Conversational AI in healthcare: what actually works

Four messages. That's the lot. An email and an SMS the day before. Then an email and an SMS on the morning of, not too early. That's the reminder set-up I'd build for a clinic, and the restraint is the point.

Every clinic needs an email reminder system. SMS on top is worth adding, because email gets missed. Go past that and you start working against yourself.

One thing to check while you're in there. I've seen clinics who thought their reminders were going out when they weren't, and it was either a wrong setting or they'd run out of message credits and hadn't bought more. Both are boring, and both got spotted fast. Still worth opening your send log today.

Most of what gets sold as conversational AI in healthcare is much bigger and much vaguer than this. Search the term in Australia and you get global platforms built for enterprise call centres. Or a doctors' body telling GPs to be careful with chatbots. Neither one tells you what to do on Monday.

The short version. Send four reminder messages and stop. Run a recall as a clinical duty, and ring first. Move intake onto an online form before the visit. Keep any offer out of a win-back message. Point an AI assistant at your procedure documents. Treat one pointed at your patient database as a separate buy, with separate care.

What conversational AI means once you take the phone out of it

Conversational AI is software that talks with a person in plain language, by text or by voice. Two very different things wear that label, and mixing them up is where clinics get nervous about the wrong one.

The first is the general-purpose kind. The RACGP fact sheet says it covers "advanced chatbots, virtual agents/assistants, and 'embodied conversational agents'", and it names "OpenAI's ChatGPT, Google Gemini, Microsoft Copilot, and Anthropic's Claude" as the big ones (RACGP). That page was last updated in September 2025. It's blunt: "GPs should be extremely careful, however, in using conversational AI in their practice at this time. Many questions remain about patient safety, patient privacy, data security, and impacts for clinical outcomes."

Read that as aimed at a GP typing a patient's situation into a general chatbot. It isn't aimed at a reminder text.

The second one is what this page is about. It's the patient messaging software your clinic probably already half owns. Reminders, recalls, intake forms, follow-up, win-backs and web chat. That same RACGP page lists "providing a platform for scheduling appointments and sending reminders to patients" and "collecting and collating patient information" as normal business uses. So the category is real in their words too.

Phones are their own problem with their own answer. If what you really want to know is who picks up when the front desk is flat out, go and read our guide to AI receptionists for medical and allied health clinics instead. Everything below is what your clinic says to a patient when nobody's on a call.

Channel by job

Reminders: four messages, and why that's enough

Here's the whole cadence. Treat it as your default and change it only when you've got a reason to.

Two days, two channels, four messages. That gives the patient a real shot at cancelling or moving the slot. That's what you want a reminder to do. A slot handed back at 8am gets filled. A no-show at 2pm doesn't.

Reminder fatigue is what happens when a patient gets too many messages and stops reading any of them. I've seen that version in other practices. More channels, more touches, a ladder of reminders that keeps climbing. It doesn't lift attendance. It teaches people to tune out.

It shows up outside healthcare marketing too. The Association of Health Care Journalists wrote about US patients being "bombarded by text and email appointment reminders". It quoted Emily Kagan Trenchard, chief of consumer digital solutions at Northwell Health: "We've recognised that our patients are feeling bombarded... [that's] not the way to keep them engaged" (AHCJ, 2 May 2024). Northwell's fix was to fold five notification systems into one. That's the US, and those clinics are far bigger than anything I'm describing here. The mechanism travels anyway.

Keeping it simple is the fix, and there's no clever version of it.

Why SMS, and where it stops working

SMS earns its spot because email gets missed. Not because it's newer.

On what patients want, the only Australian data I could find is HotDoc's own survey of over 1,000 Australians, from 11 November 2019. It found 55% preferred SMS for post-appointment follow-ups. Phone came in at 24%, email at 20% and post at 1% (HotDoc). The SMS preference held across age, gender and location. It only dropped for patients aged 75 and over. Read it with the labels on. It's a vendor's own survey, nobody audited it, and it's nearly seven years old. The drop at 75-plus is the bit most likely to matter to you. If your book skews older, don't assume the headline number is yours.

Do reminders actually cut no-shows? One checkable study sits under most of the numbers you'll see quoted. Koshy, Car and Majeed published it in BMC Ophthalmology in 2008. They looked at an eye clinic at Barts and the London NHS Trust between April and September 2006. Non-attendance ran at 11.2% with an SMS reminder, against 18.1% without. They called that "a reduction of 38% in the likelihood of patients not attending their appointments" (Koshy et al., 2008). So it's a UK hospital eye department, nearly two decades ago. Say that every time you use the figure. The direction still holds. As a forecast for your allied health clinic next month, it's a long way from home.

Recalls are a clinical duty first and a message second

A recall is you contacting a patient to bring them back in, usually about a test result. The order matters. No vendor page mentions it, because none of them are written from a medico-legal seat.

Avant is a medical defence organisation. It puts the duty plainly: "Ultimately the person ordering the test is responsible for following up any clinically significant results... If a proposed test, investigation or referral is likely to be clinically significant, you are expected to follow up with the patient" (Avant, 6 May 2026). The same page adds that putting a result on My Health Record "does not change your responsibility to follow up results and recall patients."

Automating the message doesn't hand that duty over. The software sends the text. The duty stays with whoever ordered the test.

Avant's own pattern for a recall that isn't landing runs phone calls at different times of day, then SMS, post or email, then a registered letter. Steps past that, such as contacting family or the police, are only for the truly urgent. Build your workflow around that shape. Not around whatever your software happens to automate easily.

Consent, specifically for texting

Texting gets treated differently from ringing. Avant again: "While you do not need specific consent to contact a patient by telephone, you should always be mindful of how much information is disclosed... You should specifically check whether patients agree to receiving text messages" (Avant, 19 December 2025).

There are three real places to get that consent, and Avant names all three. The registration form, for a new patient. The moment you confirm an appointment, for someone already on the books. Or an online booking or registration portal.

What you put in the message matters as much as the consent does. Avant: "Messages should not contain sensitive health information such as a description of particular test results that need to be followed up or the results of such tests (unless the patient expressly consents to this). The message should simply request that the patient contact the practice."

Intake and forms, before the patient walks in

Pre-visit digital forms are the least controversial thing on this page, and usually the quickest win. Details turn up typed, instead of copied off a clipboard. Consent gets captured properly, instead of remembered later.

One thing to get right. A patient-facing tool that takes in personal details needs the patient's okay first. The RACGP says to "obtain informed patient consent when using patient-facing AI tools; for example, if using conversational AI as part of an intake procedure" (RACGP). An intake form is named in that sentence for a reason.

Our guide to AI in healthcare administration covers the wider privacy picture for Australia and New Zealand. Getting your team to use any of it is a separate job, and that one is AI adoption in healthcare teams.

Reactivation: the answer depends on what kind of practice you run

Every source I found on patient reactivation was a marketing agency describing its own results. One promises to win patients back in as little as 30 days. No regulator, no peer-reviewed study, nothing vendor-neutral. So there's no reactivation number on this page, because there isn't one worth printing.

Reactivation is contacting a patient who has drifted off, to bring them back into care. What follows is my own view. I formed it running a multidisciplinary clinic in New Zealand, not from a case study.

A blanket recommendation across all of healthcare is wrong here. Splitting the market three ways is my call, not a published framework. Allied health behaves differently from medical, and both behave differently from everything else. Each needs its own answer.

Allied health and medical. Getting back in touch with a patient who has drifted has to be relationship-based. It can never come off as salesy. What works looks like this:

• Gifting.
• Marking a date that means something, such as a birthday.
• A friendly check-in by email or SMS with no call to action in it at all, genuinely asking how they're doing.

That last one is the hardest to pull off and the most valuable. The second there's an offer at the bottom, it stops being a check-in. A bit of templated context is fine, so the message isn't generic. Handle it carelessly and the whole thing reads as a mailout.

One idea I haven't seen anyone use, and I reckon it would work: the first treatment anniversary. A year to the day since someone first walked in is a real moment. Nobody is using it. Treat it as a suggestion worth testing, not as something anyone has proven.

Everything outside allied health and medical. Normal reactivation applies. Depending on the service or the product, there are real openings to get back in touch with past customers. Past leads who showed interest and never bought are usually the strongest group on your list.

One line of compliance before you send anything. The moment a message promotes a service, rather than just confirming an appointment, it's advertising. The rules further down apply to it.

Web chat, the quieter cousin of the phone agent

Web chat takes the same three questions off your front desk all day. Opening hours. Whether you take a particular funding type. Where to park. There are clinic-specific products in the Australian market, such as Cliniq Apps. I name it for context, not as a recommendation, because I haven't tested it.

The market itself blurs the line between web chat and a phone-answering agent. That's part of why the two terms get swapped. If your real problem is the phone ringing out, that's a different product with a different answer. It's covered in AI receptionists for medical and allied health clinics.

Where the AI assistant sits, and what it's allowed to see

The other thing people mean by conversational AI is an assistant you can talk to inside your own systems, about your own stuff.

A practice management system is the software your clinic runs its bookings, clinical notes and patient records in. Today, an assistant like this mostly comes bundled inside one, rather than bought on its own. Splose has it inside their practice management software. (We have worked with Splose in the past.) I've also been told about another vendor's assistant that took more of the user's time than it saved. Worth remembering before you assume a bundled feature is a free upgrade.

What matters is where the assistant is pointed.

An assistant pointed at your procedure documents is a fine thing to have. An assistant pointed at your patient database is a different thing, and in practice it has to be built for the job. Treating them as one purchase is how clinics get into trouble. Ask any vendor which of the two they're selling you.

Is a reminder text spam?

Not if it's only a reminder. It is if it also sells something.

Avant draws the line precisely: "If the whole or part of the purpose of the reminder text message is to advertise or offer goods or a service at the practice (such as a flu shot), you must comply with the Spam Act (Cth) 2003 and the Privacy Act (Cth) regarding direct marketing" (Avant).

Once a message is commercial, ACMA sets out what follows. "If you plan to send marketing messages or emails, you must first have consent from the person who will receive them." Every message has to "identify you as the sender", "contain your contact details", and "make it easy to unsubscribe". Unsubscribe requests get honoured "within 5 working days" (ACMA).

New Zealand reads the same way. The Unsolicited Electronic Messages Act 2007 covers "email, fax, instant messaging, and mobile/smart phone text (TXT)". A message that will "facilitate, complete or confirm a commercial transaction that the recipient previously agreed to" isn't a commercial electronic message at all, says the Department of Internal Affairs (NZ DIA). Whether an appointment reminder is caught by the Act is still unsettled over there. So if your recall messages carry anything promotional, go and get your own advice. Don't rely on a blog.

The practical rule: keep the reminder a reminder. Put the flu shot campaign in its own message, with consent, a sender name and an unsubscribe link.

Where Opal helps

We don't sell this software, and we don't take a cut on what gets recommended. The work is deciding what your practice should say, through which channel, how often. Then it's making the systems you already pay for actually do it. That's AI agents and assistants, and it usually starts by cutting messages rather than adding them. If your question is filling the book rather than tidying up the messages, that sits under AI for growth.

If AI scribes are the tool actually on your desk right now, start with choosing an AI scribe and rolling one out across the practice instead.

FAQ

Do I need patient consent to send appointment reminders by text?

Yes, and check for text specifically. Avant's guidance is that while "you do not need specific consent to contact a patient by telephone", you "should specifically check whether patients agree to receiving text messages" (Avant). Get it on the registration form for a new patient, when you confirm an appointment for someone already on the books, or through an online booking portal.

Is a reminder text message spam under the Spam Act?

Only if it advertises something. Avant's position is that where "the whole or part of the purpose of the reminder text message is to advertise or offer goods or a service at the practice", the Spam Act 2003 and the Privacy Act apply (Avant). ACMA sets what follows once a message is commercial. Get consent first. Name yourself as the sender, give contact details, and honour an unsubscribe within 5 working days (ACMA).

How do patients prefer to be contacted for recalls and reminders?

The only Australian data I found is HotDoc's own survey of over 1,000 Australians, from November 2019. It found 55% preferred SMS for follow-ups, against 24% for phone, 20% for email and 1% for post (HotDoc). The preference held across age until 75 and over. It's a vendor's own survey, nobody audited it, and it's dated. Treat it as a starting assumption, and check it against your own patient mix.