Give the AI the vet's own notes and a fixed practice brief, and ask it to rewrite them for the owner: plain words, the pet's name, what happens next, and when to call. The vet who treated the animal then checks every clinical detail before sending. Drafting takes a minute or two; the check takes another one or two.
The rule that keeps this safe is simple: AI rewrites what the vet has already decided, and never decides anything itself. The dangerous parts of an owner email are the doses, the dates, the costs and the prognosis. Those must come from the clinical record, word for word, so the prompt has to forbid the model from adding, softening or rounding any of them.
Which owner emails are worth drafting with AI
Not every message benefits. The best candidates are emails that are written often, follow a pattern, and currently get sent in jargon or not at all because nobody has time. The worst are the ones where tone matters more than information.
| Email type | What AI does well | Who checks before sending | Risk level |
|---|---|---|---|
| Discharge and aftercare instructions | Turns shorthand into numbered steps and a "call us if" list | Treating vet or nurse | High: medication and activity limits |
| Estimate explanations | Explains each line item and why the range is wide | Vet who wrote the estimate | Medium: costs must match the system |
| Test results in plain English | Explains the vet's interpretation without jargon | Treating vet | High: must not add a diagnosis |
| Chronic condition explainers (diabetes, arthritis, kidney disease) | First drafts of a reusable handout the practice approves once | Clinical lead, once, then reused | Low after approval |
| Missed follow-up and recheck nudges | Friendly wording with the reason the recheck matters | Reception can send | Low |
| Complaint replies | Calmer first draft, structure | Practice manager, and the vet involved | Medium to high |
| Anything admitting an error or incident | Leave AI out | Practice owner and your professional indemnity provider | Very high |
Start with discharge instructions. They are frequent, they matter clinically, and owners forget most of what they were told at the desk while also juggling a groggy dog and a lead.
The practice brief you set up once
Every draft should start from the same instructions, so you are not retyping rules and so every vet gets the same tone. On ChatGPT Business, Claude Team or Gemini in Workspace, put the brief into a shared Project (a Gem, becoming a skill from November 2026 in Gemini) and add your approved handouts as reference files. Business plans don't train on your content by default; on a personal plan, switch off the model-training setting in privacy settings before anyone pastes a clinical note.
You help a small-animal vet practice write emails to pet owners.
Rules:
1. Use ONLY the clinical facts in the notes I paste. Do not add
symptoms, drugs, doses, timings, costs or outcomes.
2. Copy drug names, doses, frequencies and dates exactly as written.
If a dose or date is unclear, write [CHECK: ...] instead of guessing.
3. Never promise an outcome. Keep any risk the vet mentions at the
same strength ("small risk" stays "small risk").
4. Reading level: a busy adult with no medical background.
Short sentences. Explain any medical word in brackets once.
5. Structure: what we did, what happens at home (numbered),
when to call us (bullet list), next appointment.
6. Use the pet's name and correct sex. Sign off with the vet's name.
7. Under 250 words unless I ask for more.
8. Our spelling (anaesthetic, haemorrhage). Warm but not chatty.
No exclamation marks.Rule 2's [CHECK: ...] marker matters more than any other line. It turns the model's urge to fill gaps into a visible flag the vet can't miss.
From clinical shorthand to a discharge email owners can follow
Here is how one discharge might run in a two-vet small-animal practice. The figures and case are illustrative. The vet's notes after a foreign-body operation read:
Bella, 7y FN Labrador. Ex-lap today, sock removed from jejunum,
enterotomy, no resection. Recovered well from GA.
Meds: meloxicam 1.5 mg/ml oral susp 2.3 ml SID with food x 5d,
start tomorrow. Omeprazole 20 mg PO SID x 7d.
Cone on at all times. Lead walks only 10 min TID x 14d, no stairs if poss.
Small bland meals 4x day x 3d then normal diet.
Recheck 3d (wound), stitches out 10-14d.
Call if vomiting, not eating >24h, wound swelling/discharge, lethargy.Pasted under the brief with "Write the discharge email for Bella's owner", a model returns something like this (illustrative output):
Bella had surgery today to remove a sock that was stuck in her small intestine. We made a small cut in the intestine to take it out and did not need to remove any section of bowel. She woke up well from the anaesthetic.
At home: 1. Give meloxicam (a pain relief and anti-inflammatory liquid) 2.3 ml once a day with food for 5 days, starting tomorrow. 2. Give omeprazole (to protect her stomach) 20 mg once a day for 7 days. 3. Keep her cone on at all times, including overnight. 4. Lead walks only, about 10 minutes three times a day, for 14 days. Avoid stairs where you can. 5. Feed small bland meals four times a day for 3 days, then her normal food...
Call us if Bella vomits, won't eat for more than 24 hours, seems unusually flat, or her wound looks swollen or has any discharge.
Next: a wound check in 3 days, and stitches out in 10 to 14 days.
What the vet fixed took about 90 seconds. The draft said "starting tomorrow" for the meloxicam but left the omeprazole start date out, so the vet added "starting tonight". It also turned "no stairs if poss" into "avoid stairs where you can", which was right. On a different run, the same prompt produced "She should be back to normal walks in about two weeks", which was never in the notes. That sentence is exactly the kind of kindly invention the checklist further down is designed to catch.
The time maths for a practice sending around 15 surgical discharges a week: writing each from scratch took roughly 10 minutes, and the AI draft plus check takes about 3. That's close to two hours a week back, plus owners who ring less often to ask what "SID" meant.
Explaining an estimate so it reads as care, not a sales pitch
Estimates generate more awkward phone calls than almost anything else, usually because the owner sees a list of codes and a wide range with no explanation. AI is good at turning the list into reasons. Paste the line items and the vet's two-sentence rationale, never the whole estimate history.
A before and after for a dental procedure (illustrative figures). The estimate as exported:
GA + monitoring; IV fluids; dental radiographs full mouth;
scale & polish; extractions (0-4) surgical; analgesia; LA blocks.
Total $780 - $1,240The explanation the owner receives after the AI draft and the vet's edit:
The range on Milo's estimate is wide because we can't see below his gum line until he's asleep. The lower figure covers the anaesthetic, fluids to support his blood pressure, full-mouth X-rays, and a clean and polish. If the X-rays show roots that are damaged or infected, we'll remove those teeth, and each extraction moves the total towards the higher figure. We'll phone you during the procedure before doing anything that would take the cost above the top of this range.
The one thing the vet had to add was that last sentence, a commitment the practice actually makes. AI can't know your policies unless they are in the brief, so add a line to the brief listing the promises you do make about estimates, and nothing else.
A second run of the same estimate shows why figures stay out of the model's hands. Asked to "explain how the range builds up", the draft added: "Each extraction adds about $115 to the total." It had divided the $460 gap between the two figures by the four possible extractions. The sum is right and the claim is wrong: in this illustrative fee schedule, a simple extraction of a small tooth costs far less than a surgical extraction of a large molar with a root that needs sectioning, and the range also allows for extra anaesthetic time. The vet replaced the sentence with the practice's real wording, "extractions are priced per tooth, depending on its size and how it has to be removed", and added a rule to the brief: never calculate or estimate a price; use only prices written in the notes.
Turning test results into plain English without adding a diagnosis
Never paste raw lab values and ask "what does this mean?". The model will interpret them, sometimes plausibly and sometimes wrongly, and the email will then carry an interpretation the vet never made. Paste the vet's interpretation instead, and let AI do the translation.
Write a results email for Pepper's owner (cat, 14, female).
Vet's interpretation: bloods show early kidney changes, mildly raised
values. Not an emergency. Plan: switch to a kidney-support diet over
7-10 days, recheck bloods and blood pressure in 4 weeks.
Owner is anxious. Keep it under 180 words. Offer a phone call.The illustrative output opened with "Pepper's blood test shows some early changes in how her kidneys are working. This is common in older cats and it isn't an emergency", then set out the diet change and the recheck. The vet deleted "This is common in older cats": true in general, but it wasn't in her interpretation, and it slightly shifts the message. That's the level of reading each email needs.
For conditions you explain again and again, write the explainer once with AI, have your clinical lead approve it, and save it as a reference file. Future emails then point to "the kidney care sheet attached" rather than regenerating the explanation, which removes most of the risk.
The kidney sheet in this illustrative practice ended up as one page with five headings: what the kidneys do, in two sentences; what "early changes" means for your cat; the diet switch, with a seven-to-ten-day mixing schedule; signs to watch (drinking more, weeing more, weight loss, poor appetite); and what the rechecks are for. The clinical lead made three changes to the AI's first version: she cut a paragraph on stages of kidney disease that owners at this point didn't need, removed a line suggesting particular supplements, and added the practice's own advice on tempting a cat onto a new diet. The approved version carries a date and a "review by" date a year later.
Recheck reminders reception can send without a vet
Low-risk messages are where reception saves time without any clinical sign-off, provided the prompt keeps them non-clinical. Four weeks on, Pepper's recheck hasn't been booked. The receptionist's prompt, inside the same Project:
Write a short reminder to Pepper's owner. Pepper (cat) is due a
recheck blood test and blood pressure check, which the vet asked
for 4 weeks after starting a kidney diet. It hasn't been booked.
No clinical advice. Under 80 words. Ask them to ring to book.An illustrative draft:
Pepper is now due her recheck blood test and blood pressure check, four weeks after starting her new diet. The recheck lets the vet see how her kidneys are responding to the change. You can book online in a couple of clicks, or give us a ring. We look forward to seeing her.
One fix: this practice has no online booking, so "book online in a couple of clicks" had to go. The model added a common feature of practices it has read about, the same kindly invention as in Bella's email, in a lower-stakes place. The line about "how her kidneys are responding" is fine, because it repeats why the vet asked for the recheck without adding anything. The receptionist adds "Booking: phone only" to the brief so the slip doesn't recur.
The 60-second check before any clinical email goes out
Print this and keep it by the screen. The treating vet or nurse runs it on every draft:
- Drug names, strengths, doses and frequencies match the record character for character. Brand names can differ between markets, so check the model hasn't swapped yours for another one it has seen more often.
- Every date and duration is present and correct: start dates, "for 5 days", recheck dates.
- Nothing new appeared: no symptom, reassurance, timescale or outcome that wasn't in the notes.
- Risk wording is the same strength as the vet's.
- Name, species and sex are right throughout (a "he" in a spayed bitch's email tells the owner it's a template).
- Costs match the practice management system exactly.
- "Call us if" list is complete and the out-of-hours number is correct.
- No
[CHECK]markers remain.
If the same fix keeps coming up, change the brief rather than fixing it by hand forever. Three weeks in, most practices find their brief has grown by four or five lines drawn from exactly these corrections.
Where AI-drafted owner emails go wrong
A few failures show up repeatedly, and each has a tell you can train the team to spot.
- The previous pet leaks in. Someone drafts three emails in one long chat, and the third mentions the second patient's medication. Start a fresh chat for each animal inside the Project.
- Doses get "tidied". "2.3 ml" becomes "about 2 ml" or "half a tablet" becomes "one tablet". The brief forbids this, but it still happens occasionally, which is why item 1 on the checklist exists.
- Cheerful tone for a poor prognosis. The model defaults to upbeat. Add "Tone: serious and kind" when the news is bad, or better, don't use AI for those emails at all.
- Emails get longer. Owners read the first screen. Keep the word cap in the brief and move the detail into an attached care sheet.
- Owner details in the prompt. A receptionist pastes the whole client record including address and card details. Strip those out first; anonymising data before it goes into AI covers a quick routine.
Replies need a rule of their own. Two days after surgery, Bella's owner answers the discharge email: "Her wound looks a bit pink around the stitches, is that normal? She's eating fine." A receptionist working fast could paste that into the Project and get a confident, reassuring answer, which is exactly the clinical judgement the AI must never make. The safe pattern is a holding reply the team has agreed in advance, sent by a person, plus a same-day hand-off to the nurse:
Thanks for letting us know. I've passed your message to our nurse, who will ring you this afternoon. If the wound starts to swell, weep or bleed, or Bella stops eating or seems unwell before then, please call us straight away on [practice number].
AI can help write that template once. It shouldn't be drafting fresh answers to clinical questions from owners, however routine they look.
Rolling it out across the practice in four weeks
Week one: one vet uses the brief for discharge emails only and notes every correction. Week two: update the brief, add estimate explanations, and let a second vet try it. Week three: nurses start drafting results emails from the vet's interpretation, sent only after vet sign-off. Week four: review.
What to measure is simple: minutes per email (time five before and five after), how many owners phone to ask what an email meant, and how many drafts needed a clinical correction. If corrections aren't falling by week four, the brief is too loose or the notes going in are too thin. An illustrative week-four review might show drafting time down from about 10 minutes to just over 3, clarification calls about discharge emails down from nine in the first week to four, and clinical corrections falling from 11 of the first 40 drafts to 5 of the last 45. The five remaining corrections were all missing start dates, which pointed to one more line for the brief: "Every medicine needs a start day; if the notes don't give one, write [CHECK: start]". If you already use AI to draft notes, cutting time on clinical notes pairs naturally with this, because cleaner notes produce cleaner emails. For the wider picture of tools and costs, see AI for veterinary practice owners, and for the routine messages that need no vet at all, vaccination and check-up reminders. Human health practices face the same explanation problem, and explaining dental treatment plans with AI has ideas that transfer well to vet estimates.
Questions vet teams ask before using AI for owner emails
Should we tell owners that AI helped write their email?
There is no single rule, but a short line in your practice privacy notice or website explaining that staff use AI writing tools, and that a vet or nurse checks every clinical message, is honest and rarely raises concern. What owners care about is that a named person stands behind the advice, so keep the signing clinician's name on every email.
Can nurses and receptionists use the same prompts?
Yes, for non-clinical messages such as appointment follow-ups, insurance paperwork explanations and reminders. For anything describing a diagnosis, medication or aftercare, the draft should go to the vet or nurse who treated the animal before it is sent, even if a receptionist prepared it. Build that hand-off into the workflow rather than relying on memory.
Is it safe to paste clinical notes into ChatGPT or Claude?
Use a business plan (ChatGPT Business, Claude Team, Gemini in Workspace or Microsoft 365 Copilot), which does not train on your content by default, and paste only what the email needs: the pet's first name and the clinical points. Leave out owner surnames, addresses, phone numbers and payment details. Check your data-protection obligations with your adviser if you are unsure.
What about euthanasia and bereavement messages?
Write those templates yourselves, slowly, and have the team agree the wording. AI can check spelling or suggest a gentler phrase, but a condolence message generated fresh each time risks sounding generic at the worst possible moment. Many practices keep two or three human-written versions and personalise them by hand with something specific about the animal.
Further reads
- How to Draft Customer Email Replies With AI That Sound Like You — Drafting non-clinical replies that still sound like your practice.
- Can AI Handle Customer Complaints Without Making Them Worse? — Handling unhappy owners without making the complaint worse.
- How to Set Up Human Review for AI Work Without Slowing Down — How to build the vet sign-off step without it becoming a bottleneck.
- How to Set Up Claude Projects as a Shared Team Assistant — Setting up one shared practice brief the whole team drafts from.
- AI Hallucinations Explained for Business Owners: Causes and Fixes — Why AI adds confident details that were never in the notes.
- Patient Data and AI: A Confidentiality Checklist for Small Practices — A confidentiality checklist that translates well to animal patients and owners.
- How to Map Your Customer Journey and Find Where AI Helps — A journey grid, evidence sources and the signs a step suits AI, walked through with a new puppy owner's first year at a veterinary practice.
- Explaining Lens Options and Eye Test Results With AI Help — Prompts and checked templates for explaining prescriptions, lens choices and eye test findings to patients in plain English, with the clinician in charge.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: plan and privacy pages for ChatGPT Business, Claude Team, Gemini in Workspace and Microsoft 365 Copilot (checked September 2026). Clinical examples are illustrative and were not taken from any real practice.