Course resource

Healthcare Admin Prompt Library

Prompts for the administrative side of clinical and wellness work. Administrative only.

The line

No clinical use. None.

AI must not diagnose, suggest a diagnosis, recommend or adjust treatment, interpret results, advise on medication or dosage, or triage acuity. Models produce confident, plausible, wrong clinical content, and the consequences are not recoverable.

Everything in this document is administration: scheduling, correspondence, documentation, explaining things you have already decided.

Patient data must never enter a general AI tool. Patient information is among the most protected categories in every jurisdiction. Use only an approved system covered by the relevant agreements, or fully de-identify before use. If you are unsure whether your tool qualifies, the answer is that it does not.

Explaining what you have decided

The best use in this field: turning something you have already determined into language a patient understands.

Rewrite this for a patient with no medical background:
[PASTE THE EXPLANATION YOU HAVE ALREADY DECIDED TO GIVE]

Rules:
- Plain language. Explain any term you must keep, on first use.
- Do not add any information I have not included
- Do not soften or alter the clinical content
- Do not add reassurance I did not give
- Reading age around 12
- End with what they should do next and who to contact

This is my explanation being rephrased. Do not introduce new content.

That final constraint is the whole safety mechanism. The model is a translator here, not a source.

Patient-facing information

FAQ from existing material

Here is our approved information about [PROCEDURE / SERVICE]: [PASTE]

Turn it into an FAQ of 8 questions patients actually ask.
Use only the content above. If a likely question is not answered
in the source, list it separately as a gap for us to address —
do not answer it.

The gap list is genuinely useful: it tells you what your patient information is missing.

Appointment and preparation letters

Draft an appointment letter.

Appointment: [TYPE, DATE, TIME, LOCATION]
Preparation required: [FROM YOUR PROTOCOL]
Bring: [LIST]
Duration: [TIME]
Contact to reschedule: [DETAILS]
Accessibility: [ANY INFORMATION]

Plain language. Preparation instructions as a numbered list.
Do not add any preparation step not listed above.

Documentation and correspondence

Letter drafting from your notes

Here are my clinical notes for a letter to [RECIPIENT]:
[YOUR NOTES — de-identified]

Draft the letter in standard format.
Use only what is in my notes. Do not infer, expand, or add
clinical reasoning I have not written.
Mark anything ambiguous in my notes as [CLARIFY] rather than resolving it.

Every such letter is reviewed and signed by the clinician. The draft is a time saving on typing, not on thinking.

Policy and protocol questions

Here is our protocol document: [PASTE]
Question: [ADMINISTRATIVE QUESTION]

Answer only from this document, quoting the section.
If the protocol does not cover it, say so. Do not answer from
general knowledge.

Operations

Scheduling communication

Draft messages for: appointment confirmation, 48-hour reminder,
cancellation acknowledgement, and a rescheduling offer.

Practice: [NAME]
Contact: [DETAILS]
Cancellation policy: [POLICY]

Short, warm, clear. Include how to cancel or reschedule in each.
No clinical content.

Summarising non-clinical feedback

Here are [N] de-identified patient feedback comments about our
administrative service: [PASTE]

Group by theme. Count each. Quote one representative example per theme.
Identify the three most actionable operational issues.

Ignore any clinical content — flag it for clinical review separately.

Compliance checklist

What to do when the line is unclear

If you are unsure whether something is administrative or clinical, treat it as clinical and do not use AI.

The asymmetry is the point: the cost of not using AI for a borderline administrative task is a few minutes. The cost of using it for a borderline clinical one is not measured in minutes.

The useful summary

AI belongs in this field where it rephrases what a qualified person has already decided, drafts routine correspondence, and organises operational information. It does not belong anywhere near the decision itself, and the boundary is worth being conservative about.

Back to dashboard