GP Practice DNA Policy: Wording, Reminders and Letters With AI

A GP practice DNA policy should define "did not attend", set a rolling-period threshold for a warning letter, list who is protected from automated or standard warnings, describe the reminder routine, and say who decides. AI can draft the policy wording, reminder texts and letters, but a clinician reviews each case before any letter is sent.

Most sample policies stop at the threshold: three misses and you are warned, more and you may be removed. That is the easy part. The parts that determine whether the policy is fair and whether it actually reduces missed appointments are the exceptions, the reminders and the human check. Settle those first; the drafting, where AI saves time without touching the decisions, is the easy finish.

Follow me on Instagram@sagnikteaches

Why missed appointments are worth a policy at all

Every missed slot is a clinician's time that another patient could have used. A practice that books 300 appointments a day and loses 2 to 5 per cent to non-attendance loses six to fifteen slots daily, or one to two and a half hours of clinician time at ten minutes a slot (illustrative figures). The cost is not only time but also waiting for the patients who could not get in. A policy gives staff a consistent response, tells patients what is expected, and protects the practice from accusations of unfairness because everyone is treated by the same written rule.

Connect on LinkedInSagnik Bhattacharya

A policy alone rarely changes behaviour much, though. Reminders, easy cancellation and a friendly tone do more than threats. So treat the policy as the backstop and spend most of your effort on the routine that stops the miss in the first place.

Subscribe on YouTube@sagnikteaches

Define the terms before anything else

Ambiguity causes arguments. Open the policy with short definitions in plain English, such as these:

• Did not attend (DNA): the patient had a booked appointment, did not arrive and did not cancel or contact the practice before the appointment time.
• Late cancellation: cancelled within a stated window before the appointment, for example less than two hours. Decide whether you count these; a practice may record them without ever warning on them.
• Late arrival: arrived after a stated number of minutes past the booked time and could not be seen. Say what happens.
• Practice-initiated change: a clinician is absent, the appointment is moved, or the reminder was not sent. These are never a patient's DNA.
• Rolling period: the stretch of time in which misses are counted, such as twelve months back from the latest miss.

Mention which appointment types count. A missed telephone call from a clinician, a missed blood test and a missed routine review might each be treated differently, and patients deserve to know how.

Thresholds: choose a number you can defend

The threshold is the point at which a letter is considered. The exact figure is for your partners to set with your contract, local guidance and your patient participation group, and I will not give one as a standard. A common structure has three steps, which you can adapt:

Notice that no stage in this table is automatic removal. Removing a patient from the list is a serious step with its own rules that depend on where you work and what your contract says. A DNA policy should not hide it behind a counter. If removal is possible in your setting, make it a separate decision made by named people and recorded, never by a script or a tool.

Test the threshold on your own data before adopting it. Export a year of DNA records with patient identifiers removed, count how many patients would have received a warning at two, three and four misses, and see how many of those are patients you know to be vulnerable. If the numbers show that a third of those who would be warned have complex needs, the policy needs more clinical review, not a different count.

Who is protected, and how you check before a letter goes

This is the section sample policies treat lightly and the one that matters most. A missed appointment can be the first sign that something is wrong, whether that is illness, caring duties, fear, confusion or abuse at home. Your policy should list categories of patients for whom a standard letter is not the first response, and say that a clinician looks at the record before any letter goes out.

Typical categories to consider are children and young people, adults at risk, people with mental health problems or learning disabilities, people with dementia, people nearing the end of life, those with serious or unstable conditions, carers, people with limited language or literacy needs, people with no fixed address and anyone known to be in crisis or experiencing domestic abuse. Your own safeguarding lead should shape the list.

The mechanism is simple: a pre-letter review. When the count reaches the threshold, the case goes to a named person, usually the practice manager with a clinician, who reads the notes for flags, thinks about why the person did not attend, and chooses from three outcomes: send the standard letter, send a gentler welfare-style contact, or make a phone call. Record the outcome and the reason. The review is a human step by design; it is not a task for AI, and no patient record should be pasted into a tool for it.

Reminders that stop the miss

Reminders do more than any penalty. A sensible routine has layers:

1. At booking: a confirmation message with the date, time, clinician type and how to cancel.
2. Two to three days before: a text reminder with a reply option to confirm or cancel.
3. On the morning: a short final nudge for appointment types with high miss rates.
4. After a miss: a same-day, friendly message that asks if they need to rebook.

Make cancelling frictionless. If a patient can reply "C" to a text, use a link or call a clear number, more will cancel instead of vanishing. Check your phone lines at peak times; a cancellation route that is permanently engaged is worse than none. The mechanics and cost of text reminders are covered in what automated reminder SMS really costs, and the consent side in reminder wording and consent.

Write reminder texts with a few rules: no clinical detail on the screen (a lock screen can be read by anyone), a clear sender name, one action, and a tone that assumes good intentions. Here are four drafts you might generate with AI and then adapt. They are illustrative samples.

Confirmation:  Hello [first name], your appointment at [practice] is booked for [day] at [time]. If you can't come, please cancel at [link] or call [number] so someone else can use the slot.

Reminder:  Hello [first name], a reminder of your appointment at [practice] on [day] at [time]. Reply C to cancel or call [number] if you need to change it.

Morning:  Hello [first name], we're expecting you at [time] today at [practice]. If something has come up, please let us know so we can offer your slot to someone else.

After a miss:  Hello [first name], we missed you at [practice] today. We hope all is well. If you still need an appointment, please call [number] or book at [link].

What you would check in AI-drafted texts: that none mentions the reason for the visit, that each fits within the length you pay for (long messages are billed as several), that the cancellation route really works, and that the tone does not carry any hint of blame. Draft with invented details; the real merge fields are filled by your messaging system, not by an AI tool.

The letters, with the AI prompt and a sample

Three letters cover most needs: the first-miss note, the warning letter and a welfare contact for those flagged as potentially vulnerable. Write the template once, with merge fields, and keep it under version control. An AI tool is good for the first draft and for rewording to a reading age of about 11 to 12, which helps patients who find official letters hard.

Write a letter from a GP practice to a patient who has missed several appointments in
twelve months. Tone: kind, calm, plain English, reading age about 11. No blame and no threats.
Length: 200 words or fewer. Use these merge fields only: [first name], [dates of missed
appointments], [practice name], [phone number], [practice manager name].
Include: the dates; why missed appointments matter (another patient could have used the slot);
how to cancel in future; an invitation to tell us if something is making it hard to attend;
what happens next, in one sentence, which I will fill in as [next step].
Do not mention any medical condition. Do not state a rule or number that I have not given you.

A reasonable result from a chat tool reads like this (illustrative):

Dear [first name],

Our records show you were unable to attend appointments on [dates of missed appointments]. We hope everything is all right.

When an appointment is missed, another patient may have to wait longer to be seen. If you can't come, a quick text or call to [phone number] lets us offer the slot to someone else.

If something is making it hard for you to attend, please tell us. We can often help, for example with the time of day, a phone appointment or reminders.

[next step]

Yours sincerely,
[practice manager name]
[practice name]

As the editor, you would check several things. The first sentence says "unable to attend", which is gentle but may not be true if the patient chose not to come, so make it factual: "did not attend". The phrase "we can often help" promises something, so confirm that is true. The letter must not run foul of any rule about what you may say, so have the partners approve the final template once and then reuse it. Never leave the line "[next step]" for staff to improvise, because a letter that threatens removal when the policy does not say so is worse than none.

The welfare version is different in kind. It starts with concern rather than a count, does not mention thresholds, and invites the patient to call a named person. Draft this one with extra care and have a clinician read it. Sometimes the right response is a phone call, not a letter, and the AI draft is only a script for the call.

One clause, before and after

Policies written by committee tend to sound like contracts, and patients stop reading at the second line. Compare two versions of the same clause. Before: "Patients who fail to attend scheduled appointments without providing adequate notice on more than the permitted number of occasions may be subject to action in accordance with the practice's procedures." After: "If you miss an appointment without letting us know, we will send you a friendly note. If it keeps happening, we will write to you and ask whether something is making it hard to come. We will always look at your circumstances first."

The second version is shorter, says what actually happens and signals that the practice will think before acting. AI is good at this kind of rewrite when you ask it for a plain-English version at a stated reading age and then tell it to keep every commitment unchanged. Compare the two side by side, because a rewrite can quietly soften or strengthen a commitment, and a promise you did not mean to make becomes the one a patient quotes back at you.

Handling the phone call after a letter

Some patients ring after receiving a note, often upset or embarrassed. Equip reception with a short, kind script rather than leaving each person to improvise. A good script has four moves: thank them for calling, listen to the reason without interrupting, say what the practice can do (rebook, offer a phone slot, adjust reminders), and record the reason in the record so the next review sees it. AI can draft the script and a set of likely replies, for example "I did not get a text", "I could not get through to cancel" and "I was in hospital". Use invented scenarios only, and have a clinician check that the guidance on what to say never strays into clinical advice.

Two reasons recur often enough to design for. A patient who says they never received a reminder prompts a check of whether the phone number on record is right, which is cheap to fix and worth doing at every contact. A patient who says they could not get through to cancel is pointing at a failure of yours, not theirs. Both belong in the monthly review of causes.

Language, literacy and access

A letter in dense English reaches only those who can read it. Offer the policy and key letters in the languages your patients use, with translations checked by a fluent speaker who understands the practice context, because a machine translation of a warning letter can read as harsher than the original. Consider an easy-read version, and make sure texts work for people using screen readers. Where a patient has an interpreter booked, a missed appointment may be the interpreter's failure to arrive; check before counting it.

Family situations need thought too. If a parent misses a child's appointment, the letter goes to the parent, but the child's welfare is the concern, so those cases take the welfare route first. If the same address shows several DNAs across family members within a short window, treat that as a possible sign of a household problem and raise it with the safeguarding lead rather than sending three separate letters.

A second walk-through: one patient's path

Here is an invented patient path to show how the pieces fit. A patient with a long-term condition, booked for a routine review on a Tuesday morning, gets the confirmation text at booking, a reminder on the Sunday and a morning nudge. They do not attend and do not cancel. At lunchtime the system sends the after-miss text, and the receptionist notes the DNA in the record.

Three weeks later, the same patient misses a phone appointment. The practice has set its threshold at three misses in a rolling twelve months, so this second miss gets the friendly reminder note rather than a warning. A third miss follows two months later and the count reaches the threshold. The case goes to the pre-letter review, where the practice manager and a clinician read the record and see that the patient recently lost a family member and has not been collecting repeat medicines. They decide not to send the standard letter. A clinician phones, the patient explains they have been overwhelmed, a longer appointment is offered and reminders are moved to a phone call the day before. Without the review step, a warning letter would have arrived in the same week as a bereavement.

Handling the data side as the policy matures

Keep a small record of each DNA decision: the date, which stage, who reviewed, the outcome and the reason. The record protects patients because it shows consistent treatment, and protects staff because it shows that the process was followed. Agree retention with the practice's records policy, give patients access on request in the usual way, and review the log quarterly for patterns that show bias, such as a particular group receiving more warnings than its share of misses would suggest.

Where patient data may and may not go

The policy and letters are templates and can be drafted freely with invented examples. The risk begins when real patient information enters an AI tool. Keep to three rules. Draft templates with merge fields and fictional examples only. Do not paste real names, dates of birth, national health identifiers, appointment histories or clinical notes into any chat tool unless it is an approved, contracted product for health data. And let the practice's own messaging or letter system do the merge from the clinical system, so the AI never sees the patient. For the confidentiality questions to put to a vendor, see patient data and AI: a confidentiality checklist.

Writing the policy itself with AI

The policy text is a good AI job because it is structured and not patient-specific. Give the tool your decisions and ask it to assemble the document, with a plain-English summary for the website and a staff version with procedures. The prompt should include your definitions, your thresholds, the list of protected categories, the reminder routine and the review roles, and should say "do not add rules I have not given you". Then compare the draft against your decisions line by line, because AI will helpfully add a clause that sounds sensible, such as an automatic removal after a number of misses, that you never agreed.

A realistic mistake to expect from this kind of drafting: ask a tool for a "complete" policy and it may add a sentence saying patients can be removed after three DNAs in six months, even though the partners decided on no automatic removal. A manager who ticks every clause against the decision list catches it in minutes. Without the list, a sentence like that goes onto the website and becomes the rule patients quote.

A year of data, as a worked example

Take a made-up practice with 8,000 registered patients and 60,000 booked appointments a year. Suppose 3 per cent are DNAs, which is 1,800 missed appointments. At an average of ten minutes of clinician time each, that is 300 hours, which is the equivalent of several weeks of one clinician's work. These figures are invented for illustration.

Now suppose the practice introduces a three-touch reminder routine and an easy cancel-by-text option, and the DNA rate falls from 3 per cent to 2.2 per cent. That is 480 fewer missed appointments, or about 80 hours returned. Even if only half of the fall is down to the reminders, that is a worthwhile return from a routine whose cost is a per-message fee (check your messaging provider's current rate). By contrast, a stricter warning threshold on its own would reach only the small number of repeat non-attenders, who are often among the patients with the greatest need.

Use your own figures. Count DNAs by type of appointment, time of day and day of the week; a pattern such as Monday mornings or nurse appointments with a long lead time points to where a reminder or a rebooking offer will help most.

Telling patients and measuring whether it works

Publish the policy on the website in short, plain language, add a line to the booking confirmation, and keep a printed copy at reception. Staff need a short script too: what to say when a patient rings after a warning, and when to refer the call to the manager. Review the policy yearly with the patient group, and keep a simple dashboard: DNA rate by month, number of first-miss notes sent, number of warnings, number of cases where the review chose a different route, and how many patients rebooked within a week of a post-miss message. If warnings rise while the rate stays flat, the policy is punishing without changing behaviour.

For the wider picture of fewer missed appointments, how a clinic can use AI to cut missed appointments covers the reminder side in more depth, and an AI receptionist for the morning rush shows how calls and cancellations connect. The AI hub has the rest of the series.

Questions practice managers ask about missed appointments

What does DNA mean in a practice policy?

DNA means did not attend: the patient had a booked appointment, did not come, and did not cancel or tell the practice beforehand. A good policy defines it precisely and separates it from late cancellations, late arrivals and appointments the practice itself changed, so that nobody is counted for something outside their control.

How many missed appointments should trigger a warning letter?

There is no universal number. One workable shape is a count within a rolling twelve months, such as two or three, but the right figure depends on your appointment volumes, your contract and the guidance that applies to you. Agree the threshold with the partners and patient group, and record it in the policy.

Can AI write the letters?

Yes, as drafts. AI is good at plain, kind wording and consistent structure. A person must check the facts, apply the exceptions before any letter goes out, and own the decision. Never let a tool decide who receives a warning, and never paste patient names or details into a tool that is not approved for health data.

Who should never get a standard warning letter?

Anyone for whom a missed appointment may signal a problem rather than carelessness: vulnerable adults, children, people with mental health conditions, learning disabilities, dementia, serious illness, carers, people in crisis and those who need interpreters. Build a clinical review step before any letter, so a clinician looks at the record first.

Further reads

• GP Surgery Phone System: Cloud Telephony With AI Call Handling — A phone system that supports cancellations.
• How Dental Practices Use AI for Recalls and Appointment Reminders — A recall and reminder routine from dentistry.
• Winning Back Lapsed Dental Patients With AI Messages — Messages to patients who have drifted away.
• 10 Admin Tasks a Small Clinic Can Hand to AI This Month — Other clinic admin jobs worth handing over.
• GP Surgery Website: Plain-English Pages and Wording With AI — Where the published policy should live on your site.
• AI Marketing Automation for Doctors: Recall, Reminder and Review Flows — Build recall, reminder and review-request flows in the practice software you already own, with consent, opt-outs and a worked month of numbers.
• Best AI for Paediatricians: Vaccine Recalls and School Forms — Build a vaccine recall list, letter templates and a school and camp form queue with AI, keeping parent data safe and symptom advice out.
• CQC and GP Practices: Build an Evidence Folder With AI — Organise a GP practice's CQC evidence by key question and quality statement, with an AI-built index, policy review log and mock-inspection summary.
• AI in General Practice UK: Telling Patients and Updating Privacy Notices — Draft patient notice, privacy-notice paragraph and poster wording for a GP practice using AI scribes or phone tools, with a summary of the NHS guidance.
• What Policies Should a GP Practice Have? Checklist and AI Drafts — A GP practice policy checklist grouped by risk, with a review calendar and tested prompts for drafting each policy with AI before a partner signs it off.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: general good practice for appointment policies and patient communication; the practice's own contract, regulator and professional guidance must be read directly before adopting thresholds.