Microsoft Clarity measures what AI agents do on websites: checkout in 25% of their sessions

Microsoft Clarity published a study on October 29, 2026 on what AI agents do once they reach a website.¹ It compares the actions recorded during sessions involving agents from OpenAI, Google and Anthropic with purely human sessions, on the same group of sites.¹ Clarity is advertised as a free website analytics tool: a business that already uses it can check right now whether this traffic affects it, through the new Agent Activity dashboard announced in the same study.¹ This article covers the published figures, the limits Clarity states itself, the ones PPC Land found in its method, and how to spot these sessions in your own analytics.

01What Clarity measured

The study relies on an automatic classification of sessions, not on a measurement declared by the agents themselves.¹ It details its protocol, the measurement tool behind it, and the limits it acknowledges.

The study's protocol

Clarity analyzed more than 5.2 million sessions classified as involving an agent, across 230,511 Clarity projects, between August 13 and September 9, 2026 (UTC).¹ Sessions were identified by detecting agentic browser providers across the OpenAI, Google and Anthropic ecosystems.¹ Clarity states that an agentic-classified session may also include a human handoff, and that it cannot currently separate the two parts within a single session: the findings therefore describe activity recorded during sessions involving an agent, not necessarily actions completed autonomously by that agent.¹

Smart Events, the measure behind the percentages

The compared actions are Smart Events, a Clarity feature that groups clicks and session signals into readable actions such as a payment, a booking or a login.¹ Clarity automatically detects nine event types, with no setup: purchase, add to cart, begin checkout, contact us, submit form, request quote, sign up, login and download; a project admin can also create up to twenty custom events, code-free.² Clarity is advertised as free, with no traffic limits.¹

The table comparing agents and humans

For each of the ten most frequent actions, Clarity divided the number of sessions that recorded the action by the total number of sessions in the matching group.¹ The table below reproduces the five highest-ranked rows.

02Checkout, bookings, and the gap that raises questions

The two action profiles largely overlap: eight of the ten actions appear on both published lists.¹ What changes is their weight, and PPC Land also flags a gap in the figures themselves.

Checkout and bookings, more common in agentic sessions

Checkout is the action most often recorded in agentic sessions, at 25.21%, against 14.65% in human sessions, where outbound clicks lead at 26.99% against only 14.28% for agents.¹ Booking an appointment, at 8.42% of agentic sessions, and checking availability, at 2.41%, make the agentic top ten without appearing in the ten most frequent human actions.¹³ Clarity sums up that the clearest difference between the two profiles isn't a different set of actions, but how heavily those same actions are weighted.¹

5.2 million sessions in the headline, 1.2 million in the percentages

PPC Land recalculated the base behind each published percentage, by dividing each count by its matching percentage: all ten rows of the agentic table return, to within rounding, the same base of about 1.22 million sessions, not the 5.2 million announced in the study's title and methodology.³ By that calculation, at least 78% of the 5.2 million agentic sessions recorded none of the ten most common actions.³ The study does not explain the gap, nor whether the human percentages rest on the same restricted base, which makes the comparison between the two groups hard to verify.³

The agents the study doesn't count

The study gives no breakdown by the three ecosystems it names, nor by industry, country or device.³ Its measurement window closed on September 9, 2026, one day after Meta released Muse in the US, and it covers neither Meta nor Perplexity.³ For a merchant site, an agent that reaches a checkout page still depends on the data the site feeds it through its product feed, such as the family of conversational attributes Google designed for AI-driven surfaces, detailed for product questions and answers.

03How a small business spots these agents in its own analytics

The Agent Activity dashboard doesn't measure the same thing as Bot Activity, Clarity's other dashboard dedicated to crawlers: one reads the script placed on the page, the other the server's logs.³ The diagram below summarizes that difference before detailing the levers available to a small business.

The Agent Activity dashboard, free inside Clarity

The Agent Activity dashboard is advertised as generally available for Clarity projects that receive agentic traffic, meant to identify that traffic and show how agents interact with a site.¹ For ecommerce and travel and hospitality businesses, Clarity also mentions simulations meant to surface opportunities to improve the experience, without detailing how they work.¹

A name close to another Clarity dashboard

PPC Land notes that the name Agent Activity Dashboard is close to that of an existing Clarity dashboard, the Agent Dashboard, which reports on conversations with a brand's own on-site agent rather than on outside agents arriving at the site.³ The two dashboards measure two different things, under similar names.³

Spotting agents through other means

Clarity's dashboard only covers the three ecosystems it detects.³ A site can also look for an agent's trace in its own server logs, through the bot's identifier, its IP address or a Web Bot Auth signature when present, a method detailed in our article on Google-Agent and the UCP protocol. That check belongs in a regular audit of a site's analytics, alongside the other checks of an SEO checklist.

04Frequently asked questions