Cut Phone and Web Inquiry Time at Your Family Medicine Clinic with Claude Code

It’s 30 minutes into the morning clinic. The front-desk phone has rung three times, and every call is the same: “Do I need an appointment for a flu shot?” “Can my kid get a strep test today?” “My son has a fever, how late are you open?”

The waiting room is full, the practice-management screen is still open, and the phone keeps ringing. The moment you hang up, there are five more nearly identical questions sitting in the web contact form.

I watched this play out at a family medicine and pediatrics clinic a friend runs. The doctor and the front-desk staff are all sharp people, yet every single day they lost one to two hours to the same repeated questions.

What I helped them build was not a clever AI receptionist bot. It was a foundation: writing down the common questions and the right way to answer them once, so anyone at the desk responds with the same quality. Today I’ll walk through the whole process in a copy-paste-ready form.

Key takeaways

• At a family medicine or pediatrics clinic, about 80% of inquiries are the same 20-30 questions. Just inventorying them makes phone duty noticeably lighter.
• Hand Claude Code the drafting work: FAQ drafts, phone scripts, and web reply templates. A human always does the final check.
• Never feed the AI patient names, symptoms, or phone numbers. You only give it the shape of the question.
• A realistic target is 15-20 hours saved per front-desk staffer each month, plus lower training cost.
• A check script can mechanically scan your FAQ for forbidden wording (diagnoses, absolute promises) before anything goes out.

Where the front desk actually loses time

Let me be specific about who this helps:

• The owner of a family medicine or pediatrics clinic that’s been open a few years, with two to five staff.
• The medical front-desk person who juggles reception, phones, and appointment scheduling.
• Anyone who feels “I can’t justify hiring, but the current phone load is already at its limit.”

Inquiries at this kind of clinic have a particular character. In pediatrics, anxious parents call, so the questions come out urgent and emotional. Family medicine covers a wide range: appointments, tests, prescriptions, checkups. And many questions depend on today’s situation - fever-clinic hours, vaccine stock on hand - so the answer changes day to day.

That’s exactly why “write one manual and you’re done” rarely works here. This is where the AI earns its keep.

The typical workflow, and where rework creeps in

A typical inquiry flow looks like this:

1. The phone rings, or an inquiry lands in the web form.
2. The front desk listens and takes down the question.
3. If they can answer on the spot, they do; if unsure, they check with a nurse or the doctor.
4. If a callback is needed, they note the patient’s details and call back later.
5. The same question shows up again the next day.

Rework creeps in mainly at three spots:

Rework point	Example	What happens
Inconsistent answers	The same question gets different answers from different staff	Patients get confused; complaints brew
Waiting to confirm	Too many “I have to ask the doctor first”	More callbacks, longer calls
Nothing is recorded	Handled verbally and forgotten	Every time starts from zero

All three mostly disappear once you write the question-and-answer pairs down properly one time. By hand that’s a half-day job; with Claude Code drafting it, you have something in one or two hours.

Use case 1: Inventory the common questions and draft the FAQ

The first step is to make the AI dump out every “usual question” in your head, then fill the gaps. Front-desk staff are busy, and they’re surprisingly unable to articulate what they actually answer all day.

The trick here is to teach the AI about your clinic before it writes. A generic FAQ is useless.

Copy-paste prompt template

You are an assistant who knows family medicine and pediatrics front-desk work well.
Using the clinic information below, list 30 questions patients commonly ask by
phone or web, grouped by category, with a spoken answer the front desk can use.

[Rules]
- Do not make diagnoses or state treatment plans (no medical judgment).
- For questions that should prompt a visit or a doctor's confirmation, always
  include that sentence.
- Keep each answer to 3 sentences or fewer, in natural spoken English.
- For pediatric questions, use a gentle tone that reassures anxious parents.

[Clinic information]
- Specialties: Family medicine and pediatrics
- Hours: Mon-Fri 9:00-12:30 / 15:00-18:00, Saturday morning only
- Vaccinations: By appointment, Tue/Thu 14:00-15:00
- Fever clinic: Separate entrance, call before arriving
- Online booking: Yes (follow-up visits only)
- Parking: 3 spaces, partnered paid lot nearby

Output as a table: Category / Question / Draft answer / Needs doctor check (yes/no).

This gives you a 30-question first draft in one shot. Then you sit with the front-desk staff and fix it: “that’s wrong,” “add this one too.” Editing is far faster than writing from scratch, and you catch the gaps you’d otherwise miss.

What to delegate to AI vs. decide yourself

Get this fuzzy and you’ll have an accident. Draw a clear line:

Step	Safe to delegate to AI	A human must decide
Finding the questions	Yes - great at filling gaps	Whether the categories make sense
Drafting answers	Yes - first drafts	Whether it’s medically correct
Adjusting tone	Yes - softer phrasing	Whether it fits clinic policy
Publishing/posting	No	Yes - owner gives final sign-off

One iron rule: never put AI-written medical-sounding text in front of a patient without a human check. The AI writes plausible falsehoods with total confidence. Whether it’s correct as front-desk wording must be reviewed by the doctor or a nurse, every time.

Use case 2: Build a first-response phone script

Once the FAQ exists, turn it into a script you can read aloud on the phone. When a new front-desk hire starts, just having this script makes their ramp-up dramatically smoother.

Here’s the prompt I tested:

Using the FAQ table below, create a first-response phone script for the front desk.
For each question, break it into four steps:
"opening line -> what to confirm -> answer -> closing line".

[Must include]
- For high-urgency complaints (fever, seizures, difficulty breathing, etc.),
  always add a branch that does NOT answer and instead directs the caller to
  "seek care now / go to the ER".
- For cases needing a callback, list what to collect (name, patient ID,
  contact number, brief description of symptoms).
- For pediatric calls where a parent sounds anxious, open with a line of empathy.

The branch that sends high-urgency complaints straight to care instead of answering them can be a matter of life and death in a clinical setting, so this is the one part you must never drop. Even after the AI writes it, the doctor checks it word for word.

What changed, before and after

Here’s the rough before/after from my friend’s clinic:

• Before: 4-5 minutes per call on average. New hires stalled and checked with a senior staffer every time. The same questions came in 10+ times a day.
• After: 2-3 minutes per call, reading from the FAQ and script. New hires answer on the spot. For routine web-form questions, the staff just copy the draft reply and send it.

In time terms, that’s roughly 40-60 minutes per staffer per day, or 15-20 hours a month freed up. Even valued at an hourly wage, that’s a few hundred dollars a month, while the AI itself costs a dollar or two. ROI was never something we had to agonize over.

Use case 3: Draft replies to web inquiries

Web-form inquiries take longer than phone calls because you have to reply carefully in writing. The AI speeds this up too.

But do not paste the patient’s actual message into the AI (reasons below). Instead, tell it only the type of question, have it generate templates, and let a human fill in the proper nouns by hand.

Create 5 web-inquiry reply templates for a family medicine and pediatrics clinic.
Types: (1) how to book, (2) confirming hours, (3) booking a vaccination,
(4) how to use the fever clinic, (5) booking a health checkup.

[Rules]
- Open with thanks; close with "Take care / We look forward to your visit", etc.
- Don't touch personal information or make symptom-related assertions.
- Mark spots a human fills in later with double braces, like {{name}} {{date}}.

The front desk just drops the patient’s name and date into the returned template and sends it. A reply that took 3 minutes drops under 1.

Privacy and security you can’t skip

This is non-negotiable for a medical practice, so it gets its own section.

• Do not paste a patient’s real name, ID, phone number, or specific symptoms into the AI input box. Sending it to a cloud AI service = sending it outside your walls.
• Give the AI only the shape of the question and your clinic’s public info (hours, etc.). Never hand over medical records or billing data.
• Before publishing any FAQ or script, confirm it contains nothing that could identify a specific patient.
• Put a one-page rule sheet on the wall: what staff may enter, and what they may never.

In short, use the AI as a writing tool, never as a processor of individual patient data. Hold that line and you stay well within the spirit of health-data safety guidance. In the US, the HHS HIPAA guidance for professionals is worth reading so everyone understands the rules.

A check script that screens your FAQ mechanically

Finally, add one gatekeeper that doesn’t rely on human eyes alone. This script scans the FAQ for absolute or diagnostic phrasing the front desk should never use (“it’s definitely…”, “you’ll be completely fine”). It runs on Node.js.

Export your FAQ to a CSV (faq.csv, columns question,answer), then run:

import { readFile } from "node:fs/promises";

// Phrasing that's dangerous if the front desk uses it casually
// (medical judgment, absolutes, over-reassurance)
const NG_WORDS = [
  "definitely", "you'll be fine", "no need to worry", "nothing to worry about",
  "it's a", "diagnos", "just take the medicine", "safe to ignore",
];

const csv = await readFile("./faq.csv", "utf8");
const rows = csv.trim().split("\n").slice(1); // first line is the header

let hit = 0;
rows.forEach((line, i) => {
  const answer = line.split(",").slice(1).join(",").toLowerCase();
  const found = NG_WORDS.filter((w) => answer.includes(w));
  if (found.length > 0) {
    hit++;
    console.log(`[fix needed] line ${i + 2}: ${found.join(" / ")}`);
    console.log(`  -> ${answer.trim()}`);
  }
});

console.log(hit === 0
  ? "OK: no dangerous absolute phrasing found."
  : `${hit} item(s) need review. Send them to the doctor.`);

Run this once before publishing and you prevent a dangerous line like “you’ll be completely fine” from slipping into patient-facing material. Pair it with the checklist below as your pre-publish gatekeeper.

Pre-publish checklist (print it and pin it at the desk):

• No medical assertions or diagnoses?
• Is there a branch that directs emergencies to “seek care / ER”?
• Nothing that could identify a specific patient?
• Do hours, booking methods, etc. match current operations?
• Did the doctor or a nurse give final sign-off?

FAQ

Q. Can I use this without programming knowledge? A. To build FAQs and templates, you just paste the prompt into a browser chat. Only running the check script takes a bit more effort, so start with the writing side. For the full picture of getting set up, the getting-started guide for first-time Claude Code users is a good reference.

Q. Will this transfer to a new front-desk hire? A. It actually makes handoff easier. The script and FAQ live in writing, so you spend less time training verbally. The mindset for making things usable by non-technical staff is covered in Claude Code for non-engineers.

Q. What if the AI produces wrong medical information? A. That’s exactly why “a human always checks before publishing” is built into the process. The AI is a drafter; final responsibility is human. The check script plus the doctor’s review is a two-layer guard that stops mistakes.

Q. I want to tailor the prompt more to my own clinic. A. The more clinic-specific detail you provide, the better the accuracy. For tips on writing prompts well, see advanced prompt engineering in practice, with concrete examples. Once you’ve got a working FAQ, productivity tips help you keep the loop fast.

What happened when I actually tried it

I ran the prompts above at my friend’s pediatrics clinic and built a 30-question FAQ and a phone script in half a day. The biggest win was that the inventory step surfaced seven or eight questions nobody realized they were answering all the time. Questions the front desk handled on autopilot finally became words on a page.

I ran the check script too, and it caught “no need to worry” buried in two of the AI’s draft answers. In a clinical setting that single phrase is dangerous. That was the moment I was glad I’d put a mechanical gatekeeper in place.

When I checked back two weeks after rollout, they told me they “no longer brace themselves for the repeat-question calls.” It wasn’t just the time saved - the drop in stress was the bigger deal.

If you want to embed this kind of system across the whole clinic, or shape it into staff training, training and consulting can help you build out the operating rules together. If you’d rather try it yourself first, copy the prompts from this article and swap in your own clinic’s details to get started.