Microsoft Clarity can now show what AI agents do on your site. A guide for CPG brands.

If an AI agent tries to buy your protein bars and gives up at the flavor picker, what do you see? On most brand sites, nothing. The session looks like a short visit with no purchase. Nobody files a complaint, and the shopper who asked the agent simply ends up with someone else's bar.

On November 4, Microsoft Clarity announced tools aimed at exactly that blind spot. In a post by Nabil Naghdy, the free behavioral analytics product introduced an Agent Activity dashboard, simulated agent tests, wider reach for its Brand Agents shopping assistant, and an AI analyst called Ask Clarity.

This guide is about the first two, because they answer a question every DTC brand now has: are agents shopping on my site, and can they finish?

What Microsoft announced

Clarity describes three overlapping web journeys: people browsing, AI recommending, and agents acting on a shopper's behalf. The new features map to those.

Feature	What it does	Availability
Agent Activity dashboard	Shows how agentic traffic changes over time, which supported agents visit, and what they do: searches, form submissions, log-ins, downloads, bookings and checkout	Announced as available globally to all Clarity users
Simulation assessments	Test whether an agent can complete a key task; report completion, time taken, number of interactions and where to improve	Starting with commerce and travel scenarios
Brand Agents	On-site AI shopping assistant; for eligible WooCommerce and Shopify merchants, products can also appear in Copilot, Bing, MSN and more	Beta, websites only
Ask Clarity	Answers questions about site behavior with links to session recordings as evidence	Closed beta for retailers, with a waitlist

Sources: Microsoft Clarity and Microsoft Learn.

The headline number in the announcement is about growth. "Weekly sessions from AI agents nearly doubled in three weeks to almost 2 million," Microsoft writes, based on its analysis of 5.23 million agentic sessions across 230,511 Clarity projects between August 13 and September 9, 2026. It detected agents from OpenAI, Google and Anthropic, and names Muse and Dots as examples of agents "shopping, booking, and filling out forms."

Why this matters for a CPG brand

Two recent studies explain why we would bother.

The first is about who is buying. Fraud-prevention company Forter looks at agent orders across its merchant network. It found that agents which drive a store's own checkout pages, rather than a special protocol, went from 1% of agentic orders in late July to 70% in the week of October 12. We covered the details in our Forter post.

Show the data

Period	Value
Late July 2026	1%
Week of Oct 12, 2026	70%

Source: Forter, AI Agent Order Data (Oct 23, 2026). Forter groups Muse, Instinct and ChatGPT Agent among the agents that drive merchants' own checkout pages.

In other words, most agents are now using the same product pages and checkout your customers use. If those pages trip a person up occasionally, they may trip an agent up often.

The second is about where agents fail. In General Machines' CartTrace-100k sandbox of more than 101,000 simulated shopping journeys, reading product facts failed in 37.5% of journeys and clarifying the shopper's needs in 28.9%, while cart, checkout and payment each failed in under 1%. The agent also added items straight from search results, without opening a product page, in 37.6% of journeys. That was a lab test with stand-in stores, not real ones, but the lesson carries: agents mostly stumble on information, not on the payment form.

Clarity's tools are a way to check both things on your own site, with your own pages.

Setting it up

If your site already runs Clarity, the Agent Activity dashboard should appear in your project. If it doesn't, Clarity installs with a tracking script or through platform apps for Shopify, WordPress and WooCommerce. Microsoft says the product is free.

A few setup notes before you rely on it:

• Check consent and masking settings. Clarity records sessions. Make sure your cookie banner and privacy notice cover it, and that sensitive fields like checkout details are masked.
• Decide who owns the dashboard. In small brands, analytics belongs to whoever has time. Give agent traffic a named owner, even if it is a monthly look.
• Keep your existing analytics. Clarity sits alongside GA4 rather than replacing it. Our GA4 guide covers how to see AI referrals there, which is a different signal: people arriving from an assistant, not agents acting for them.

What to look for in the dashboard

The dashboard shows which agents arrive and what they do. For a CPG brand, we would read it in this order.

Which agents and how many

Note the agents by name and their weekly sessions. A growing line from one agent usually follows a product launch on that platform. Compare it with what you know: did Muse or ChatGPT ship a shopping feature that week?

Where they land

Do agents arrive on product pages, collection pages or search results? If they start at your site search, the quality of your search results matters more than you think. CartTrace's finding that agents often add to cart straight from a results list suggests the card itself, with its name, size and price, may be all an agent reads.

What they do next

The dashboard lists actions such as searches, form submissions, log-ins and checkout. The ratio between "visited a product page" and "reached checkout" is the number to watch. A low ratio isn't automatically a problem, since many agents are researching rather than buying. A ratio that drops after a site change is.

Running a simulation

Simulation assessments are the more practical half. Instead of waiting for real agents to fail, you ask Clarity to run a task and see what happens. Clarity reports whether the task was completed, how long it took and how many interactions it needed.

Pick tasks a shopper would actually hand an agent. For packaged goods, we would start with three:

1. The reorder. "Buy the 12-pack of the chocolate flavor." This tests variant selection, pack sizes and whether the default option is the one people want.
2. The constrained pick. "Find a gluten-free option under $30 and add it to the cart." This tests whether filters and product facts are readable.
3. The subscription. "Set up a monthly delivery." This tests subscription widgets, which are often built as custom pop-ups.

When a task fails or takes many steps, the usual suspects on CPG sites are familiar: an email capture pop-up on first visit, an age gate, a flavor picker built from images without labels, a quantity selector that needs a hover, a forced account creation, or a bundle builder that only works with drag and drop. Tip 33 covers whether to let agents buy at all; this is the test of whether they can.

Fix one blocker, run the same task again, and record the result. Then run all three after any theme update, checkout change or new app install. We have a longer checklist in our holiday checkout audit.

The Brand Agents question

Brand Agents is a separate decision. It puts a Microsoft-powered shopping assistant on your site, grounded in your catalog, shipping and returns policies and brand voice. Microsoft's documentation says that when you accept the terms, your catalog is published to Microsoft Merchant Center and made discoverable through Bing, Copilot and other AI experiences. You can also use it only for that discovery, without showing the assistant on your site.

That is attractive for a small Shopify brand that has never set up Microsoft Merchant Center. Before you turn it on, check two things. First, read what data is shared and how the catalog is used. Second, make sure the catalog is accurate, because whatever is in it is what Copilot will describe. Brand Agents is still in beta and available for Shopify and WooCommerce websites only.

Limits to keep in mind

• Clarity can only report on agents it can detect. Microsoft refers to "supported" agents, and some agents present themselves as ordinary browsers.
• The growth figure covers Clarity's own projects over four weeks. It is a signal of direction, not a market estimate.
• A simulation is one agent's attempt. A real agent built by another company may behave differently. Treat a passed test as a good sign, not a guarantee.
• The dashboard shows what agents do on your site. It doesn't show what assistants say about your products before anyone visits.

That last gap needs a different check: asking the assistants the questions your shoppers ask and reading the answers. Some teams use tools such as SearchPanel from XENA Intelligence for that side; a monthly list of questions in a spreadsheet covers the basics.

A monthly routine

Put together, a small brand can run this in under an hour a month:

1. Open the Agent Activity dashboard and note agents, sessions and landing pages.
2. Compare product-page visits with checkout reached for agent sessions.
3. Run the three simulations: reorder, constrained pick, subscription.
4. Fix the first blocker you find, and rerun.
5. Note any agent that appeared for the first time, and check what launched that month.

Agents are already shopping on store pages built for people. For the first time, a free tool will show you some of what they see.