The customer asked an AI to book a plumber. Here is how it decides whose calendar to open.

What changed in 2026, platform by platform

For two years the AI story for local businesses was about being mentioned. An assistant would name three roofers and the customer would still have to call one. That ended, in stages, over the last fourteen months, and each stage has a date you can check.

Google AI Mode. On August 21, 2025, Google announced agentic booking in AI Mode: the assistant searches several reservation platforms for real-time availability and presents a shortlist with open slots. It launched for U.S. Google AI Ultra subscribers in Labs with restaurant reservations, naming OpenTable, Resy, Tock, Ticketmaster, StubHub, SeatGeek and Booksy as partners, and said it was “expanding soon to local service appointments and event tickets.” In November 2025 the Labs experiment added event tickets and wellness appointments. On April 10, 2026, Google’s Rose Yao announced that restaurant booking was free and worldwide; Semrush documented the flow five days later and noted that the final step still opens a Reserve with Google panel through a partner such as OpenTable or SevenRooms, so it is “not fully agentic yet.” At I/O in May 2026, Google announced expanded agentic booking for specific local services and calling agents for U.S. users; trade press named beauty, pet care and home repair as examples. Google’s own business-calling feature, which phones a business to ask about availability and pricing, had already launched in the U.S. on July 16, 2025. We covered the phone side in detail in Google’s AI is calling Denver businesses for quotes.

ChatGPT. OpenAI’s Help Center release notes for August 10, 2026 say ChatGPT can now find and book restaurant reservations through OpenTable (global), Resy (U.S.) and Yelp (U.S. and Canada), on all plans except ChatGPT Work. On October 20, 2026, Booksy published how its ChatGPT integration works: a customer types something like “find a barber open after 6 PM near me,” ChatGPT shows matching providers with live, bookable time slots in a carousel, and the customer confirms in Booksy. Any business with an active Booksy profile is included; no activation step is required. One thing that did not survive: ChatGPT’s in-chat Instant Checkout for retail, launched with Stripe in late 2025, was pulled back on March 24, 2026 in favor of merchants’ own checkouts. The underlying Agentic Commerce Protocol spec is still maintained (latest stable version dated April 17, 2026), but the lesson for a service business is that the platforms are moving toward handing the final step to you, not away from it.

Claude. On September 2, 2026, Anthropic released Claude Commerce Agents, an Apache 2.0 open-source blueprint with two parts: a customer-facing shopping agent that searches, compares, builds a cart and hands off to the store’s own checkout, and a staff-facing merchant agent whose proposed changes need human approval. Shopify connected it to its stores the same day. It is retail-first, but the pattern, agent gathers and compares, business keeps the final transaction, is exactly what service booking looks like.

Gemini and Google’s commerce rails. Google released the Universal Commerce Protocol in January 2026 with Shopify, Etsy, Wayfair, Target and Walmart as collaborators. Under UCP the merchant stays the seller of record and a checkout button appears on eligible listings in AI Mode and the Gemini app, with payment through Google Pay. Google’s developer page lists industry modules including lodging with real-time pricing and availability, and at Google Marketing Live on May 20, 2026, Google extended UCP to hotel booking and local food delivery (reported by PPC Land; a single source, so treat the date as press-reported). Universal Cart, announced at I/O 2026, lets a user collect items from different retailers across Search, YouTube, Gmail and Gemini.

Notice what is missing from that table: a plumber, an electrician, a roofer. Trades are the last category to get a rail, and the reason is not demand. It is that a restaurant has a table and a salon has a chair, both of which fit a calendar slot, while a water heater replacement needs a price, a service area check and a yes from a dispatcher. The platforms are solving that now, and the businesses that publish those three things in a form an agent can read are the ones that will be first in line when the rail opens in Denver.

How an agent actually picks a business

People imagine the assistant “searching Google” and reading the first result. What it actually does is closer to what a careful office manager does with a stack of quotes. Google describes AI Mode as issuing several sub-searches at once (the “fan-out” we explained in our AI Mode guide), collecting candidates, then checking each against the customer’s constraints: time, location, price, availability, ratings, accessibility, cancellation policy and whether it can be booked at all.

Ignite Visibility’s June 4, 2026 analysis of Google’s agentic booking describes two behaviors worth memorizing. The first is a pattern of consensus: the agent crawls several sources looking for businesses that appear reliable, legitimate and open, which means your hours on Google, Yelp, Apple Maps, Facebook and your own footer had better agree. The second is a risk rule: when the agent finds conflicting data, such as inconsistent hours or service lists, or a dead end such as a static contact form, it moves to a competing business. That is the whole game in one sentence. The agent does not argue with you, email you, or wait until Monday. It leaves.

BrightLocal’s 2026 Local Consumer Review Survey (1,002 U.S. adults, published February 11, 2026) shows how many customers are now on the other end of that decision: 45% used an AI tool for a local business recommendation in the past year, up from 6% the year before, 23% used Google’s AI Mode for one, and 42% say they trust AI recommendations as much as reviews. Yext’s 2026 global survey of 3,848 people found 44% had used AI for local search in the past month and 71% of those use it more than they did a year ago. These are not early adopters anymore. They are a third to a half of your market, and the assistant is doing their shortlisting.

A worked example from a Denver ZIP code

Here is how the decision runs, step by step, for a request of the kind we see every week: “My water heater just died, I’m in 80205, who can come tomorrow morning and roughly what will it cost?” The assistant pulls candidates from the map pack, from review sites and from service pages that mention Five Points, RiNo and Whittier by name. Then it drops candidates. A company that lists 24/7 service on Google and 8 to 5 on its website fails the consensus check. A company with a contact form and no prices hits the risk rule as a dead end. Of what remains, the assistant can read a price range from one site, find live availability for one through its scheduler, and do neither for a third. The business it recommends, and offers to book, is the one that can be quoted and scheduled without a phone call, with enough recent reviews and a license number in the footer to be safe to recommend. Nothing about that outcome requires ranking first. It requires being a business the software can finish the job with.

The five things an agent needs from you

Every platform above, whatever its protocol, is reading for the same five inputs. If you get nothing else from this article, get this list.

1. One identity. Name, address or service area, phone, hours and service list that match on your website, Google Business Profile, Apple Business Connect, Bing Places, Yelp and the directories. The consensus check runs before anything else. Our entity signals guide covers how to audit this.
2. A price the agent can repeat. Not a number you are locked to, a range it can say out loud. “Water heater replacement typically runs $1,650 to $3,200 in Denver depending on tank size and venting” is a quotable sentence. “Call for a quote” is a dead end, and the risk rule treats dead ends as a reason to leave.
3. Availability it can read. A calendar, a scheduler the platform partners with, or an endpoint that answers “what is the next open slot in 80205?” For restaurants and salons this is OpenTable, Resy, SevenRooms, Booksy, Mindbody or Boulevard. For trades, Ignite Visibility lists Jobber, Housecall Pro and ServiceTitan as Google-partnered schedulers; a private booking endpoint does the same job for assistants that read tools directly.
4. A way to book that is not a form. A static contact form is, to an agent, a wall. It cannot see what happens after submit. A booking link with time slots, a Reserve with Google integration, or a tool the agent can call with the customer’s approval is the difference between “booked” and “moved on.”
5. Proof it can verify. License number, insurance, review count and rating, how long you have been in business. The agent reads these to decide whether it is safe to recommend you to a person who will never see your website. BrightLocal’s thresholds (31% require 4.5 stars, 47% skip businesses with fewer than 20 reviews) are now being applied by software as well as people; we worked through them in the 2026 review numbers for Denver businesses.

The Denver angle: distance, seasons and service areas

Denver has three features that change how this plays out compared with the national coverage you will read elsewhere.

Distance still rules the map, but not the agent. Google’s own help page lists relevance, distance and prominence as the local ranking factors, and a business in Lakewood will always have a harder time in the Highlands Ranch map pack than a business in Highlands Ranch. An agent working a booking request is less distance-bound. It is working from the customer’s constraint (“someone who can come to 80205 tomorrow morning”) and from your stated service area. If your service area is stated clearly and consistently, including the suburbs you actually drive to, you can be booked for jobs the map pack would never have shown you for. If it is vague, the agent assumes the worst.

Seasons compress demand into hours. The first hard freeze on the Front Range produces a wave of furnace and frozen-pipe calls in the same evening; the first hail cell of June does the same for roofers; the first 95-degree week does it for air conditioning. Those are exactly the moments a customer hands the job to an assistant because they do not want to call eight companies. A business whose availability is readable at 9 pm on the night of the freeze gets booked. A business whose website says “call us” gets a voicemail, maybe.

Service-area businesses hide their address, and agents notice. Many Denver contractors run from a home or a yard and hide the address on Google, which is correct per Google’s rules. The consequence is that the agent has one fewer thing to verify, so the rest has to be stronger: a visible service area on the site, a license number, insurance, a physical mailing address in the footer if you are comfortable with it, and reviews that mention the neighborhoods you work in. Our Denver metro local SEO guide covers how the suburbs differ from the city for both humans and software.

Restaurants, salons and clinics: the partner rails

If you run a restaurant, salon, barber shop, spa, clinic or anything else that already lives on a scheduling platform, you are closer than you think, and the to-do list is short.

Restaurants. Be on a Reserve with Google partner (OpenTable, SevenRooms, Resy, Tock) and make sure the integration is live on your Google Business Profile. Semrush’s April 2026 write-up warned about an earlier problem where Google had swapped third-party reservation links for its own calling feature on some profiles without telling the owner, and in some cases venues appeared to have no availability at all. Check your own profile from a logged-out browser and from AI Mode. In ChatGPT, OpenTable, Resy and Yelp listings are what get surfaced; a PCWorld test in April 2026 found the agent completed an OpenTable booking only after “about five minutes of agonizingly slow clicking,” which is a reminder that clean partner data matters more than the AI’s cleverness. Our restaurant SEO page is built around the assistant booking the table.

Salons, barbers, beauty and wellness. Booksy’s integration puts every active Booksy profile into ChatGPT results with live slots as of October 20, 2026, and Booksy is also one of Google’s named AI Mode partners. Booksy’s own advice to its businesses is the same as ours: keep the service list clear, hours current, and a few recent photos uploaded, because the agent is drawing on the calendar and services you already maintain. Mindbody and Boulevard are the other names that come up for Reserve with Google in this category.

Clinics and practices. Zocdoc began powering real-time, in-network appointment booking inside Yelp’s iOS app and Yelp Assistant in April 2026. We have not found a Zocdoc app for ChatGPT as of this writing. For a Denver dentist, chiropractor or med spa the practical move is to make sure the scheduler you already use exposes availability to Google (ask the vendor directly whether they are a Reserve with Google partner) and to keep insurance and service lists consistent everywhere, because healthcare is where agents are most cautious.

Trades: what to do while the rails are still being built

If you are a plumber, HVAC company, electrician, roofer, locksmith, garage door company or any other trade, you are in the category Google said it was expanding to at I/O 2026 and that the other platforms have not formally opened yet. There are three ways to be ready, and they stack.

First, make the five inputs readable on your own site. A dedicated page per service with a price range, a visible service area with the suburbs named, hours, license and insurance, and LocalBusiness and Service schema with a ReserveAction in potentialAction pointing at your booking link. This is foundational onsite work; it is the kind of thing covered task by task on our SEO à la carte menu, and it is what the AI Overviews and ChatGPT citations are reading today while the booking rails catch up.

Second, use a scheduler the platforms already talk to. If you are on Jobber, Housecall Pro or ServiceTitan, ask your vendor what their Reserve with Google status is and turn on online booking with real slots. The Google Business Profile setting that matters is under Business Profile settings, then Advanced settings, under “Google automated calls and text messages”: the toggle labeled “Bookings and inquiries from customers” is the one that lets Google’s agents book or ask on a customer’s behalf. A verified profile is required to change it. If an office manager turned all three toggles off in 2025 to stop the hours-confirmation calls, that one is now worth turning back on.

Third, publish a private booking endpoint. This is the newest option and the one that works for assistants that read tools directly rather than through a partner. The open standard the assistants use to connect to tools is MCP; a business can expose a small server that answers three questions, is this ZIP in your area, what does this job cost, what are the next open slots, and takes a booking with the customer’s approval. Our sister product Doorhanger.ai does exactly this for trades, including a free scan that grades your site on the four stages an agent walks through (found, trusted, quoted, booked); we explain the stages on our agentic booking setup page. You do not need to understand the protocol to benefit from it, any more than you needed to understand OpenTable’s API to take reservations.

What to do when the booking arrives

Here is the part that most coverage skips. An agent-sourced booking is a lead with an unusual profile: the customer has already compared you to competitors, already heard your price range read back to them, and already picked a time. They expect the next thing that happens to be confirmation, not a sales call. Two failure modes follow.

The first is silence. If the booking goes to an inbox nobody watches until the morning, the customer, who was comfortable enough to let software book them, will let the same software book someone else. Fronetics’ often-cited figure is that up to 50% of sales go to the vendor that responds first; our own experience in home services says a lead that waits an hour is worth a fraction of one answered in a minute. This is why we wired AI Speed to Leads to confirm by email in seconds and, when the customer agreed to it on the form, to place a natural-sounding call within about a minute that asks your qualifying questions. It never cold-calls and every call says it is an AI, which is also what the booking platforms require of their own agents.

The second failure mode is renegotiation. If the agent quoted $1,650 to $3,200 from your published range and your tech arrives and says $4,800 with no explanation, you have taught one customer and, through the review they leave, every future agent that your published data is wrong. The fix is not to hide prices; it is to publish honest ranges with the variables named (tank size, venting, permit) and to train the team to walk the customer from the range to the number.

A third, quieter problem is attribution. When a job arrives through a partner platform, the platform usually tells you where it came from. When it arrives through an assistant that read your site and called your booking link, the only evidence may be a customer who says “ChatGPT found you” at the door. Ask. Add a “how did you find us” field to the booking confirmation with an AI option, and have the office note it on the ticket. Within a quarter you will know whether the work in this article is producing jobs, which is the only test that matters.

Measure it. GA4 can be set up with an AI Assistant channel for ChatGPT, Gemini and Perplexity referrals; Search Console folds AI Mode and AI Overview clicks into Web data; booking platforms report their own source. We built the setup in how to track AI search traffic and leads, and the number to watch is booked jobs by source, not sessions.

A 30-day plan

Week one: consensus. Pull your name, address or service area, phone, hours and service list from Google, Apple, Bing, Yelp, Facebook and your own footer into one spreadsheet. Fix every mismatch at the source. Check the “Bookings and inquiries” toggle on your Google Business Profile.

Week two: prices and proof. Write a price range with named variables for your top five services and put it on each service page and in your FAQ. Add license, insurance and years in business to the footer and the About page. Ask every customer this week for a review; mention the neighborhood when you ask, so the review does.

Week three: availability and booking. Turn on online booking with real slots in your scheduler, or add a booking endpoint. Replace the “contact us” form on service pages with a booking link that shows times. Add ReserveAction schema pointing at it.

Week four: response and measurement. Route every booking and lead to a phone that is answered, with an instant reply in place for the hours it is not. Set up the GA4 AI channel. Then run the six prompts in our free AI visibility self-check and ask each assistant, in plain words, to book you. Write down what it says. That transcript is your real audit.

If you would rather have someone pull all of that for you, the free audit now includes an agent-readiness pass: we try to get an assistant to book you and show you where it gave up. You can also ask our AI assistant in the header of this page what it can see about your business right now; it reads the same public signals the big assistants do.

Keep going