Practice writing splits cleanly in two. On one side: recalls, booking reminders, frame copy, review requests, patient explainers. Low risk, high volume, repetitive — ideal territory. On the other: clinical records, referral letters, anything describing findings or a diagnosis. That side carries professional and legal obligations, and it is where generation should stop.
Where AI clearly helps
- Recall campaigns. Due date, what the recall is for, booking options, what happens if nothing is found. Sent hundreds of times a month.
- Patient explainers. What a prescription means, why lens coatings differ, what to expect from an eye exam — written once at a plain reading level.
- Frame and lens descriptions. Product-shaped copy with changeable inventory and consistent structure.
- Insurance answers. What benefits typically cover and what paperwork is needed — accurate at a general level, kept separate from individual cases.
- Appointment logistics. What to bring, whether drops affect driving, how long it takes, parking. Answers the same ten questions forever.
- Review requests and follow-ups. Short, once, with the link.
- Team documents. Reception scripts for common enquiries, and consistent wording across staff.
Where it nearly got us in trouble
- Clinical findings and records. What was observed and what it means. These are professional records with legal weight; a generated sentence here is indefensible.
- Referral and correspondence letters. Addressed to another clinician about a specific patient. Write them, or generate structure and fill the clinical content entirely by hand.
- Prescription-adjacent statements. Whether a prescription has changed, what acuity was measured, what is "within normal limits". Figures come from the record, not from prose.
- Health claims about products. Blue-light filtering, lens coatings, supplements. Generated benefits tend to be more confident than the evidence supports.
- Insurance promises. "Usually covered" written as "covered". The difference is a phone call three weeks later.
- Urgency language. Recommending someone be seen quickly for symptoms nobody has assessed, or the reverse — reassuring language about symptoms that warrant assessment.
Explaining what you found, in writing
Patients rarely read clinical correspondence, but they do read the message or leaflet you hand them, and that text decides whether someone books the follow-up or quietly worries for six months. Two rules held up. Separate the observation from the significance. "Your prescription has changed slightly" and "this is a normal change at this age, and here is what we will do" are different sentences, and merging them is what generates anxiety. State the next step and the timing — recheck in twelve months, see us sooner if X — because an unexplained wait is where reassurance collapses. Generated explanations tend to hedge in the wrong place: confident about significance, vague about timing. Reverse that.
Editing time, from our 30-day benchmark
Our 30-day test measured roughly 22 minutes of editing per 1,000 words on the low-cost dedicated writer against about 8 minutes on the general assistant — near $7 per piece of your time at $30 per hour. For a single-location practice the saving lands almost entirely in recalls and explainers; everything clinical was untouched, which is where the time actually goes and why the benefit is smaller than vendors suggest.
A workflow that holds up
Draw the line explicitly and write it down for your team: administrative yes, clinical no. Build recalls and explainers as templates with every variable — dates, intervals, products — supplied from your practice records, then check each variable against the source. Keep hazard-free defaults in face of uncertainty: if a symptom could matter, the message should say see us, and no generated text should be reassuring about symptoms nobody has looked at. Store product claims only where you have a supplier source. And review a sample of outbound messages quarterly, because the errors repeat and they will cluster in whatever field nobody checks.
Administrative yes, clinical no
Template the recalls and explainers; keep findings, referrals and any measurement out of the generator.
Try Rytr →Frequently Asked Questions
Can AI write clinical records or referral letters?
No. Those carry professional and legal weight. You can use a template for shape, but every clinical statement has to be written by the clinician.
What practice writing is safe to generate?
Recalls, appointment logistics, product and frame descriptions, general patient explainers and review requests. All repetitive, none clinical.
Can it help with insurance questions?
At a general level — typical coverage and required paperwork. Avoid any promise about an individual claim; say what is usually true and point to the insurer.
What about health claims for lenses or coatings?
Keep every claim to what supplier documentation supports. Generated benefits are usually more confident than the evidence.