Clarity AI agent traffic: which visits become real enquiries?

By KiwiFlow ·

TL;DR

Microsoft's November 4 Clarity announcement gives service businesses a reason to examine AI-assisted customer journeys. Keep agent activity, human referral visits and confirmed enquiries separate in your report. We would start with one service page and check what reaches your booking system before buying another tool.

What Microsoft announced

Microsoft Advertising's November 4 announcement introduces Agent Activity, conversational analysis and Brand Agents. It describes visibility into agent visits and actions, answers supported by session evidence, and help with product discovery or service bookings. These are product capabilities, not a new organic search ranking rule.

The full Clarity announcement sets narrower conditions. Agentic Analytics supports agent operators from Google, OpenAI and Anthropic. Simulations initially cover commerce and travel. Local-business Brand Agents require an eligible WordPress business and a supported CRM. Product discovery across Microsoft's AI experiences applies to eligible Shopify and WooCommerce merchants. Ask Clarity is described as a closed beta for retailers; AI Visibility and advertising data are planned additions.

For a Canadian or US service business, check the eligibility of the feature you intend to use. The announcement's broad global language does not establish that your website, account and booking provider qualify. Installing a Clarity tag alone does not complete the separate Brand Agent setup.

Ask what the report is counting

Suppose your supplier sends a report with twenty AI visits and three booking actions. This is a hypothetical example, not a KiwiFlow client result. Before calling those actions new business, ask what each number means.

An agent activity record describes a detected visit or action on the website. A human arriving through an AI referral is a different journey. A confirmed enquiry requires evidence that your business received a request it can respond to. Keep those definitions beside the numbers so the next person reading the report does not silently change them.

For our own reporting decisions, we would use the following distinctions:

ObservationWhat we would recordWhat still needs checking
Detected agent visits a service pageAgent activity on that pageWhether it was working for a prospective customer
Visitor arrives with an identifiable AI referrerAn observed AI referral visitWhether that visit produced an enquiry
Booking button is clickedIntent to start bookingWhether the booking system accepted anything
Inbox or CRM receives a valid service requestConfirmed enquiryService fit, attribution and whether the person books

A visitor can appear at several stages. Adding the rows would double-count journeys. For the same reason, a booking action count should not be divided by an unrelated total of AI referrals and labelled a conversion rate. Agree on a consistent set of eligible visits first, and leave the rate unreported if you cannot connect the records reliably.

Follow one service request through the website

Imagine a homeowner trying to book a leaking-tap repair. The website describes plumbing work, but the booking menu offers only a general consultation. That mismatch gives the owner a specific question to investigate: can someone request the repair they just read about?

Start with the service page. Check whether it explains the work you accept, the area you cover and what happens after an enquiry. If pricing needs an inspection, say that. A vague promise of an immediate quote can create the wrong expectation before anyone reaches the form. The same questions matter when a customer reads the page directly. Our local SEO foundations guide explains how clearer service information supports the next decision.

Then inspect the booking destination. Does the service name match? Can the customer select a supported location? Is the available appointment a confirmed slot or a request awaiting your response? Use the wording your business can honour. We would repair a misleading booking promise before adding an assistant that repeats it.

If your tools show a failed journey, use the evidence to choose the repair. A missing service option calls for a booking-menu change. A broken link needs a link fix. Neither finding, by itself, justifies rewriting every page for GEO, the practice of improving visibility in generative search experiences.

Any controlled test should use your own designated test details and be labelled so it does not become a sales lead. Avoid repeated production submissions. Confirm the result in the destination system, then exclude the test from business reporting. A success message in the browser is only one part of that check.

Keep a record you can compare next week

Pick one reporting window and one service page for the first review. Record the tool, its filters and the date. Keep a note of changes to consent settings or tracking during that window, because a change in collection can look like a change in customer interest.

For each observed enquiry, ask whether the source is known and whether the request fits the business. An out-of-area request can be genuine while still being unsuitable. Keep unknown attribution as unknown. You can improve the customer's experience even when you cannot prove which search or AI interaction started it.

A small business may have too few observations to calculate a useful trend. In that case, save the individual failure pattern without presenting a percentage improvement. One repair request that never arrived deserves investigation; it does not establish that an entire channel performs poorly. Likewise, one successful booking cannot show that installing a tool caused extra demand.

We would also keep website behaviour separate from claims about authority. An agent accessing a page does not establish an independent recommendation of your business. If someone reports an AI citation, ask for the actual answer, query, platform and date. That evidence documents one observed appearance. Repeated relevant discovery and qualified enquiries over time are more useful than a screenshot presented as permanent visibility.

Decide what to fix before adding another feature

Bring one service URL, the booking destination and the evidence of the suspected failure to your next website review. Ask your supplier to explain the smallest repair, how they will check it and which result will count as success. If they recommend a new integration, have them establish eligibility before pricing the implementation.

For a business with little observed AI activity, improving a confusing enquiry path can still be worthwhile. The reason should be the problem you found, with a result you can check. You do not need to predict how quickly customers will adopt agents to correct a service menu that sends them to the wrong appointment.

If the uncertainty starts with what customers can find and understand on your service pages, see what KiwiFlow's SEO Foundation reviews.

How many jobs did you lose this month?

You don’t know, and that’s costing you. Tell us how customers find you and what happens after they get in touch. We’ll show you where the work goes missing.

Book a call with KiwiFlow ↗