Orthodontic Referral Form Template and an AI Completeness Check

An orthodontic referral form should hold the patient's identifiers, the referrer's details, the reason for referral, relevant dental and medical history, radiograph details, consent and contact preferences. AI can check each incoming referral against that list and flag what is missing, but it should report gaps only. Deciding priority stays with a clinician.

That split is the whole design. A referral form template is a list of fields. A completeness check is a rule for each field: required, conditional or optional, with a plain-English reason. Once the rules are written down, an AI tool can read an email, a PDF or a form submission and tell the coordinator in seconds what to chase. The sections below give the template, the rules and examples of each.

Follow me on Instagram@sagnikteaches

The template, field by field

Most pages that rank for this search offer a downloadable blank. A blank form is a start, but it does not tell the person completing it what "complete" means. I would write the form with a short instruction under each heading, because that is also what the AI check will read.

Connect on LinkedInSagnik Bhattacharya

1. Patient details

Full name, date of birth, address, a phone number and an email address for the patient or the parent. For a child, add the name of the person with parental responsibility and their relationship. The most common gap here is a second contact number, and the most common error is a date of birth in a different format from the one your software expects. The check should flag both: "no second contact" is a note, and "date of birth does not look like a valid date" is a stop.

Subscribe on YouTube@sagnikteaches

2. Referrer details

Name, practice, professional registration number if your system uses one, phone, email and the address replies should go to. Include a line asking the referrer to say how they want the outcome communicated. A referral that arrives with no return address is one you can accept but cannot answer, which is why this field matters more than it looks.

3. Reason for referral

A free-text box and a tick-list of common reasons: crowding, spacing, overjet, crossbite, missing or impacted teeth, hypodontia, jaw discrepancy, space maintenance, patient request. The box should ask for one or two sentences, not an essay. The check looks for an empty field or a reason too vague to act on, such as "orthodontic opinion", and asks the referrer what the concern is.

4. Dental and medical history

Current dental health, previous orthodontic treatment, extractions, trauma, habits such as thumb-sucking, and relevant medical conditions, allergies and medicines. Keep this section as a structured list so the person completing it cannot skip a row. The check treats "allergies" as required even when the answer is "none known", because an empty field and "none" mean different things.

5. Radiographs and images

Whether radiographs were taken, which type, the date and how they are being sent. Panoramic radiographs are common in referrals, but whether one is expected, and whether other views or photographs are required, depends on your practice protocol and the guidance that applies to you. Put your own requirement in the form. The check compares what the referrer says they sent with what actually arrived.

6. Consent and communication

That the patient, or a parent where relevant, agrees to the referral and to sharing information with your practice. Add contact preferences and any language or access needs. The form should state who obtained consent and when. The check looks for an empty box and flags an unticked one, but it never infers consent from the rest of the text.

7. Urgency flag, completed by the referrer

If your form has a "routine or urgent" box, keep it as the referrer's own statement and mark it as such. The AI reads it as data and passes it on. It does not agree or disagree. For example, if the box is ticked "urgent" and the reason field is blank, the output is "urgent box ticked, reason missing, route to the coordinator now", not a judgement on the urgency itself.

Turning the template into rules an AI can apply

A model follows a clear rule much better than a vague instruction like "check this referral is complete". I would write a rule table with three levels, because they decide what happens next.

Your levels will differ from mine. The point is that each field has a level written down before any AI is involved, and the level is agreed by the clinician who owns the referral process. If that agreement does not exist yet, write it first. It is the useful part of the whole project.

The prompt that runs the check

Here is a prompt built on the rule table. It asks for a structured result, which makes it easy to paste into a tracking sheet. It tells the model what not to do, because that is where the risk lies.

You are an administrative assistant checking an incoming orthodontic
referral for COMPLETENESS ONLY. You are not a clinician. Do not assess
urgency, suitability, diagnosis or treatment. Do not infer anything not
written in the referral.

For each field below, report PRESENT, MISSING or UNCLEAR, and quote the
exact words you relied on. If a field is not in the text, say MISSING.

STOP fields: patient full name; date of birth; patient or parent phone;
referrer name; referrer return contact; reason for referral
CHASE fields: radiograph type and date; allergies (an explicit "none"
counts as present); relevant medical history; consent statement
NOTE fields: second contact number; access or language needs;
previous orthodontic treatment

Then give: (1) a one-line status: STOP, CHASE or COMPLETE; (2) a list
of what to request from the referrer, in plain words; (3) any
inconsistencies, such as a date of birth that gives an implausible age,
or a radiograph mentioned but not attached.

Referral text:
[paste referral here, with identifiers removed if your setup requires]

Notice the instruction "quote the exact words you relied on". It makes the model show its working, so a coordinator can see at a glance whether a "PRESENT" is real. Models can say a field is present when they have merged two lines, and a quote exposes that immediately.

What the output looks like, and what you fix

Here is a trimmed illustrative result for a referral that arrived as an email with one attachment.

STATUS: CHASE

STOP fields: all PRESENT
 - Reason: "crowding in the lower arch, parent concerned"

CHASE fields:
 - Radiograph: UNCLEAR. Email says "X-ray attached", but no type or
   date given.
 - Allergies: MISSING.
 - Medical history: PRESENT ("asthma, uses inhaler").
 - Consent: MISSING. No statement that the parent agrees to referral.

NOTE fields: second contact MISSING; access needs MISSING.

REQUEST FROM REFERRER:
 1. Please confirm the type and date of the radiograph.
 2. Please confirm any allergies, including "none known".
 3. Please confirm the parent has agreed to this referral.

INCONSISTENCIES: none found.

What I would change in the prompt after seeing this: the model flagged "second contact" as missing even though it is a NOTE field, so I would add "do not list NOTE fields in the request to the referrer". That keeps the chase email short. Small edits like this are how the prompt improves. Keep a version number on it so the team knows which one produced which result.

The radiograph and consent checklist

These two items cause most delays, so they get their own checklist. Write yours to match your protocol, because requirements differ between practices and sometimes between types of referral.

1. Radiograph type named. Does the form say which view or views were taken?
2. Date given. A recent image matters more than an old one, and your protocol will say what counts as recent.
3. Transfer method stated. Attached to the email, uploaded to a portal, on disc, or to be taken here.
4. Image received and readable. A person, not the AI, opens the file and checks it is the right patient and quality.
5. Consent to refer recorded. By whom and on what date.
6. Consent to share information recorded. Separate from consent to treatment, which comes later.
7. Guardian named for a child. And their relationship to the patient.
8. Preferred contact method noted. Particularly important for teenagers and for parents who work shifts.

The AI can verify the first three and the last four by reading the text. It cannot verify item four, because it cannot confirm the right patient is in the image. That is why the checklist keeps a human step, and why the output wording says "mentioned" and never "received".

Referral guidance differs between places

Referral guidance is published by national bodies, professional organisations and sometimes by the payer or health system your practice works with. The details differ: which radiographs are expected, what justifies a referral, whether a scoring index is used to decide eligibility for funded treatment and what the form must record. Do not copy another region's list into your rules. Find the guidance that applies to your practice, read the current version, and write your STOP, CHASE and NOTE fields from it.

Two systems show how far apart the rules can sit. In the UK, the NHS says a dentist who thinks a child needs braces refers them to an orthodontist, and the British Orthodontic Society's guidance for dentists says most orthodontists in primary care can treat only patients under 18 on the NHS, with eligibility depending on the IOTN grade (a five-grade index of treatment need running from "no need" to "severe need"). The same guidance asks referrers to give the IOTN grade if they know it, or else describe the problem (an overjet over 8mm, say, or severe crowding at the front), to include recent relevant radiographs, to mention previous orthodontic treatment and any trauma to the front teeth, and to explain any urgency in the letter. It also asks them to make sure the patient knows they are being referred and has good oral hygiene with no active decay, unless the referral is only for advice on extractions, and not to refer a patient who does not want treatment or would not wear a brace. In the US, the American Association of Orthodontists recommends a first check-up with an orthodontist by age 7 and tells parents they do not need to wait for their dentist to refer, so a practice there may see self-referred families alongside dentist referrals.

Turned into rules, the UK list adds three lines to the template: an IOTN line (CHASE, where "grade not known, problem described" counts as present), an oral-health line (CHASE) and a "patient aware of referral" tick beside the consent box. A practice that also takes self-referred families needs the form to work without a referrer, so the referrer fields become STOP only when a dentist is referring. Both changes are completeness rules, not clinical ones: the check reports that an IOTN grade is missing, never what the grade should be.

Where the same practice receives referrals under two different sets of rules, for example funded and private patients, keep two rule tables and add a field at the top of the form saying which one applies. The AI then loads the matching table. A practice that sees both can keep a single form and use a switch in the prompt, which I prefer to two separate forms because the referrer only needs to learn one.

Three practices, three setups

A single-site practice with email referrals

Referrals arrive as free-text emails with attachments. The coordinator pastes the email body into an approved business AI tool, runs the check prompt and copies the "request from referrer" lines into a reply. Setup takes an afternoon: write the rule table, test the prompt on ten past referrals with identifiers removed and compare the AI's result with what the coordinator would have said. If the two disagree more than once in ten, the rules need tightening.

A multi-site group with a web form

The form is structured, so the first layer of checking happens in the form itself: required fields cannot be blank. The AI check adds what a form cannot: whether the reason is specific enough, whether the radiograph description matches the upload, and whether the dates are plausible. Because the data is already structured, the prompt can be shorter and the output can go straight into a tracking sheet.

A practice receiving scanned letters

Here the weak point is the scan. If the text is not readable, the model will either miss a field or invent a plausible value. Put a rule in the prompt: "if the text is garbled or incomplete, report UNCLEAR, do not guess". Have a person read the scan itself for the STOP fields. This is the setup where I would be slowest to trust the AI, and I would run it alongside manual checking for a month before relying on it.

Weak and strong entries, field by field

The fastest way to teach referrers (and the AI) what "complete" means is to show both. Here is a pair for each of the seven sections, all illustrative. Put them on a guidance sheet that travels with the form.

Notice that none of the weak entries is wrong, exactly. They are things a busy referrer types at the end of a long clinic. The strong entries cost the referrer about thirty extra seconds, which is why a guidance sheet with examples reduces chasing far more than a stricter form does.

A worked month, with numbers

Say a practice receives 60 referrals in a month (an illustration, not a benchmark). Before the check, the coordinator reads each one and chases by phone, spending about 8 minutes on each that has a gap and 3 minutes on each that does not. Suppose 24 have gaps. That is 24 times 8 plus 36 times 3, which is 192 plus 108, so 300 minutes, or five hours.

With the AI check, reading and listing the gaps takes roughly 2 minutes per referral, including pasting and reading the quotes: 120 minutes. The chase emails still need sending, say 3 minutes each for the 24 with gaps: 72 minutes. Total 192 minutes, about 3 hours 12 minutes, a saving of around 108 minutes a month. That is not dramatic. The bigger gain is consistency: every gap on the list is found every time, and the tracking sheet tells you which referrers need a nudge. Count your own minutes for two weeks before and after, and trust those numbers, not mine.

A chase email the AI can draft for you

The check's "request from referrer" lines become a short reply. A second prompt turns the list into polite wording, and the coordinator edits it before sending.

Write a short, courteous email from our orthodontic practice to a
referring dentist. Thank them for the referral. Ask for ONLY the items
listed below, as a numbered list. Do not mention urgency or give any
clinical comment. Sign off as "[your name], Referrals Coordinator".

Items to request:
1. Type and date of the radiograph
2. Confirmation of any allergies, including "none known"
3. Confirmation that the parent agreed to this referral

A typical result is five or six lines, ready to send after a read-through. What I would fix: models often add "we look forward to seeing the patient shortly", which hints at an appointment you have not offered. Add "do not promise or imply an appointment date" to the prompt.

An inconsistency the check catches that a blank form never would

Consider a referral that gives a date of birth making the patient three years old, while the reason mentions "lower second molars not erupted". A hurried reader may miss it; a model told to look for implausible combinations will list it under inconsistencies. Most of the time the cause is a typing slip in the year, and the fix is a one-line query to the referrer. The same pattern catches a radiograph "dated" after the referral, a referrer phone number with too few digits and a parent named as the patient. None of these is clinical. They are the typing slips that otherwise surface at the first appointment, when the front desk finds the contact number does not work.

Write three or four of these consistency rules into the prompt, and no more. Each extra rule is another chance for a false alarm, and a coordinator who sees too many false alarms soon stops reading the list.

What to tell referrers about the check

Referrers do not need a lecture on AI, but they are entitled to know how their referral is handled. A single line on the form or the acknowledgement email is enough: that incoming referrals are checked for missing details, that a member of staff reviews every result and that nothing is decided automatically. Say which tool category you use if asked, and point to your privacy notice for the rest. A referrer who understands that a request for a missing radiograph date is routine is far less likely to read it as a criticism.

Tracking sheet: the part that makes it a process

A check that nobody records is a check that nobody learns from. Add a simple sheet with these columns: date received, referral reference, status from the check (STOP, CHASE, COMPLETE), what was requested, date requested, date resolved and who handled it. Do not put patient names in the sheet if a reference number will do.

After a month the sheet answers practical questions. Which referrers send incomplete forms most often? Which field is missing most? How long does a chase take to resolve? If one referrer's radiographs are always undated, a friendly call or a tweak to your published form can fix the cause instead of chasing the symptom. A short guidance page for referrers, with the weak and strong entries above, can sit on your website and be linked from every request email.

Realistic ways this goes wrong

• The model fills a gap. It writes "allergies: none" because the form said nothing. Fix: the prompt says MISSING when absent, and the quote requirement exposes any invention.
• Staff treat COMPLETE as "safe". The status only means the fields are present. Rename the status "fields complete" and say so in the staff guide.
• Drift in the prompt. Someone adds a line asking the AI to "prioritise urgent cases". That turns an admin check into triage. Lock the prompt, version it and have the clinical lead approve any change.
• Identifiers in the wrong tool. A coordinator uses a personal free account. Fix: a written rule, an approved tool and a quick read of the confidentiality checklist.
• Silent failure. The tool is down, so referrals pass unchecked. Keep a manual fallback checklist, the same list as above printed on one page.

Testing the check before you rely on it

Build a small test set of ten referrals, written by you, not real ones, covering the cases you care about. Include one perfect referral, one with no radiograph details, one with no consent statement, one with an impossible date of birth, one with a conflicting urgency flag and one with a garbled paragraph. Run the prompt on each and record whether the output matched what you expected.

A fair target is that every STOP and CHASE gap in your test set is caught every time, across three runs of the same prompt. Models vary between runs, so three runs matter. If a gap is caught twice out of three, the rule is ambiguous and needs rewording. Do this again whenever you change the prompt or switch tool.

Where this fits in the wider front desk

A completeness check is one narrow job. Calls about consultations and urgent breakages come through a different path, described in orthodontic AI receptionist consults. Once the referral becomes a patient, reminders and reviews follow, and patient recall letters cover the wording for those. If your practice software supports it, AI write-back to practice management systems explains what to check before anything is saved to a record. The general pattern for choosing admin tasks suited to AI is in which dental front desk tasks AI can take over.

If you would rather map your own referral path with someone, that is what my AI implementation consultation covers, and you can book a call here.

A short staff guide to put beside the screen

Keep this to six lines and pin it up.

1. Open the referral. Remove identifiers if your setup needs it.
2. Run the check prompt, version shown at the top.
3. Read the quotes, not just the status.
4. Copy the request lines into a reply to the referrer. Edit for tone.
5. Log the result in the tracking sheet.
6. Pass the complete referral to the clinician who sets priority. Never change the referrer's urgency flag.

A new coordinator can follow those six steps on day one. The rule table is the training, the prompt is the tool and the sheet is the evidence that it works.

Reviewing and updating the rules

Review the rule table twice a year and whenever a guideline you rely on changes. Look at the tracking sheet first: if a field is missing in more than a third of referrals, either the form is unclear or the requirement is unrealistic for referrers. Change the form before blaming the referrers. If a field is never missing, ask whether it still needs a rule, or whether it has become a habit and can drop to NOTE. Record each change and its date, so you can explain later why a referral was accepted under one version of the rules and queried under another.

Questions about checking referrals with AI

Can AI decide whether a referral is urgent?

It should not. Urgency is a clinical judgement, and a model can miss context a clinician would catch. Use AI only to report what is missing or inconsistent, then route the referral to the person who decides priority under your own protocol.

Do referrals need to be anonymised before the AI sees them?

Prefer a tool covered by a data-processing agreement, and strip identifiers where you can. If you cannot, use an approved business tool, not a free chat account. Your data-protection lead should sign off the setup before real referrals go in.

What happens if the AI says a field is missing but it is there?

Treat that as a failed check and look at why: usually a scanned PDF the model could not read, or a field label it did not recognise. Fix the rules or the input, and keep a small log of misses so you can see whether accuracy improves.

Should we reject incomplete referrals automatically?

No. Send a polite request for the missing items from a person, and keep the referral on file. An automatic rejection can delay care and annoy the referrer. The check should create a to-do, not a decision.

Further reads

• Dental Practice AI Mistakes: Consent, Data, and Over-Automation — Spot consent and over-automation slips early.
• Explaining Dental Treatment Plans to Patients With AI — Explain the plan to the patient once the referral is accepted.
• Best AI for Dental Insurance Verification: Which Tools Write Back — See which checks can run before a first visit.
• Heidi Health for Dentists: Setup, Templates and Patient Consent — Review consent wording for any AI tool in the surgery.
• Orthodontic Payment Plan Template: Fee Scripts Drafted With AI — Draft the fee conversation that follows a first consultation.
• Best AI for Orthodontics: Following Up Free Consultations — Compare the tool types that chase free orthodontic consultations, from texts and drafts to AI phone agents, and track consult-to-start.
• Marketing for Orthodontic Practices: AI for Referrers and Parents — Add two audiences to your orthodontic marketing, referring dentists and parents, with AI drafts that pass a claims and consent check.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: British Orthodontic Society pages for dentists on making a referral, NHS eligibility and when to avoid referring; NHS page on orthodontics; American Association of Orthodontists child orthodontics page (all read 4 November 2026). Referral guidance varies, so check the current version that applies to your practice. No vendor pricing is quoted.