Microsoft Clarity’s analysis of 5.2 million website visits points to a clear behavioral shift in the AI economy. AI agents reach checkout pages in 25% of sessions, well ahead of the 15% rate seen with human visitors. As large language models move from simple conversation to handling actual tasks, the data offers an early empirical view of how these autonomous systems interact with commercial sites.
Quick Take
- The data: Microsoft Clarity logged AI-agent interactions across 230,000 projects, finding agent sessions have a significantly higher presence on checkout, booking, and form pages than human traffic.
- The catch: Reaching checkout does not necessarily mean autonomous buying; researchers and analysts caution that many of these sessions likely end in a “human handoff” to finalize the transaction.
- The infrastructure shift: The rise of transactional AI traffic is accelerating the need for new standards like the Universal Commerce Protocol (UCP), which allows agents to read inventory and build carts programmatically.
- Why it matters now: For developers and brands, the web is no longer just rendering HTML for human eyes-it must now serve structured, agent-readable commerce endpoints to capture AI-driven revenue.
What happened
Microsoft Clarity examined 5.2 million sessions tagged as agentic, covering systems built on OpenAI, Google, and Anthropic models. The data shows focused, task-driven movement. Checkout pages appeared in 25.21% of agent sessions versus 14.65% for humans. Form submissions hit nearly 20% in agentic traffic, while bookings showed up in more than 8%.
Analysts note key limits in the numbers. A 25% checkout-page rate does not translate to a matching autonomous purchase rate. Event tracking mixes steps, and current tools cannot reliably separate an agent that completes a purchase from one that simply prepares a cart before handing control to a person. The effective base for checkout events also sits closer to 1.2 million sessions, underscoring how legacy analytics struggle with the non-linear paths agents take.
Why it matters now
Frontier models now target multi-step agentic work across the open web, yet most sites still rely on visual layouts, CAPTCHAs, and tangled DOM structures that block smooth machine navigation. The Clarity numbers show agents pushing hard to close that gap and bring purchase intent straight to merchants. That pressure is driving new commerce standards. Work on the Universal Commerce Protocol (UCP), the Model Context Protocol (MCP), and Agent-to-Agent payment systems aims to let agents query stock and complete transactions through clean APIs instead of scraping HTML.
For the wider AI field, the pattern confirms that heavy spending on tool use and reasoning is producing measurable commercial behavior. If agents become the main entry point for digital sales, the teams that control the agent platforms-whether Google’s Gemini integrations or OpenAI’s operator tools-stand to shape how transactions begin and end.
Who is most affected
This shift fundamentally alters the roles of analytics, ecommerce, and AI product teams, who must now learn to segment, serve, and secure non-human traffic without misattributing business metrics.
| Stakeholder | Impact | Why |
|---|---|---|
| Analytics & Marketing | Misattribution risks | Traditional analytics conflate AI agent session prep with human intent, skewing conversion funnels, ROI models, and lead-gen metrics. |
| Ecommerce & DevOps | Infrastructure restructuring | Sites must adopt agent-readable structured data and protocols (like UCP) while simultaneously preventing bot fraud and navigating CAPTCHA blocks. |
| LLM Providers | Monetization opportunities | Companies that successfully enable seamless “human handoffs” or autonomous AI payments will capture the lucrative agentic commerce layer. |
About the analysis
Researched and written with AI assistance under the editorial responsibility of the i10X News team. The title image is an AI-generated illustration. Read our AI Use Policy. Spotted an error? Request a correction.
This independent, research-based i10x analysis synthesizes first-party telemetry data and industry scrutiny to contextualize the rise of agentic commerce for CTOs, developers, and AI strategists.
i10x Perspective
The web is quietly splitting into two layers: one visual layer built for people and another structured, API-first layer built for models. Clarity’s numbers supply the first large-scale sign that agents are arriving ready to transact, not merely to gather data. The real test ahead is whether fraud controls, analytics platforms, and payment systems can adjust quickly enough to handle this traffic-or whether they will treat the next wave of commerce as a threat to be blocked.
