Microsoft Clarity analyzed 5.23 million sessions that involved AI agents and found that Checkout was the most frequently recorded action, appearing in 25.21% of them, against 14.65% of human sessions.
What Clarity measured
The analysis, which Microsoft published on the Clarity blog on October 29, covers sessions across 230,511 Clarity projects between August 13 and September 9, 2026. Clarity identified the sessions through detected agentic browsers from three operator ecosystems: OpenAI, Google and Anthropic. It then counted which Smart Events fired in each session. Smart Events are Clarity’s labels for business actions such as checkouts, form submissions, appointment bookings, searches, downloads and logins.
Microsoft attaches two limits to the numbers, and both matter to anyone who plans to quote them. An agentic session may include a human handoff, and Clarity cannot yet split the agent part from the human part. The company says the findings describe what was recorded during sessions involving agents, and that an agent may not have completed every recorded action on its own. The counts are also independent of one another across different sites and industries, so they do not form a conversion funnel.
With those limits stated, the ranking for agentic sessions was Checkout (306,621 sessions, 25.21%), Submit form (241,391, 19.85%), Outbound click (173,655, 14.28%), Book appointment (102,401, 8.42%) and Search (87,096, 7.16%). Login, Show more, Check availability, Add to cart and Download filled out the top ten.
Where agent sessions differ from human ones
Human sessions in the same cohort of projects ranked differently. Outbound click led at 26.99%, followed by Submit form at 18.56% and Checkout at 14.65%. Microsoft’s reading is that the two groups perform many of the same actions and the difference is in the weighting. Form submissions were close, at 19.85% for agentic sessions and 18.56% for human ones. Search made up 7.16% of agentic activity and 4.79% of human activity. Outbound clicks made up 14.28% of agentic activity, about half the human share.
For a marketing team, the outbound click gap is the number to read first. Much of the measurement stack was built around a person clicking away to an ad, a partner or a social profile. Clarity’s data says agent visits spend a smaller share of their recorded actions on that and a larger share on task actions such as checkout and booking. A dashboard that blends the two groups will describe neither.
Reading the event names carefully
The two top-ten tables use event names that a reader can easily merge. The agentic list carries Checkout at 25.21% and Add to cart at 2.33%. The human list carries Add to cart at 3.31%, Order success at 1.79% and Begin checkout at 1.62%, with Checkout in third place at 14.65%. Checkout, Begin checkout and Order success appear as separate Smart Events, and the study does not define how Clarity tells them apart. A Checkout count in the agentic ranking therefore cannot be read as completed orders. It records that the session reached the event Clarity labels Checkout.
The rest of the agentic list is also task-shaped. Book appointment (8.42%), Check availability (2.41%) and Login (6.83%) sit alongside Show more (4.31%) and Download (1.69%). Microsoft says these results show that agentic sessions are not limited to passive browsing or information retrieval and that they reach interactions associated with commercial and task-oriented intent. The study pools every industry in the cohort, and Microsoft’s methodology note says the counts span different industries, so the profile on any one site can differ from the pooled ranking.
What Microsoft shipped on November 4
On November 4, Microsoft announced three additions to Clarity, available globally: Agentic Analytics, Brand Agents and Ask Clarity. It said weekly sessions from AI agents nearly doubled in three weeks to almost 2 million, and the October 29 study puts the rise in agentic session volume at 88% over the same three weeks.
The Agent Activity dashboard shows how agentic traffic changes over time, which agents visit a site and what they do once there. Simulation assessments let a team test whether an agent can complete a task before a real customer does, and report whether the task finished, how long it took and how many interactions it needed. Microsoft says simulations start with commerce and travel scenarios. The study describes the dashboard as generally available for projects that have agentic traffic.
Ask Clarity answers plain-language questions about site behavior and links its findings to session recordings. It is in closed beta for retailers. Microsoft cites its July to August early-access feedback survey, which a footnote puts at 16 participants, and reports that 88% said they discovered something new about their business or website. One of them, Kelly Wellman, eCommerce Manager at Pela Case, said: “I use this feature to dive deeper into user behavior, especially around new launches.” She added that it “showed that 80.2% of pageviews on our PowerBank landing page ended with no onward click.”
The sample is small and the quote comes from the vendor’s own announcement, so it describes what the feature is meant to do rather than how it performs across stores. It does show the use Microsoft has in mind: an analyst asks why a launch page leaks and gets recordings as evidence.
The catalog that Brand Agents publishes
The second addition carries a data decision that sits outside analytics. Microsoft says eligible WooCommerce and Shopify merchants who use Brand Agents can have their products appear across Microsoft Copilot, Bing and MSN when shoppers ask AI for recommendations, and that shoppers complete the purchase on the merchant’s own website. For merchants who already have Brand Agents installed, the announcement says no additional catalog upload or configuration is required.
The documentation spells out the mechanism. The Shopify installation guide says that by accepting the Brand agents terms, a merchant consents to having its catalog published to Microsoft Merchant Center, which lets Bing and Copilot discover the products. The Brand agents overview lists catalog discovery among the key capabilities and marks the product as in beta and limited to websites, with mobile apps excluded. The announcement also lets a merchant choose to offer the on-site Brand Agent experience or to use Brand Agents only to make products appear across AI experiences.
That choice puts a merchandising and data-sharing question on a marketing leader’s desk that used to belong to whoever installed an analytics tag. A team that adopts the free analytics product to watch agent traffic is also one setting away from syndicating its catalog into a Microsoft shopping surface.
Agents on the buying side
The agent traffic Clarity counts has a counterpart on the user side. Meta’s Muse for Small Business announcement in September described a personal AI agent that connects to tools a business already runs on, including Klaviyo, Shopify, Stripe, Canva and Slack, as well as Facebook and Instagram business accounts. Meta said nothing publishes, sends or spends without the user’s approval. We covered the Klaviyo and Shopify connectors in Meta’s Muse Agent Now Connects to Klaviyo and Shopify.
Cameron Adams, co-founder and CPO of Canva, said in the announcement: “Muse is an exciting leap towards agents that genuinely help us work, create and make our lives easier.” Whether the agent arriving at a checkout page came from a shopper’s assistant or from a business owner’s own tooling, Clarity’s classification only records the operator ecosystem. The study lists OpenAI, Google and Anthropic, and the announcement says supported agents for Agentic Analytics are limited to those three operators, so an agent from another provider would not appear in the dashboard.
What it means for the marketing leader
The data is early, vendor-published and drawn from sites that use one analytics tool. It still gives a marketing team five concrete things to do this quarter.
First, split reporting. Ask the analytics owner to report agentic sessions and human sessions separately for the same period, using the same Smart Event definitions. The Clarity cohort shows different mixes of actions, so a blended conversion rate averages two behaviors. Our earlier piece on measurement disputes moving from the number to the definition makes the same point about writing the definition down.
Second, record what the handoff caveat means for your own sites. Because Clarity cannot separate agent and human portions of one session, a Checkout event in an agentic session is not proof that an agent bought something. Treat the 25.21% as the share of agentic sessions that recorded the event and keep it out of revenue attribution until your team can see the order records behind it.
Third, run the simulation on your checkout and booking paths if you sell products or appointments. Microsoft’s simulations report completion, time and interaction count for a task, which are figures a product owner can put a target against. Pick the one path that earns the most revenue and test it first.
Fourth, decide on the catalog before someone else does. The terms for Brand Agents publish the catalog to Merchant Center. In our reading, that decision belongs with the people who own pricing, promotions and channel strategy, and it should be made on purpose, with the change logged, rather than arriving through an installation step. If the answer is yes, check what the shopper sees on the merchant site after the click, since Microsoft says the purchase completes there.
Fifth, put a date on the next read. The October 29 study covers four weeks ending September 9, and Microsoft’s November 4 post says weekly agent sessions had nearly doubled in three weeks, so a figure from this study will be out of date by the time a quarterly plan is approved. Ask the analytics owner to re-run the same comparison monthly and keep the definitions fixed between runs. Clarity is free, and the dashboard needs agentic traffic to show anything. A site with none will see an empty view, which is itself a finding about how far agents have reached that site. A site with some can put numbers next to the argument over how much of its traffic is no longer a person browsing.
Source: Microsoft Clarity
