How a Veterinary Clinic Can Speed Up Owner Letters, Booking Confirmations, and Post-Op Care Notes with Claude Code

It’s 6 p.m. at the animal hospital. Three families are still in the waiting room. On the front desk sits one unfinished job: reprinting the “post-op care sheet” for the owner of a cat that just had a spay. Open last patient’s template, swap the name, fix the suture-removal date, add the cone-of-shame note…

Then the phone rings. It’s a booking confirmation for tomorrow. When that’s done, back to the sheet. And somewhere in the name-swapping, one field got missed: a sheet addressed to “Bailey” still had a different patient’s weight on it. I caught it before it went out the door, but my stomach dropped.

I realized that this exact job, “rewriting nearly-identical text by hand every single time, with one tiny thing different,” is something I can hand off to Claude Code in large part. This is a practical, hands-on look at how a vet clinic front desk and its veterinarians can stop burning energy on owner-facing writing.

Key takeaways

• Of all the writing a clinic does, “owner explanation letters, booking confirmations, and post-op care notes” follow a fixed template, so an AI draft works well.
• What you hand to Claude Code stops at “drafting and formatting.” Medical judgment and the final check are always done by a person.
• I’ve included copy-paste prompt templates and a check-script that generates a batch of templates at once.
• Owner names, contact details, chart numbers, and other personal data are, as a rule, never given to the AI. They go in “merge fields” filled in afterward.
• If 30 minutes of daily writing drops to 10, that’s about 7 hours a month, roughly one front-desk person’s load lifted.

Who this is for: a picture of the clinic floor

The reader I have in mind runs a neighborhood animal hospital with one to three veterinarians and a handful of staff. The front desk juggles bookings, billing, phones, and letters all at once, and the vets write up explanations and referral letters between appointments. That’s the scale I’m picturing.

The pain point for this group is clear. The writing is “not hard enough to start from scratch, but quietly heavy to redo every time.” Vaccine reminders, dental-care nudges, senior wellness-check outreach, pre-op fasting instructions, post-op precautions. You know every word of it by heart, yet putting it into print eats five or ten minutes each go.

Breaking down the clinic’s writing workflow

First, let me lay out how it works now. Most clinics will recognize this.

1. The exam or procedure ends.
2. The vet explains things to the owner out loud.
3. The front desk or the vet writes up the sheet to hand over, or the message to send.
4. Patient-specific details, name, dates, drug amounts, get dropped in.
5. It’s printed, or sent via the clinic’s text/email channel.

Steps 3 and 4 eat the time. Step 3 especially, “writing it up,” is where many clinics keep tweaking a similar document by a hair and reusing it. That’s exactly the part an AI draft fits best.

By contrast, step 2 (the explanation itself) and step 4 (entering that animal’s numbers) are human territory. Blur those into the AI’s job and you cause accidents. I’ll get to why below.

Use case 1: drafting an owner explanation letter

Preventive-care nudges can backfire: even when the content is correct, if it lands as “preachy” or “fear-mongering,” it does the opposite of what you want. Tone tuning is precisely where AI shines.

For a dental-care letter, for example, a prompt like this produces a draft.

You are the front desk at a neighborhood animal hospital. Write an owner-facing
"start a tooth-brushing habit" letter.
Conditions:
- Audience: the owner of a 3-year-old small-breed dog
- Tone: gentle, never blaming, never fear-mongering
- Readable in under 30 seconds; 120 words or fewer
- Include the "don't force it, start by getting them used to having their mouth touched" stage
- Don't pressure them into a visit; close with "feel free to ask us anytime"
Do not include any proper nouns or personal data. Use [Owner Name] for the name
and [Breed] for the breed as merge fields.

The key is that last line. Don’t enter the real owner or pet name; leave merge fields like [Owner Name]. Do this and you can save the output as a reusable template, and you never hand personal data to the AI.

Use case 2: mass-producing booking confirmations

Booking confirmations are a block of boilerplate. That’s exactly why asking the AI to “make several variations at once” is a huge relief. First-visit, recheck, vaccine, and pre-op each need different cautions.

The table below organizes what changes by type. Paste it straight into the AI prompt and you’ll miss fewer items.

Booking type	Cautions to include	Merge fields
First visit	What to bring (past lab results), pre-visit stool sample	Date/time, clinic name
Recheck	Bring notes on changes since last visit	Date/time, attending vet
Vaccine	May postpone if unwell that day; rest on the day	Date/time, vaccine type
Pre-op	Fasting/no-water start time, consent form	Date/time, fasting start time

Keep the merge fields separate and you get a safe split of labor: the AI makes the “template” only, and a person enters the specific numbers, date/time, fasting start time, afterward.

Use case 3: lining up post-op care notes by case

Post-op notes are the area where a mistake is scariest. So I hand the AI “the skeleton of the text and its readability” only, and the veterinarian locks down the medical substance.

The split I use is this checklist.

• The vet fills in dosing frequency, amount, and duration (never let the AI invent numbers)
• The cone-of-shame and rest period are confirmed by a person to fit the case
• The “call us right away if this happens” danger signs are pasted in from the clinic’s own standard
• Suture-removal and recheck dates are left as blank merge fields in the output
• The AI is only asked for “gentle, short, bulleted”

Even just building the skeleton is far faster than writing on a blank page. The safety of the substance stays in human hands while only the writing chore gets cut.

What to delegate to the AI, and what a person must decide

This is the most important part, so I’ll draw the line with a table.

Task	Delegate to Claude Code	A person must decide
Explanation letter draft	Structure, tone, length	Whether the content is medically correct
Booking confirmation	Mass-producing and formatting boilerplate	Specific values like date/time, fasting time
Post-op care note	A readable skeleton	Dosing amount, danger signs, durations
Multilingual support	First-pass translation into, say, Spanish	Final check of clinical terms

The principle behind the line is simple. Anything that, if wrong, affects an animal’s body or causes a dispute with an owner is decided entirely by a person. The AI stays the “make it fast and readable” assistant. Not blurring this boundary is the foundation for using AI safely in a medical setting. For the basic way to ask, Claude Code for non-engineers is a good reference.

What changed before and after

Before, making a single letter meant “find a similar past document, open it, edit it, double-check the proper nouns,” for five to ten minutes. With several a day, the front desk’s focus got chipped away in fragments. Because letters got written between phone calls and walk-ins, swap-the-wrong-field mistakes happened easily.

After, I have the AI pre-build templates for the letters we use most, saved with only the merge fields left blank. On the day, you just fill in the name and date. The time spent “thinking about the wording” dropped to near zero. When a new kind of letter is needed, throwing one prompt gets the foundation back in 30 seconds.

A rough ROI estimate

Let me do quick math. Say a front desk that spent 30 minutes a day on letters and confirmations gets that down to 10 minutes a day by templating.

• Time saved: 20 minutes/day x 22 working days/month is about 7.3 hours/month
• At a front-desk wage of, say, $18/hour, that’s roughly $130/month of labor cost
• The freed-up time goes to checking in with the waiting room or verifying billing

Bigger than the dollar figure is that the front desk stops having its head split open by writing. Mistakes drop, and the care given to owners who walk in gets more attentive. That’s where it pays off at a vet clinic.

Copy-paste: a check-script that generates templates in a batch

For anyone who finds “typing a prompt every time” a hassle too, here’s a setup that drafts a list of common letter types all at once. It runs with Node.js and an Anthropic API key. The output is saved to files, so afterward you just fill in the merge fields.

First, set up.

mkdir clinic-letters && cd clinic-letters
npm init -y
npm install @anthropic-ai/sdk

Next, the main file (generate.mjs). List the “letter types” in an array, and it writes each draft into out/. It hands over no personal data at all and is instructed to leave only merge fields.

import Anthropic from "@anthropic-ai/sdk";
import { mkdir, writeFile } from "node:fs/promises";

const client = new Anthropic();

// List only the letter types you want to make. No personal data.
const letters = [
  { id: "vaccine-reminder", topic: "Dog combination-vaccine reminder" },
  { id: "dental-care", topic: "Encouraging a tooth-brushing habit for cats" },
  { id: "post-surgery", topic: "At-home care after a spay (skeleton only)" },
];

const rules = [
  "You write owner-facing letters handed out at a neighborhood animal hospital front desk.",
  "Gentle and short; never blame, never fear-monger. 120 words or fewer.",
  "Owner and pet names go in [Owner Name] and [Pet Name] merge fields.",
  "Specific numbers like dates, drug amounts, and fasting times: do not write a value; leave a blank [   ].",
].join(" ");

await mkdir("out", { recursive: true });

for (const letter of letters) {
  const res = await client.messages.create({
    model: process.env.ANTHROPIC_MODEL || "claude-sonnet-4-6",
    max_tokens: 800,
    system: rules,
    messages: [{ role: "user", content: `Topic: ${letter.topic}. Write a draft of this letter.` }],
  });
  const text = res.content.find((b) => b.type === "text")?.text ?? "";
  await writeFile(`out/${letter.id}.txt`, text, "utf8");
  console.log(`Written: out/${letter.id}.txt`);
}

Running it is just this.

node generate.mjs

Three drafts appear inside out/. Every specific value comes out blank, so before printing the front desk fills in the name and date, and for post-op care the vet locks down the drug amount. That’s how you safely cut only “the chore of writing.” When you want to add a type, just add one line to the letters array. When you want to raise prompt accuracy, advanced prompt engineering helps.

By the way, writing your letter rules (tone, word count, merge-field rules) into the project’s CLAUDE.md saves you from re-stating instructions every time. How to do that is in CLAUDE.md best practices.

Privacy and security notes

At an animal hospital too, owner names, addresses, phone numbers, and chart numbers are personal data. Treat the pet’s name as information that can help identify the owner; that’s the safe stance.

Remember just one line to defend. What you give the AI is the “template” only. Specific information goes in merge fields, filled in later by a person. That’s exactly why the prompt and script above use the merge-field method.

On top of that, if you use this for work, choose a plan that doesn’t train on your input data, or use it via the API. Avoid pasting owner information into a free consumer chat. Just setting one in-clinic rule, “no personal data goes into the AI,” prevents the majority of accidents. For a fuller take on safe use, also read the Claude Code getting-started guide. You can confirm the official details in Anthropic’s privacy policy.

FAQ

Q. Can a front-desk person who isn’t great with computers use this? A. For just drafting letters, you only copy-paste the prompt and send it. The script can be built once by the clinic owner or a tech-savvy person, and after that it runs by “adding one line per type.”

Q. Is it OK to hand over the medical content the AI wrote as-is? A. No. Drug amounts, danger signs, rest periods, anything that touches the body in numbers and judgment, must be locked down by the veterinarian. The AI handles the skeleton and readability of the text.

Q. Wouldn’t it be easier to have it written with the owner’s name in it? A. Easier, yes, but that means handing personal data to the AI. With the merge-field method you can reuse the template and there’s no leak risk. Over the long run this way is both faster and safer.

Q. Can I use it for English-speaking owners, or Spanish-speaking ones? A. For a first-pass translation, yes. But have a person check the clinical terms and cautions at the end. For daily efficiency tips, productivity tips is also a good reference.

What I confirmed when I actually tried it

I actually had the script draft three types, dental care, vaccine, and post-op care. I checked two things. First, whether it really returns specific numbers as blanks. With the [ ] instruction in place, it didn’t invent drug amounts or dates; it returned proper blanks. That matters a lot for safety.

Second, the tone. Just adding the one line “no blaming, no fear-mongering” softened the dental-care letter considerably, from “if you don’t, the teeth will go bad” to “start by getting them used to having their mouth touched.”

Honestly, at first I thought “letting an AI touch post-op notes is scary.” But once I drew a clean line, have it build only the skeleton and keep the substance in human hands, the worry vanished. The biggest win is that, with less time spent thinking about wording, I have room again to actually talk to the owners. If you want to make this a rule across your clinic, you can work through the operational design with us at the training and consulting desk.