We Checked 1,096 Contractor Websites for AI Agent Readiness

On Nov. 1, 2026, we checked 1,096 US home-service contractor websites from Google's local map results. About 3 in 4 readable homepages showed no dollar amount, and only about 1 in 8 loaded a scheduler or request form. That leaves a concrete opportunity for a contractor preparing for AI assistants: publish the information they need to compare a service and give them a working route to request it.

Last verified: November 1, 2026.

Google's help page for Gemini in Chrome already says its auto browse can "schedule appointments" for US Google AI Pro and Ultra subscribers. For a contractor, that puts a published price and a booking link ahead of any agent file, and the first fix below is a test you can run on your own form this week.

Agent discovery files are still rare

On Nov. 1, we asked every site in the census for five agent discovery files. None served JSON at /.well-known/mcp.json, the path in the abstract of MCP's earlier SEP-1649 draft, and 57 answered that request with an ordinary web page. None served a Personal Agent Protocol file at /.well-known/poppy.json or an A2A agent card at /.well-known/agent-card.json.

Two sites had something, and only one file was the contractor's own. It's an AI Catalog at /.well-known/ai-catalog.json that lists the homepage, an llms.txt file, an agents.md file labeled as the agent entry point, an XML sitemap and the company's booking page on a third-party scheduler, which documents several entry points. We did not test whether an agent could use them to complete a booking. The other site carried an MCP server card at /.well-known/mcp/server-card.json, but a home-service marketplace builds and hosts that contractor's website, and the card describes the marketplace's own server.

None of the 976 homepages we could read registered WebMCP tools or carried a QuoteAction in schema, and 6 (0.6%) marked up a ReserveAction.

The standards are still moving. MCP's final server card extension, SEP-2127 (created Jan. 21, 2026), makes server cards optional and moves domain discovery to /.well-known/ai-catalog.json, which is why we requested both the old paths and the new catalog. Sierra and Meta announced the Personal Agent Protocol on Oct. 6, and its spec is still a draft. WebMCP, a draft with contributors from Microsoft and Google, lets a page register tools for an agent to call.

Meta put connectors, which plug into Meta AI "through an API or MCP server," into developer preview at Connect (recap dated Sept. 24). The examples on its connectors page are fitness, productivity, entertainment and smart home, and local services aren't among them.

What a contractor's homepage code tells an agent

Everything below comes from raw homepage code as served, with no JavaScript run, so a browser agent may find more than we did.

The basics are mostly in place. Of the readable homepages, 86.0% had a clickable phone link and 54.6% carried LocalBusiness schema, but the facts thin out inside that schema: 47.0% put the phone number there, 49.3% the address and 42.0% the opening hours. A form of some kind sits on 48.3% of homepages, though that count includes search boxes and newsletter sign-ups.

LocalBusiness schema gives search systems explicit fields for your name, phone, address and hours, alongside the information customers see. It is part of making each branch easy to understand. Google's guide to its generative AI features, updated July 10, 2026, keeps the emphasis on established search practices rather than a separate set of AI-only markup requirements. Start with our LocalBusiness schema guide for multiple locations and match every field to the branch's visible page.

Operating partners comparing sites across a portfolio can start with page weight. The median contractor homepage served 173 KB of HTML, and 51 (5.2%) served more than 1 MB before any image or external script loaded; if a portfolio homepage is one of them, ask its web vendor what's in the HTML.

3 in 4 contractor homepages show no dollar amount

We counted generously and still found few prices. Just 23.5% of readable homepages showed a dollar figure anywhere in the page text, with coupons and financing amounts counted.

Schema carried even fewer. Only 3 homepages (0.3%) had a price above zero in Offer or PriceSpecification markup. The priceRange field turned up on 30.9% of the homepages we re-fetched, but 264 of those 285 values were only dollar signs, which tells a machine nothing about what a service call costs.

Links point the same way: 6.8% of re-fetched homepages linked to a page with a word like pricing or rates in its web address, while 18.4% linked to specials or coupons and 19.3% to financing.

An agent asked to collect quotes needs something to compare, and so does the customer reading over its shoulder. Our guide covers what to publish, and how to word a range. If a price also goes into schema, mark up only prices that appear on the page, in the same words, since Google's structured data guidelines, updated July 10, 2026, say markup must be a "true representation of the page content."

Booking software appears on 15.8% of homepages; about 12% load a scheduler

Booking or field-service software code turned up on 154 readable homepages, most often from ServiceTitan or Housecall Pro (65 each). Vendor code on a page doesn't mean anyone can book from it, though. When we re-fetched and sorted them, 117 loaded a scheduler or online request form, and the rest carried only the vendor's other code, named a vendor with no URL we could classify or didn't answer.

That puts the floor at 12.0% of readable homepages, and we didn't send a request through any of those schedulers to see whether it lands. Count every homepage that shows "book online" or "schedule service" text or carries booking software of any kind, and the share rises to 40.7%.

The listings in our scans rarely carry a booking link: 9.4% of competitor listings in Google's local map results showed one, against 80.9% with a website link. Google's AI answers link one more often. In the Cheers Market Baseline, 29.3% of US Google AI Mode answers linked a booking or scheduling page, which counts links in answers, not bookings.

Agent sessions on other sites already go looking for this step. In Microsoft Clarity's study of 5.2 million agentic sessions, published Oct. 29, 2026, 19.85% included a submit-form event (18.56% of human sessions did), 8.42% a book-appointment event and 2.41% a check-availability event. For the form labels, location routing and confirmation states that let an agent finish a booking, see Can AI agents book appointments from your website?

Check the CAPTCHA on your quote form

About a third of re-fetched homepages (34.5%) name a CAPTCHA or anti-spam tool somewhere in their code, most often Google's reCAPTCHA. A name in the code isn't a check on the page, though. A CAPTCHA script actually loads on 16.8% of re-fetched homepages, and on 30.9% of the 430 that have a form, nearly double.

Test the full path on your own form: a loaded CAPTCHA script and a blocked submission are different things. TechCrunch reported on Oct. 6 that some Walmart shoppers using Meta's Muse couldn't finish purchases at a button that asks visitors to click to verify they're human, and Walmart said the blocks weren't intentional. That's a checkout button, a different mechanism from a contact-form check.

DataDome's Q3 2026 numbers show how fast the traffic grew. Agentic browser requests on its network went from about 12.5 million in August to 58.2 million in September, and Muse made up at least a third of September's, a share measured only over the month's last nine days (PPC Land, Oct. 15; Help Net Security, Oct. 19). DataDome also reported that 99.3% of sampled Muse requests came from cloud and CDN infrastructure and that Muse doesn't identify itself as an agent. Whether your form's check lets that traffic through is something only a test on your own form will show.

What to fix before the next census

Six fixes, starting with the one you can run this week.

• Run a real request through your own form this week. If you have Gemini in Chrome's auto browse (US, 18 and over, a Google AI Pro or Ultra plan, desktop Chrome), ask it to request a quote for a service that won't send a truck, and tell your office it's a test; Google says Gemini aims to ask before it submits a form. Watch where it stops, whether that's a CAPTCHA, a phone-only step, a pop-up scheduler it may not be able to open or a thank-you page for a lead that never reaches your CRM, then check that the lead landed and delete it. Gemini can work inside your own desktop Chrome, so a clean pass there doesn't show what a cloud-hosted agent like Muse meets; if you have ChatGPT's cloud browser, which OpenAI's help page says is available in ChatGPT Work on paid plans, run the same request there.
• Copy your phone, address and hours into LocalBusiness schema so search systems can confirm what your page already says. Most homepages already show the phone, but far fewer repeat it in schema, so the work is copying facts you already publish. Have your web vendor match the schema to the phone and address on the page, then check the hours against the ones your office keeps.
• Publish at least one price you can stand behind, such as a diagnostic fee, a trip charge or a "starting at" range with its conditions spelled out, as plain text on a crawlable page. Put it in schema only after it's on the page, in the same words.
• Give your scheduler its own plain URL. If it opens as a pop-up on the homepage, put it on its own page, link that page from the homepage and use the same link as your Google Business Profile booking link. If you pay for field-service software, click your own Book button too, since 23 of the homepages with that kind of code carried only its call tracking, chat, review or portal pieces.
• Ask your web vendor what your form's spam check does with a visitor coming from a cloud server. A blocked submission should show a visible error with your phone number, never a fake success page, and a thank-you page should mean the request arrived.
• Add agent integrations around a real customer task. Use an MCP server, AI Catalog or Personal Agent Protocol connection when an agent or scheduling partner can use it to check availability, request service or book. Keep the public website path working alongside those integrations as standards develop.

We expect customers to hand more of this work to personal assistants: find the right company, compare the offer and arrange the visit. A website that an assistant can read and use puts your branch in a better position for that buying journey.

Our own site, cheers.tech, publishes an MCP server card and an AI Catalog. For a contractor, we start with the customer-facing information and booking path, then add integrations that connect those tasks to the systems the business already runs.

robots.txt and llms.txt: what we counted

No valid robots.txt turned up on 13.6% of contractor sites. Of the sites with one, 6.1% blocked at least one AI bot and 1.6% an AI search or user agent, while 1.5% blocked Googlebot itself. We counted the rules but didn't judge them; which AI crawlers to allow covers that decision.

llms.txt is on a third of these contractor sites (33.7%), and another 6.4% return an HTML page at that address instead of a file. Our llms.txt study looks at who writes the files and whether AI reads them.

Before changing crawler rules, read the Muse blocking guide and distinguish access controls from the customer actions you want to support.

The Personal Agent Protocol guide explains when a connector project has a practical business case and what remains under development.

Method

The panel comes from Cheers' weekly scans of Google's local map results in markets where 10 Cheers home-service customers operate: five scans starting Sept. 28, 2026, holding 6,305 website domains. It isn't a national random sample, and a few markets supply most of the listings.

We took every domain that held a top-three spot in at least half its appearances (282), plus 600 random domains from the 1,727 that held one less often and 600 from the 4,296 that never did. We removed directories, social networks, platforms, government and big-box domains, then Cheers customers, then sites that weren't US home-service contractors, judged from listing names, schema type and domain (433 read by hand). That left 1,096: 110, 527 and 459 from the three groups. Re-weighting them to the frame moves no rate by more than 3.0 points.

One pass ran from 06:57 to 07:00 UTC on Nov. 1 with a desktop Chrome user agent, fetching each homepage, robots.txt, llms.txt and the five discovery paths; 976 homepages were readable. We read inline homepage HTML only: no JavaScript, inner pages, external scripts, form submissions or bot impersonation. We requested discovery files again from 10:13 to 10:17 UTC and re-fetched readable homepages from 12:54 to 12:55 UTC (922 answered) to check prices, schedulers, links and CAPTCHA code.

A discovery file counts if it exists and parses as non-empty JSON; we didn't connect to any server a file lists. A scheduler counts only if the page loads a vendor's booking or request-form address, and a schema price only if it's a number above zero.

Presence rates are floors and absence rates are ceilings: a browser agent may find forms and schedulers a tag manager adds later, a site could publish an agent file at a path we didn't request, and we can't say how firewalls treat real agents. The booking-link rate covers all 15,795 competitor listings in the scans, and the AI Mode rate covers 6,307 US answers from Oct. 9 to 31. Clarity's sessions span 230,511 projects on other websites, Aug. 13 to Sept. 9, 2026; its events overlap and can include a person taking over, so they don't show an agent finished anything. DataDome doesn't publish its network size or Muse sample size.

Cite as: Cheers Local Agent-Readiness Census, edition 1: 1,096 US home-service contractor websites drawn from Google's local map results, crawled Nov. 1, 2026 (UTC). Dylan Allen-Arnegard, Cheers.

Sources

• Ask Gemini in Chrome to complete tasks for you with auto browse, Google Chrome Help, read November 1, 2026. Auto browse can schedule appointments for US AI Pro and Ultra subscribers.
• SEP-1649: MCP Server Cards, HTTP Server Discovery via .well-known, modelcontextprotocol GitHub, October 14, 2025. The earlier /.well-known/mcp.json path.
• SEP-2127: MCP Server Cards, HTTP Server Discovery, Model Context Protocol, January 21, 2026. Server cards are optional; discovery moves to the AI Catalog.
• Introducing Personal Agent Protocol, Sierra, October 6, 2026. The Sierra and Meta announcement.
• Personal Agent Protocol draft spec, personalagentprotocol.org, read November 1, 2026. Draft status and the poppy.json path.
• WebMCP, webmachinelearning on GitHub, August 13, 2025. Draft for pages that register tools for agents.
• Meta Connect 2026: The end-to-end recap, Meta for Developers, September 24, 2026. Connectors via API or MCP server in developer preview.
• Connectors, Meta AI developer page, read November 1, 2026. Example categories, with no local services.
• Optimizing your website for generative AI features on Google Search, Google Search Central, July 10, 2026. No special files or markup needed for AI features.
• General structured data guidelines, Google Search Central, July 10, 2026. Markup must match the page content.
• What AI Agents Do on Websites: Insights From 5.2 Million Sessions, Microsoft Clarity, October 29, 2026. Form, booking and availability events in agentic sessions.
• The next hurdle for AI agents: getting websites to let them in, TechCrunch, October 6, 2026. Muse shoppers blocked at a Walmart verify button.
• Meta's Muse gains a 33% share of AI browser traffic in 9 days, DataDome finds, PPC Land, October 15, 2026. DataDome's agentic traffic counts and Muse share.
• Half the AI bot traffic hammering your website comes from Meta, Help Net Security, October 19, 2026. Muse requests from cloud infrastructure, unidentified as an agent.
• Using cloud browser in ChatGPT, OpenAI Help Center, read November 1, 2026. Cloud browser availability on paid plans.

Sources read November 1, 2026.

Dylan Allen-Arnegård is the CEO & Co-Founder of Cheers, the done-for-you local visibility partner for established home-service companies and PE-backed portfolios.