What stops an AI agent from booking with a local business

“Book me a cleaning next Tuesday afternoon somewhere near work.” This is the kind of request people are starting to hand to assistants. To complete it, an agent has to do five things on your website. Most local business sites make at least one of them impossible — not through any exotic problem, but through ordinary decisions that were fine for humans.

Here are the five tasks, why each fails, and what fixes it.

Task 1: understand the offer

The agent needs to know what you do, for whom, and where — in text. A hero image that says “Smile with confidence” tells a human what you are; it tells a model nothing. A page whose only description of services is inside a slider that loads after the page renders is invisible.

Fix: one paragraph of plain text near the top of the home page stating the business type, the services, and the area served. Then a LocalBusiness (or Dentist, Plumber, Attorney) schema block that repeats the same facts in JSON-LD.

Task 2: find the price

An agent asked to “find a dentist for a crown under $1,500” has to find a number. Prices in a PDF are unreadable to most agents; prices in an image are unreadable to all of them; “call for pricing” ends the task.

You do not need to publish an exact price list. A range — “crowns from $900 to $1,400 depending on material” — is enough for the agent to include you rather than skip you.

Fix: a pricing or services page with ranges as HTML text, and Offer markup with priceRange in the schema.

Task 3: find the way to book

This is where most sites fail. The agent looks for a link that leads to a booking step. What it finds instead:

• a “Book now” button that opens a JavaScript modal with no URL of its own,
• a third-party widget embedded in an iframe from a domain the agent has not been told about,
• a form behind a CAPTCHA wall on page load,
• a phone number as an image, to prevent scraping.

Every one of these is a reasonable choice for a human visitor. Every one of them ends the agent's attempt.

Fix: a real URL for booking (/book), reachable by a normal link, that leads to a page where the booking form or the phone number exists as text. The widget can stay; give it a page of its own with a plain-text fallback.

Task 4: check the conditions

Hours, cancellation policy, service area, insurance or payment methods. The agent needs to confirm that Tuesday afternoon is inside your hours and that you serve the customer's part of town. If hours are only in the footer as an image, or the service area is only a map, the agent cannot confirm and will prefer a business where it can.

Fix: hours and service area as text, openingHoursSpecification and areaServed in the schema, and one page — /policies or /faq — with cancellation and payment terms in plain sentences.

Task 5: reach a human

Even a fully automated flow needs an escape hatch: a phone number or contact path the agent can hand to the user. A number rendered as an image, or a contact form that requires solving a puzzle, closes the last door.

Fix: phone number as text with a tel: link, email as text, and contactPoint in the schema.

Why this is not a redesign

None of the fixes above changes how your site looks. They add text where there were images, URLs where there were modals, and a block of JSON-LD in the page template. A developer can do the full set in an afternoon; llms.txt and robots.txt are text files you can upload yourself.

How Anserra tests this

The Agent Readiness check runs exactly these five tasks with a real browser agent, follows only links already on the page, never submits a form, and reports the step where it stopped with a screenshot. The result maps to levels L0–L3 from the Agent-Ready Local Services specification. The free scan includes a quick version; paid plans run the full check monthly or weekly.

The point is not the score. The point is the screenshot of the exact place where a customer's assistant gave up on you — and the one-line fix next to it.