | WebMCP coverage of real agent traffic is unstated. |
| Experiments | Human approval stays in Sense's test loop. | Auto-coded variants touch your storefront. |
| Price | Free and Growth plans are public. | Agent features sit on custom-priced tiers. |
| Availability | Beta open to selected customers now. | No dates for general availability. |
How it compares
Two realistic alternatives for most stores:
Stay where you are and add segments. Most analytics tools let you build segments for known AI referrers and suspected agent user agents, and your server logs show the rest. It's rougher and more manual, but it's yours and it costs engineering time rather than a contract.
Keep your testing tool, add the split later. If you already run Optimizely or a similar tool, the human-versus-agent question can be answered first in analytics and fed into tests you already know how to run. Contentsquare's own partner integrations suggest it expects many buyers to do exactly that.
Verdict, and what to ask on the demo
As of October 16, 2026: worth requesting beta access if you're already a Contentsquare customer, or if agent and AI-referred traffic is big enough that your weekly numbers are arguing with each other. Not worth a rushed annual contract for a platform still in phased beta.
On the demo, ask four things. Show us our own agent visits split from human sessions. Show what's visible for agents that don't use WebMCP. Show how agent traffic counts against the event terms. Show exactly what an approved Sense experiment changes on the page, and who else signs it.
Our ecommerce work is about getting the storefront, inventory, payments and fulfillment to tell one story. If you want help reading a contract like this one, or setting up the human-and-agent split in the analytics you already have, talk to us.
Related work: Paddle8.
We build alongside the people doing the work.
Bring us the problem you’re thinking about.