AI agents are getting to your booking form. Can your analytics tell?

Cloud AI cites who it can see. Now some of it shows up on the page and tries to finish the job.

On October 29, 2026, Microsoft Clarity published What AI Agents Actually Do on Websites. It's first-party data from Clarity's own analytics: more than 5.2 million agentic-classified sessions across 230,511 Clarity projects, recorded between August 13 and September 9, 2026. Sessions were flagged using "detected agentic browser providers across the OpenAI, Google, and Anthropic ecosystems."

The headline isn't that agents are coming. It's where they show up once they arrive: checkout, forms and booking.

The short version

• Agents reach the money pages. Checkout was the most-recorded action in sessions involving agents (25.21%), then form submits (19.85%), outbound clicks (14.28%) and appointment booking (8.42%).
• Humans weight it differently. In human sessions on the same projects, outbound clicks led (26.99%) and checkout was 14.65%. Form submits were close: 19.85% agentic vs 18.56% human.
• Agents search your site more. On-site search was 7.16% of agentic activity vs 4.79% for humans.
• It's growing fast in Clarity's data. Clarity says agentic-session volume "increased by 88% in just three weeks."
• It's not proof an agent finished anything. Clarity says sessions "may also include a human handoff" and the numbers are not "proof that an agent independently completed a task."

What Clarity actually measured

Clarity's Smart Events are the business actions it already tracks: checkouts, form submissions, appointment bookings, searches, downloads, logins. Clarity counted how often each one appeared in sessions involving an agent and compared that with human-only sessions on the same projects.

Two caveats are Clarity's own, and they matter:

1. Agent and human are mixed. "Because Clarity cannot currently separate the agent and human portions into different sessions, the findings describe activity recorded during sessions involving agents, not necessarily actions completed autonomously by an agent."
2. It's not a funnel. The events are "independent, non-mutually-exclusive session counts across different websites and industries."

So read it as: when an agent is in the session, the session leans toward doing things, not just reading.

Why a visibility company cares about this

Getting cited and getting used are two different jobs. We've written about the first one for months. This is the second one showing up in real analytics.

An agent doing a task on someone's behalf reads the same first-party pages a model cites: your service page, your hours, your booking form, your availability. If the form is a maze, the hours are only in an image, or the "book" button only works after a pop-up, the agent and the person behind it hit the same wall. Clarity's own question for site owners is the right one: "Where do agents appear to progress smoothly, and where might human intervention still be required?"

That's a first-party page problem. Same pages, new reader.

Conditions you control

1. Put the task facts in text. Services, service area, hours, what a booking needs. In crawlable HTML, not a PDF or an image.
2. Make the action reachable. One clear path to book, quote or buy. Labeled fields. No required step hidden behind a chat widget or a login you don't need.
3. Keep the human in the loop where it counts. Payments, bookings and account changes should ask for a confirmation. Our ChatGPT site-tools Field Guide covers why auto-submit should stay off on anything that costs money.
4. Don't trust a user-agent string alone. Google's own verify requests doc exists because "spammers or other troublemakers" claim to be Google. Verify with reverse DNS or published IP ranges before you allowlist. Signed requests are identity, not a citation (our Web Bot Auth note).
5. Know which bot is which. OpenAI's bots page says ChatGPT-User visits pages for user actions, "robots.txt rules may not apply," and it "is not used to determine whether content may appear in Search." Search opt-outs go through OAI-SearchBot. Our robots.txt Field Guide maps the rest.

Measure agents separately (about 20 minutes)

1. Pick your three task pages. Usually: book or contact, the top service page, and pricing or quote if you have one.
2. Look for agent sessions in whatever analytics you run. If you use Clarity, the post says the Agent Activity Dashboard is generally available for projects with agentic traffic. Clarity's separate Bot Activity dashboard reads server-side logs through supported CDN integrations (Clarity lists Fastly, Amazon CloudFront and Cloudflare).
3. Keep three numbers apart. Bot requests (something fetched the page). Agent sessions (something tried to act). Citations and referrals (an answer pointed a person at you). They're different, and Microsoft says so: bot activity "doesn't indicate that content was retrieved, grounded, cited, or surfaced in AI generated responses."
4. Walk the task yourself. Open the booking page with no cookies on a phone. Can you finish in one pass? If you can't, an agent can't either.
5. Repeat monthly. Monitoring tells you agents arrived. It doesn't fix the form.

What this is not (honesty block)

• Clarity's data, not ours, and not the web. It's Clarity projects that received agentic traffic, across mixed industries. Microsoft hasn't said it represents all sites, or local service businesses in particular.
• Not autonomous completion. Clarity says agent and human portions can't be separated yet, so "checkout recorded" doesn't mean an agent paid for anything.
• The denominator isn't fully spelled out. The tables say "percent of agentic sessions," while the text also calls them shares of "agentic actions," and the checkout count (306,621) is far below 5.2 million. Treat the percentages as Clarity's relative weighting, not a conversion rate.
• The 88% growth is a three-week window in Clarity's own data. It isn't a forecast.
• No one said agent-friendly pages get you cited. Not Microsoft, not OpenAI, not Google. Cited and used are separate outcomes. We're not claiming one buys the other.
• Detection is provider-based. Clarity names OpenAI, Google and Anthropic ecosystems. Agents it doesn't detect aren't in the numbers.
• We don't sell Clarity. It's one free analytics tool with published data. Use whatever you already run, as long as you can separate agent sessions from humans.

Related New Reward Field Guides

• ChatGPT can now use tools on your page. Getting cited did not buy you that.
• Google is signing some agent requests. That is identity, not a citation.
• Google AI Mode Can Call Your Business: What to Fix Now
• AI sends a sliver of your traffic. That's a reason to do the work in the right order, not to skip it.
• ChatGPT UTM tags are missing. Direct traffic is not a visibility crash.

Want the fast read first?

The free AI-visibility score is a quick check: email only, no credit card, a 0–100 read across the engines we track. It won't tell you what agents did on your site. It will show whether AI can see your business clearly.