AI agents and local business: what to have ready

What is an AI agent that books local services?

It is software that takes a task from a customer, such as “find a plumber open Saturday and book the first slot,” and works through websites and profiles to do it. Google’s guide defines agents as “autonomous systems that can perform tasks on behalf of people” and uses booking a reservation as its example.

Is this the same as an AI receptionist?

This is the customer’s agent, not yours. Many pages ranking for this topic sell AI phone or chat agents that answer a business’s own calls.

That is a separate purchase, and the last FAQ answer below keeps the two apart.

• Booking agentActs for the customer. Reads your public pages and profiles.
• AI receptionistA tool you deploy to answer your own calls or chats.

What does Google say about agents?

Google’s own guide to optimizing for generative AI features was last updated 10 July 2026 on the page. It says browser agents may analyze screenshots, inspect the page’s DOM and read the accessibility tree.

The changelog shows that agent guidance was added on May 15, 2026, described as “initial guidance.”

Google also lists a Google-Agent fetcher (page updated 19 August 2026). Agents hosted on Google infrastructure use it to browse the web and take actions on a user’s request.

1. May 15, 2026The changelog shows agent guidance added as "initial guidance."
2. 10 July 2026The generative AI guide was last updated.
3. 19 August 2026The user-triggered fetchers page was updated.

What does Google actually tell businesses to prepare?

Very little that is specific, which is the honest starting point. The table shows what the guide says and what it leaves out.

That leaves a gap between what is documented and what an owner can do on Monday morning. The rest of this post fills it with our view.

Where we infer instead of cite, the text says so. Each inference is cheap to act on, because it also helps human customers.

For the bigger AI search picture, see our guide to AI search optimization and the list of LLM SEO myths.

Which business data should be ready for an agent?

Publish eight facts as plain text on a page, matching your Google Business Profile, with one owner for each. An agent comparing providers, or a person using one, can only act on facts it can read.

This list is our view of the minimum. We built it from how agents are documented to read pages.

For you, that means each row needs one owner. That person changes the fact first and then checks every other place it appears.

Pricing and service area are the two rows small-business sites most often fail. So they get more detail here.

How should you state your pricing?

You do not need a price list. State the method (flat fee, hourly rate, or priced after inspection). Then say what changes the number.

If you publish a range, publish only one you honor today. Our FAQ pages post shows how to write that answer without inventing figures.

The second version names the method and what moves the price. It holds no number the shop would have to defend.

How should you state your service area?

Write the cities you serve as text on the page. Google’s Business Profile hours help covers hours only. Our service area guide covers the Business Profile side.

Our rule: if you would turn down a job in a town, leave that town off the list.

How do you keep hours correct in every place an agent might read them?

Pick one source of truth, change it first, then copy it to every other surface the same day. In our experience, hours are the fact that most often differs between a website, a Business Profile and third-party listings.

A wrong hour is an easy failure for a customer to notice.

What hours settings does a Business Profile have?

Google’s hours help page describes three settings on a Business Profile:

• Main hours.
• Special hours for holidays or temporary changes.
• “More hours” for features on different schedules.

Google reviews edits before they go live. In select regions and categories it may call the verified phone number to confirm details.

What does an hours mix-up look like?

Here is a hypothetical reconciliation for a drain-cleaning shop. The places disagree.

Decide the real hours first. Then fix the profile, the site and the directory in that order, and note the date in a shared sheet.

Because Google reviews profile edits before they go live, make them well before a holiday, not on it.

1. Decide the real hoursSettle what the shop truly offers before touching any page.
2. Fix the Business ProfileUpdate main hours and any special hours first.
3. Fix your websiteShow the same hours as visible text.
4. Fix old directoriesCorrect the listings that still show old hours.
5. Note the dateRecord the change in a shared sheet.

Write hours as visible text on your site. You can add structured data too.

Google’s local business structured data documentation (updated 8 September 2026) lists openingHoursSpecification for opening hours and gives times in hh:mm:ss format. It also says to use both validFrom and validThrough to define seasonal hours.

This fragment uses made-up hours and sits inside your existing business markup. Repeat the first block for each weekday:

"openingHoursSpecification": [
  {
    "@type": "OpeningHoursSpecification",
    "dayOfWeek": "Monday",
    "opens": "08:00:00",
    "closes": "17:00:00"
  },
  {
    "@type": "OpeningHoursSpecification",
    "dayOfWeek": "Saturday",
    "opens": "09:00:00",
    "closes": "13:00:00",
    "validFrom": "2026-11-01",
    "validThrough": "2027-03-31"
  }
]

Markup must match visible content. Google’s general structured data guidelines say not to mark up content that readers cannot see.

Whether any agent reads this markup at all is unproven. Visible text on the page is the safer bet.

Our Business Profile optimization checklist lists the other profile fields worth checking while you are in there.

Can an agent finish your booking form?

Test it yourself. Try to complete the booking with a keyboard alone, in under two minutes, without an account. If you cannot, a hurried agent or customer is likely to stall too.

That is our hypothesis, not a measured result. The test costs nothing and finds real problems.

Google says browser agents may read the accessibility tree. It also says semantic HTML helps screen reader users too. Taken together, a form that works for assistive technology is a plausible fit for an agent.

Run these checks on your main booking path:

1. Every field has a visible label that is tied to the field, not placeholder text alone.
2. The submit button says what it does (“Request a visit”), not “Submit.”
3. You can reach and finish the form using only Tab, Shift-Tab and Enter.
4. The time-slot picker shows its options as text, not a picture or canvas.
5. Required fields are marked, and error messages appear next to the field in words.
6. No account, app download or phone call is needed to send a basic request.
7. The confirmation screen states what happens next and when you will hear back.
8. The page still works at phone width, since many customers book from a phone.

Which step will most likely fail?

We expect step 4 to be the common failure. Some third-party scheduling widgets load inside a frame that an agent may not read well.

We have not tested that. Ask your scheduler’s vendor.

Our rule: if the vendor cannot say how the widget works with a screen reader, add a plain text request form beside it.

Our website accessibility and SEO post covers the human side of these same fixes, and why they are worth making anyway.

Should you block AI agents or let them in?

Leave public booking and information pages open, and put private pages behind a login. Do not rely on robots.txt for either goal.

What does robots.txt really control?

Google says user-triggered fetchers “generally ignore robots.txt rules,” because a user asked for the fetch.

Google’s robots.txt introduction (updated 10 December 2025) adds that robots.txt is “not a mechanism for keeping a web page out of Google,” and that compliance is voluntary. So robots.txt controls well-behaved crawlers, not agents acting for a user.

What is the risk of blocking too much?

The practical risk runs the other way: you block something you wanted. Firewalls and bot filters that challenge unknown visitors can stop a legitimate customer or agent before they see the form.

Our decision rules:

• Booking and service pages: allow. Test the path from a clean browser session.
• Customer portals, invoices, quotes with personal data: require login. Never depend on robots.txt.
• Bot filters: review the rules that apply to your booking URL and keep logs of what they block.
• Verification: Google warns user agent strings can be spoofed. It says it is testing Web Bot Auth with the identity https://agent.bot.goog. Treat names as a hint, not proof.

We have not checked client logs for Google-Agent traffic, so we cannot say how common it is on local sites. If you can read raw logs, searching for the user agent name is a cheap first look.

How can you test whether agents handle your business?

Run a small, dated, repeatable test, and write down the failures along with the successes. One run proves nothing, because answers vary between runs and accounts.

Our GEO measurement limits post explains why. The same caution applies to agents.

This is the protocol we would use. As of 22 October 2026 we have not run it for a client, so it is a method, not a result.

1. Fix the prompt set. Write five to ten prompts a real customer would type, such as “book a drain cleaning for Saturday morning near a named ZIP code.” Include one that names your business and several that do not.
2. Pick the engines your customers use. Common candidates are Google’s AI Mode, Gemini, ChatGPT, Perplexity and Microsoft Copilot. Record the product name and date, since products change often.
3. Run each prompt three times, from the same location and with the same signed-in state. Log each run separately.
4. Log six outcomes per run: business named, hours right, price method right, booking page reached, completed or handed off, and where it stopped.
5. Stop before you pay or confirm a real appointment. Use a test mode or cancel at once, and tell your team so nobody gets a surprise job.
6. Repeat after each fix, and change one thing at a time.

The blank log below is the one we would use. Its only row is empty on purpose, because nothing has been run yet.

Our Copilot visibility post covers one engine’s quirks. Our AI referral traffic in GA4 post covers how to see any resulting visits.

What is still unproven about AI agents and local business?

Most of the headline claims are unproven. This is our running list as of 22 October 2026.

It is long on purpose, because a short one would make the topic look more settled than it is.

• Scale. We found no verified figure for how many consumers use agents to book independent local businesses. Do not plan budgets around volume.
• Markup effect. Google says no special schema is needed for its AI features. We have no evidence that adding any markup makes an agent choose you.
• llms.txt. Google says its Search does not use these files and that creating one will “neither harm nor help” its Search visibility. We make no claim for other agents.
• Business Profile and selection. Google recommends Business Profiles for visibility in AI responses. It does not say profile completeness decides which business an agent books.
• robots.txt. Google says its user-triggered fetchers generally ignore it. We have not verified what other vendors’ agents do.
• Other vendors. Google’s documentation describes Google’s agents. We have not verified how OpenAI, Anthropic, Microsoft or Perplexity agents read local sites.
• Commerce protocols. Google mentions the Universal Commerce Protocol as emerging. We could not open its documentation, so we make no claim about whether it covers service bookings.
• Frames and widgets. Whether agents can use booking widgets loaded in frames is untested by us.
• Our own tests. We have no first-party test results to publish yet. Anyone who shows dated screenshots of agents booking a specific business should show the prompt, the engine and the date too.

What should you do this week?

Fix the human-facing basics in this order. Each one pays off with or without agents.

1. Reconcile hours across your site, Business Profile and top listings, and set special hours for the next holiday.
2. Add or rewrite a plain statement of how you price on each core service page.
3. List your service area as text and your services as separate pages.
4. Run the eight booking-form checks and fix the failures.
5. Check firewall or bot rules against your booking URL.
6. Run the dated prompt test once as a baseline, then again after fixes.

Our Local SEO checker can show whether your profile and site basics line up before you start.

Where does the wider work belong?

The wider work belongs with local SEO and with AI search optimization. There we handle it as one system with your content and profile.

If you want those fixes planned against your own site and profile, we can build that into a growth plan.

Get Your Organic Growth Plan

Frequently asked questions

Sources

• Guide to optimizing for generative AI features on Google Search, Google Search Central, last updated 10 Jul 2026, accessed 22 Oct 2026
• Google’s user-triggered fetchers, Google Search Central, last updated 19 Aug 2026, accessed 22 Oct 2026
• Changelog of Google Search Central documentation, Google Search Central, accessed 22 Oct 2026
• Local business structured data, Google Search Central, last updated 8 Sep 2026, accessed 22 Oct 2026
• General structured data guidelines, Google Search Central, last updated 10 Jul 2026, accessed 22 Oct 2026
• Introduction to robots.txt, Google Search Central, last updated 10 Dec 2025, accessed 22 Oct 2026
• Edit your business hours, Google Business Profile Help, accessed 22 Oct 2026
• AI features and your website, Google Search Central, last updated 10 Dec 2025, accessed 22 Oct 2026