Could an AI agent book an appointment at your practice?

AI search is moving from answering questions to completing tasks. Google’s AI Mode has been live in Australia since October 2025, and at its I/O conference in May 2026 Google announced the next step: Search that checks pricing and availability for local services and books on the user’s behalf — starting in the US, in categories like home repair and beauty, and expanding from there. When that behaviour reaches Australian healthcare, the practices that win bookings will be the ones whose websites a machine can read, price, and book without a phone call. Five checks tell you whether yours is one of them — and every single one also pays off with human patients today.

From answering questions to doing the job

For the past two years, the story of AI search has been about answers. A patient asks ChatGPT or Google’s AI Mode “how much does a psychology session cost in Geelong?” and the AI assembles an answer from whatever websites it can read confidently. We’ve written before about why most practice sites never get cited in those answers — and that problem hasn’t gone anywhere.

But the ground is shifting again. The next phase isn’t answering the patient’s question; it’s doing the patient’s errand.

At I/O in May 2026, Google demonstrated Search handling requests like a private karaoke room for six people on a Friday night — returning live pricing, availability, and booking links rather than a list of websites to go read. For some categories, Google will even ring the business on the user’s behalf. The rollout is US-first and the launch categories are things like home repair, beauty, and pet care rather than healthcare. But Google has already begun testing healthcare ads inside AI Mode, and the direction is not ambiguous: the assistant is becoming the person who does the looking, the comparing, and increasingly the booking.

The demand side is already here. Roughly one in ten Australian adults asked ChatGPT health-related questions in a recent six-month window — around 1.9 million people, according to a nationally representative survey published in the Medical Journal of Australia — and about four in ten of those who knew of the tool but hadn’t yet used it for health questions said they were considering it. Those people aren’t going back to typing keywords and opening ten tabs. The moment their assistant can also book the appointment, a meaningful share of them will let it.

The timing for Australian allied health specifically is uncertain — months, more likely a year or two. The preparation is not uncertain at all, because it’s the same work that gets you cited today.

What an agent actually does on your website

An AI agent is the most literal-minded visitor your website will ever have. It doesn’t infer, it doesn’t ring around, it doesn’t squint at a screenshot of your fee table and take a guess. Walking through the errand step by step shows exactly where practice websites fail.

Say the request is: “Find me a psychologist near Box Hill who sees adults for ADHD assessment, takes Mental Health Care Plan rebates, and has something available in the next fortnight — and book it.” The agent has to:

1. Work out who you treat. It reads your service pages looking for a plain statement: who you see, what for, and — just as important — who you don’t see. A page of warm brochure copy about holistic wellbeing journeys gives it nothing to match against. It moves on to a competitor who states it outright.
2. Establish what it costs. It looks for fees, rebate amounts, and gap costs as text on a page. If your fees live in a downloadable PDF, or behind “contact us for pricing”, the agent can’t complete the comparison step — so you lose at the shortlist stage, silently.
3. Check availability and book. It looks for a booking path it can follow: a clearly labelled online booking link that leads to real, visible appointment times. A “request an appointment” form that promises a callback is a dead end for a task the user asked to have finished.
4. Hand over cleanly. Even when the agent can’t complete the booking itself, it hands the patient the last step: “this practice matches, here’s the booking link, here’s the fee.” The practices that survive to that handover are the ones whose sites answered steps 1–3.

Notice what’s missing from that list: your colour scheme, your logo, your mission statement. The agent is indifferent to all of it. It cares whether your site states facts a machine can read and offers actions a machine can take.

Five checks that make a practice site agent-ready

None of this requires new technology. It requires the information you already have, published the way machines — and, not coincidentally, busy humans — need it.

1. Your key facts exist as plain text on real pages. Fees and rebates with actual dollar figures, who you treat and who you don’t, telehealth availability, referral requirements, realistic wait expectations. Every one of these answers should be findable as ordinary text — not a PDF, not “enquire for details”. This is the same direct-answer structure that gets you cited by AI search; agents just raise the stakes.
2. Schema markup that matches what the page says. Structured data — LocalBusiness, MedicalBusiness, FAQPage, your address, phone, and opening hours — is how a machine confirms what it thinks it read. It’s a one-time job, and we’ve written a complete schema guide for allied health practices to work from.
3. A booking path a machine can reach — ideally complete. A visible “Book online” link on every service page, leading to a booking engine that shows real availability without demanding an account first. If your booking system offers appointment types with clear names and durations, an agent (and a human) can act on them. A contact form where a booking should be already costs you bookings; against agents, it’s disqualifying. Our piece on structuring booking pages covers what good looks like.
4. Nothing important locked inside images, PDFs, or widgets. A fee table as an actual HTML table, not a screenshot of one. Service lists as text, not an infographic. If a machine can’t select the text, assume the machine can’t use it — and neither can the patient using a screen reader, which is the accessibility win hiding inside this whole exercise.
5. AI crawlers allowed in. Your robots.txt decides whether GPTBot, ClaudeBot, PerplexityBot, and Google’s AI crawlers can read your site at all, and an llms.txt file tells them what matters. Some businesses choose to block AI crawlers on principle; for a practice that wants to be found and booked, invisibility is a high price for a principle. This is a five-minute check your web person can do today.

If you’ve already done the GEO fundamentals from our GEO vs SEO explainer, you’re most of the way there. The new work is mostly item 3: making sure the action — not just the answer — is reachable by something that isn’t a person.

What you don’t need to do

Every platform shift produces a wave of packages you’re urged to buy. Most of it, as usual, can be skipped.

• An “AI agent optimisation” retainer. The acronyms are new (AEO, agentic SEO, answer-engine optimisation); the work underneath is the fundamentals above. Anyone selling it as a proprietary ongoing service is selling you a lawn they intend to mow forever.
• A chatbot on your website. A patient’s agent doesn’t want to negotiate with your chatbot, and neither do your patients. Publish the answers as text instead of hiding them behind a conversation.
• A panicked rebuild. If your site is structurally sound, this is a content-and-markup exercise measured in days, not a new website.
• Taking the phone number off. Google’s agents make phone calls, and plenty of patients — especially older ones — still prefer to ring. The phone line and the website are now answering the same query; keep both sharp.

The honest timing question

Should you drop everything for a capability that isn’t in Australian healthcare yet? No — and that’s not what this is.

Everything on the five-point list improves your conversion with human visitors this month: clear fees reduce no-show enquiries, real online booking captures the after-hours decision-maker, plain-text answers win AI citations that are already sending patients today. The agentic layer, whenever it arrives here, will be built on exactly the same signals. That’s what makes this the rare trend where the preparation is free of regret — you’re not betting on a feature, you’re publishing your practice’s facts in the form every reader, human or machine, prefers.

The practices that lose this shift won’t lose it in a dramatic moment. They’ll lose it the way sites without mobile design lost the 2010s: quietly, one invisible shortlist at a time.

At ABCS we build allied health websites that are structured for both audiences — the patient and the machines acting on their behalf — with the schema, content structure, and booking flows this piece describes. If you’d like to know whether an agent could book your practice today, that’s a conversation we’re happy to have.