The distinction that matters in this setting is not between good text and bad text. It is between documents that describe logistics and documents that describe what will happen to a person's body. The first category can be drafted. The second cannot be originated by a language model, and the review burden is high enough that drafting it often costs more than writing it.
Three document types carried the value: appointment logistics, consultation follow-ups, and review requests. Four categories must stay with a clinician: outcome claims, contraindications, recovery timelines, and anything about devices or products.
Where it helped
Appointment logistics
Where to park, which entrance, what to bring, how long to allow, whether to arrive with a clean face, what happens if you are late. This is pure information, it is identical for most clients, and generated versions were consistently clearer than the handwritten versions we replaced. Arrival instructions and parking are the two details clients most often phone about, and both are easy to get right once.
Consultation follow-ups
After a consultation, the client has questions they did not ask in the room. A short follow-up that restates what was discussed, notes that a decision is not required today, and names who to ask — drafts handled this well because the structure is standard. The one rule: the summary of what was discussed must be written from your notes, not generated from a template.
Treatment descriptions (non-clinical parts)
The parts of a treatment page that are not about the treatment: what a first visit looks like, how many visits are typical, what the appointment involves in sequence, payment and membership terms. Keep the clinical description separate and clinician-written, and let the generated material handle everything around it.
Where it failed — and where it must not be used
Outcome claims
Drafts produced statements about what treatments would achieve — smoother, tighter, gone — with a confidence nothing supported. In this category, that is not a copywriting problem; it is a regulatory one. We removed all outcome language from generated material and left it entirely to the clinician, with no exceptions. This is the category we would not automate even with review.
Contraindications and screening
Generated lists of who should not receive a treatment were plausible, incomplete, and in some cases wrong in ways that mattered. Screening content comes from your clinical protocols and the clinician's assessment of the individual. A model cannot know a patient's history, medications or conditions, and no amount of editing fixes content built on missing information.
Recovery timelines
Drafts gave specific recovery periods. Recovery varies by individual, by treatment and by aftercare, and a stated timeline becomes an expectation. Our rule: aftercare sheets describe what to watch for and when to call, never how long healing should take.
Device and product claims
Generated text referred to devices and products with attributes — approved, cleared, clinically proven — that we could not verify. Any statement about a device or product must match the manufacturer's own documentation, quoted or paraphrased from it, with nothing added.
What patients actually read
Someone considering a treatment reads the aftercare sheet more carefully than any other document. What they want is knowing what is normal and what is a reason to call. In our testing, aftercare sheets organised that way — expected sensations, then signs that warrant contact — drew fewer anxious calls than sheets organised chronologically by day.
The second thing they read is the number of sessions and the total cost, stated plainly. Pages that deferred this to a consultation converted worse than pages that gave a range.
Two document types are worth maintaining separately from the rest. A screening explanation, written by the clinician and issued before a first treatment, so the client's questions are answered in one place rather than across a message thread. A consent explanation in plain language, which the clinician walks through — generated text can organise the layout, but the wording of a consent document needs reviewing clause by clause.
Editing time, from our 30-day benchmark
Appointment logistics went from about 20 minutes to 5, near $8 saved at $30 per hour. Consultation follow-ups dropped from roughly 15 minutes to 4. Review requests from 8 to 2. Clinical content showed no saving and was not handed to a tool.
A workflow that holds up
Generate the logistics and follow-up structures once and reuse them. Request reviews after the follow-up appointment, not the treatment. Outcome claims, contraindications, recovery timelines and device statements stay with the clinician — written from your protocols, reviewed against the individual assessment, and signed off by whoever carries the accountability.
Draft the logistics, not the medicine
Parking, timing and follow-ups scale. Anything about what a treatment will do does not.
Try Rytr →Frequently Asked Questions
Can AI write our treatment descriptions?
The non-clinical parts — visit sequence, duration, payment terms. The clinical description stays with the clinician.
Can it write aftercare instructions?
It can format them. The content — what is expected and what warrants a call — must come from your clinical protocols.
What about before-and-after claims?
Not generated. Outcome language in this category carries regulatory risk regardless of how carefully it is edited.
Can it screen patients for contraindications?
No. A model cannot access a patient's history, medications or conditions.
Can it state recovery times?
Our rule is no. Recovery varies by individual, and a stated timeline becomes an expectation.
Is it useful for a single-location spa?
Yes — appointment logistics and review requests are where the time goes.