Field Notes2 August 20268 min read
What NZ health and wellness businesses are actually doing with AI
From physio clinics to personal trainers, AI is showing up in NZ wellness businesses in three specific places.
From physio clinics to personal trainers, AI is showing up in NZ wellness businesses in three specific places.
The NZ health and wellness sector has a size problem, but not the one you’d expect. It’s not undersized: physiotherapy, osteopathy, personal training, nutrition, and allied health collectively employ tens of thousands of New Zealanders across hundreds of small practices. The problem is that most of those practices are run by one or two practitioners who are also their own marketing department, admin team, and scheduling coordinator. A physiotherapist running a four-room clinic in Hamilton is a clinician for forty hours a week and a small business operator for another fifteen.
That structural constraint, the clinical load plus the operational load carried by the same person, is where AI tools are making their first measurable difference in NZ health and wellness businesses. Not in clinical decision support, which remains appropriately regulated and human-led, but in the surrounding work that was always administrative: communicating, educating, scheduling, and getting found.
The starting point for most NZ health businesses looking at AI is not automation. It is visibility. When a Wellington resident tears their ACL and asks a language model to recommend a physiotherapist who specialises in sports rehabilitation in Te Aro, the practice that appears in that answer is not necessarily the best one. It is the one with enough specific, structured, publicly available content that an AI search tool can confidently cite it.
AI search optimisation changes the calculus for health businesses because the search behaviour for health services is inherently conversational. Nobody types “physio” in quite the same way they ask “which Wellington physiotherapist treats ACL injuries and sees patients on Saturdays.” But they do ask those longer, more contextual questions to AI search tools, and the practices that have written plainly about their clinical focus, their techniques, and the specific conditions they treat regularly are far better positioned to appear in those answers.
A Christchurch osteopathic practice we worked with had a six-page website with about 400 words across all pages. Most of it described what osteopathy is in general terms, rather than what this specific practice does, for whom, and how to book. After a content overhaul that produced twelve pages, including condition-specific content for lower back pain, headaches, and pregnancy-related musculoskeletal issues, plus a structured FAQ drawn from the questions patients most commonly asked at intake, the practice’s organic search sessions increased 87% over five months. Booking enquiries from new patients more than doubled.
The AI-drafted content itself took roughly two hours to produce per page: a practitioner recorded a ten-minute voice note describing what they wanted a new patient to understand, a language model drafted the page from that brief, and the practitioner reviewed it for clinical accuracy before anything went live. The constraint was never the writing. It was that the writing had never been prioritised against clinical work. AI drafting changes the economics of that trade-off without removing the practitioner’s judgment from the final content.
The most immediate time saving we see in health businesses is patient communication: pre-appointment reminders, post-appointment follow-up notes, and recall sequences for patients who should be returning but haven’t booked. These are high-volume, low-judgment messages that follow consistent patterns and consume significant administrative time when handled manually.
An AI automation layer handling this communication can run without staff involvement once the logic is set up. The setup investment is real: mapping which client types receive which sequences, when, and with what content. But once that logic is established, it runs consistently without someone remembering to send a text or an email.
A personal training studio in Auckland with 120 active clients was sending post-session follow-up messages manually. At 35 sessions per week, that represented roughly three hours of the studio owner’s time. With a structured automation layer, those messages generate from session notes the trainer logs immediately after the session, which they were already doing. The follow-up goes out within the hour without additional action. Over a year, that is approximately 150 hours returned to client-facing work or, more accurately, rest.
The harder part of building this layer is maintaining it. Sequences that go stale, recall messages that fire for clients who stopped treatment by choice rather than by lapsing, follow-up content that no longer reflects what the practitioner is actually doing in clinic: these require periodic review. The automation handles the repetition; the practitioner still needs to audit the content once a quarter. Businesses that set the system up and forget about it for twelve months find it has quietly become inaccurate in ways that damage rather than support patient relationships.
Health practitioners hold significant expertise that most of their patients would benefit from between appointments: exercise progressions, dietary adjustments, postural habits, condition management strategies. Most of that knowledge does not get shared because producing written or video content takes time that does not exist in a clinical week.
AI drafting makes this tractable. A physiotherapist can describe a home exercise programme for a common presentation in ten minutes of recorded notes. A language model produces a draft the clinician then reviews and edits for clinical precision. The result is a content asset that can be shared with every patient presenting with the same condition, published on the website for AI search value, and reposted on social media as educational content over time.
The constraint is clinical accuracy. Health content has to be right. The review step is not optional and is not a formality: language models produce plausible-sounding clinical language that may describe correct general principles while getting condition-specific details wrong. Every piece of health content produced with AI assistance needs a practitioner’s review before it reaches patients. The time saving is in drafting. The judgment remains human.
NZ health practitioners operate under professional body guidelines that govern what claims can be made in promotional content. Physiotherapy New Zealand, the New Zealand Nutrition Society, and equivalent bodies publish advertising standards that health businesses must observe. This creates specific constraints on AI-generated social media content.
The failure pattern we see: a health business generates promotional content with AI, publishes it without clinical review, and the content includes outcome claims that exceed what the evidence supports or what the professional body permits. The risk is not only regulatory. It is reputational in a market where a practitioner’s credibility is their primary commercial asset.
What works within those constraints is educational content rather than promotional content. A weekly post explaining the physiology of a common complaint, the difference between two approaches to a condition, or how to assess whether something warrants clinical attention is valuable to patients, publishable within regulatory standards, and makes the practitioner visible as a credible source in a way purely promotional content does not. Social media marketing for health businesses performs best when it treats the audience as capable of understanding real information, rather than delivering messages designed only to prompt a booking.
An Auckland nutritionist we worked with published one educational post per week for six months: each one addressing a common misconception in her field, written from her clinical perspective. At month three, enquiries began referencing specific posts. By month six, she attributed four new client bookings directly to content they had found while researching a health question. The posts were not promotional. They were useful, which in professional services is the more durable form of promotion.
The line that cannot be crossed is clinical. An AI-generated response to a patient query about whether their symptoms require treatment is not appropriate, regardless of how well the system is designed. Practitioners who have tried to use AI customer service layers to handle clinical triage questions find the output either too cautious to be useful or too confident about conditions it cannot assess without examination.
The position that works: AI handles administrative and scheduling communication; clinical questions route directly to the practitioner. That boundary needs to be maintained strictly. Patients who contact the business about a symptom should reach a person, or at minimum a clear pathway to a person. An AI layer that attempts to assess whether they need to come in is a liability, not a convenience.
The second failure mode is depersonalisation in a sector where the relationship is a significant part of the product. Many health businesses, particularly in allied health, operate on long-term relationships with a patient base that chooses them specifically for continuity with a practitioner they trust. Communication that reads as automated breaks that signal of continuity in ways that are hard to recover from. The test for any automated communication is simple: could this message have come from any business in any industry? If the answer is yes, it probably should not come from a health practice where personal continuity is what the patient is paying for.
The NZ health and wellness businesses building the right foundations now are not investing in AI that replaces clinical judgment. They are investing in AI that removes the administrative obstacle between a practitioner’s expertise and the people who need to find them.
The content infrastructure required for good AI search visibility, specific, structured, condition-level information about what the practice does, is the same infrastructure that powers a useful FAQ, accurate patient communication, and better intake forms. Building it to support one function makes every other function work better at no additional cost.
The next meaningful shift will likely be in AI-assisted after-hours voice triage: systems that can handle calls from patients with non-urgent queries outside business hours, capture the nature of the concern accurately, and route appropriately without either dismissing the caller or pretending to provide clinical guidance. The technology is close. Practices building their knowledge base now, whether for website content or patient education, will be ready to extend to that layer when it is reliable enough to deploy. The investment is the same in either case, and it is worth making regardless of what eventually runs on top of it.
The gains in NZ manufacturing aren't on the factory floor. They're in the admin layer that surrounds it.
Three areas where NZ construction firms are finding real returns from AI, and one where most are still leaving time on the table.
Local search, email, and social: where AI is changing the economics of NZ physical retail.