AI prompt to sort clinic tasks: automate, assist or keep with a person
Paste your list of clinic tasks into this prompt and it sorts each one into three groups: AUTOMATE (software does it), ASSIST (software drafts, and a person checks before anything is sent or saved) and KEEP WITH A PERSON. Each row gets a reason and a small first step. Diagnosis, triage (judging how urgent a case is), urgent symptoms and prescriptions never go in the first group, and the prompt stops if your list contains patient names.
Fill in and copy the promptCopy the prompt
Your values are marked in colour.
You are an operations analyst helping a physical therapy clinic decide which administrative tasks to automate. You are not giving legal, clinical or compliance advice. Classify every task in the list below. <tasks> Send appointment reminders Answer questions about opening hours Decide if a new symptom is urgent </tasks> <context> Systems in use: a booking system and a shared inbox Constraints: two front-desk staff, no IT team </context> Classes: - AUTOMATE: rule-based, low harm if wrong, and a mistake is easy to spot and undo. - ASSIST: software prepares or drafts, and a staff member reviews before anything is sent, saved or billed. - KEEP WITH A PERSON: clinical judgment, urgent or sensitive conversations, anything where a wrong action could harm a patient, move money, or break a legal duty. Return a table with these columns: Task | Class | Why (one sentence) | What the person still checks | First small step to try | Data touched (none, scheduling, contact, clinical). Rules: - If a task description is too vague to classify, write NEEDS MORE INFO and ask one specific question. Do not guess. - Never classify anything involving diagnosis, triage, urgent symptoms or prescriptions as AUTOMATE. - Do not estimate hours saved unless the user gave minutes and frequency. If they did, show the arithmetic. - End with the three best tasks to start with and one sentence on why each is safe to start. - Do not ask for or use patient names or identifiers. If the input contains any, stop and tell the user to remove them first.
How to use it
- 1Write your tasks as a plain list, one per line. Add minutes and how often if you know them, so any time saved is worked out from your own numbers.
- 2Fill in the four fields and copy the prompt. Name the systems you use, such as your booking system, because the first small step depends on them.
- 3Paste it into a chat AI. Do not include any patient names.
- 4Read the table. Start with one of the three tasks it suggests, and check the reason yourself before you build anything.
- 5Give the tasks you chose to whoever builds your workflows. The "What the person still checks" column becomes the review step.
A good result looks like
A clean table. Vague tasks are flagged instead of guessed. Triage and prescription tasks are never in the automate group, and any time savings come with the arithmetic shown. For the three example tasks I would expect "decide if a new symptom is urgent" to land in KEEP WITH A PERSON, and the reminders and opening-hours answers to land in AUTOMATE or ASSIST depending on where the answers come from. That is my reading of the rules, not a recorded result.
Stay safe
- No patient data. This sorts tasks and is not compliance advice.
- Whether a workflow may handle patient information is a decision for your compliance adviser.
If it goes wrong
4 known problems, each with a fix
- 01
Problem: It invents time savings when you gave no minutes.
Fix: Tell it "only use numbers I gave you".
- 02
Problem: It marks "answer patient questions" as automate with no source of answers.
Fix: Ask which document the answers come from.
- 03
Problem: It puts almost everything in assist to stay safe.
Fix: Ask it to justify each assist against automate.
- 04
Problem: It marks a task KEEP WITH A PERSON with no reason beyond "it is sensitive".
Fix: Ask which exact step is risky, and whether the rest of the task could be ASSIST.
Has it been tested?
Not yet. This one was written for this page and I have not run it on a specific model yet. Results will vary by tool, so run it once, check the result, and use the fix table above as your checklist. When I test it, the model, the date and the result will be listed here.
Questions
Is this safe to use at a clinic?
It works on task descriptions, not on people, so no patient data is needed. Keep it that way. Treat the result as a starting point for a conversation with your team and your compliance adviser.
Can AI decide which patients are urgent?
This prompt never puts triage or urgent symptoms in AUTOMATE. Judging urgency is clinical work for a qualified person. Software can at most pass a message to that person quickly. This is not clinical or compliance advice.
What do I do after I have the sorted list?
Pick one AUTOMATE task, decide what counts as a wrong result and who gets alerted, and build it with a person approving anything risky. I build AI integrations, bots and workflow automations, and you can describe your task list on the estimate page to get a free estimate.
Build this for you
Want this built into your own workflow?
Tell me what you do by hand today. I will say what is safe to automate, what needs a person, and roughly what it would take. I reply within 24 hours.