How to Write an AI Prompt That Gets You Useful Output

Three things every AI prompt needs

Define your goal

Start by being specific about what you're trying to create and why. Is it a chart note? A follow-up email to a patient? An Instagram caption? The AI needs to know the purpose before it can get the format or tone right. It also helps to name the outcome you're after. Not just "write something about my services" but "write a short About Me paragraph that sounds approachable to new patients."

Specify the format

Tell the AI exactly what structure you want: bullet points or paragraphs, how long it should be, what sections to include. If you're writing a SOAP note, say so. If you want three sentences, say three sentences. Leaving format open usually means you'll get something that technically answers the brief but doesn't fit where you need to use it.

Add context and constraints

This is where most prompts fall short. Context means giving the AI the background it needs: your dictation, the patient scenario, your target audience, your tone. Constraints are the guardrails: things to include, things to avoid, rules around language or formatting.

A few examples of constraints that make a real difference in practice:

• "Use Canadian spelling throughout."
• "Do not include clinical recommendations I haven't mentioned."
• "Write in a warm but professional tone. This is going to a patient, not a colleague."
• "Keep the total length under 150 words."

One important note on privacy: when using a general AI tool like ChatGPT for anything related to patient communication, leave out names and identifying details. Use a placeholder like "Patient A" or a generic description instead.

Good AI prompt examples

Here's how those three pieces come together in four common scenarios. The structure (goal, format, constraints, context) stays the same each time. What changes is the specifics you feed in.

Drafting a chart note from a live AI Scribe recording

When your AI Scribe is recording your session as it happens, the context it's working from is the live conversation. Your prompt should specify how to pull the most relevant context from real-time speech.

You are an expert scribe that is responsible for summarizing the patient's appointment. Your task is to briefly summarize the session from a live recording. Write a SOAP note in point form, with three to five bullet points per section. Only document clinical information related to the session. If something is unclear or not mentioned, write "not documented" rather than guessing. Do not add clinical recommendations that were not stated aloud during the session. Here is my recap: [insert recording]

Drafting a chart note from a post-session AI Scribe dictation

When you're talking to your AI Scribe after the session to summarize what happened, the context is your recap, not a live recording. Your prompt should reflect that you're the only voice, and that you're summarizing from memory:

You are an expert scribe that is responsible for documenting a treatment session from a post-session dictation. Write a SOAP note in point form, with three to five bullet points per section. Use only the information I provide in my dictation. Do not expand on or interpret what I say. If something is unclear or not mentioned, write "not documented." Here is my session recap: [insert recording].

💡 If you’re just getting started with an AI Scribe for soap notes and not sure which option will work best for you, take a peek at our breakdown of ambient listening vs. dictation to learn the pros and cons of each.

Drafting a Google review response

One thing worth adding to this prompt is a line about your clinic's tone or personality. Review responses are public-facing and often the first impression a prospective patient gets, so the more the AI knows about how you communicate, the more authentic the output will feel.

You are the owner of a small allied clinic responding to a Google review. Write a two to three sentence response that is warm, understanding, and acknowledges the reviewer's feedback specifically. Our clinic values professionalism, so ensure the tone of the response is formal and concise. Do not mention any clinical details or the patient's condition. Here is the review: [paste review].

Drafting a reply to a patient email

Keep in mind that not every patient email needs the same tone. A quick address confirmation and a charge dispute are very different conversations, so it's worth adding a line to your prompt that describes the nature of the email and how you want the response to land.

You are the front desk coordinator at a small allied health clinic. Write a reply to the following patient email regarding a charge on their credit card. Keep the tone warm, clear, understanding, and professional. The response should be three to five sentences. Do not make any clinical recommendations or promises about billing outcomes. Sign off with my name: [your name]. Here is the email: [paste email].

Once you've got a version that works for each scenario, save it as a template so you're not rebuilding a new prompt from scratch each time.

How to apply the same formula across your whole practice

The same approach applies to a lot of the business and marketing work that piles up. A few other places a well-thought out AI prompt can make an impact:

• Your website's ‘About’ page: Give the AI a few bullet points about your background and approach, then ask it to turn them into a short, compelling paragraph written for prospective patients. Specify the tone ("approachable and professional, not overly clinical") and the length.
• Your brand voice: If your AI-written content keeps sounding generic, try this. Paste a few examples of your own writing, like a newsletter intro, a social post, or anything that just sounds like you. Then, ask the AI tool to identify what makes your voice distinctive. Save that description and include it in future prompts. The output will start to sound a lot more like you wrote it yourself.
• Social media marketing and content repurposing: Got a blog post that did well? Paste it in and ask the AI to turn it into three Instagram captions, a short email teaser, or a script for a short video. Specify the platform, the length, and the tone for each.

What to do when the AI output still isn't right

If the first attempt misses, adjust one thing at a time rather than rewriting everything. The most common fixes:

• Output is too long: add a word count limit to your constraints.
• Tone feels stiff: add "plain, conversational language" to the format section.
• It's adding things you didn't say: add "only include information I provided. Do not infer or assume."
• Format keeps drifting: paste a brief structural example directly into the prompt.

Prompting is a skill, and it does take a few rounds to dial in for each type of task. Most practitioners find that a short investment of time upfront, testing a few variations and saving what works, pays off quickly once the templates are set.

Frequently asked questions

Do I need a separate AI prompt for each type of note or task?

It helps to have one for each note type. A SOAP note for a chiro visit, a session note for a psychotherapist, and a progress note for an OT all have different structures and documentation conventions. Build a base prompt and adjust the role, format, and rules for each one rather than starting from scratch every time.

How do I stop the AI tool from adding things I didn't say?

Add this instruction to your constraints: "Only include information I provided. Do not infer, assume, or add clinical recommendations unless I stated them. If something is unclear, write 'not documented.'" This is especially important for clinical notes, where accuracy isn't optional.

Is it safe to use a general AI tool like ChatGPT for patient-related content?

For administrative tasks like Google review responses, website copy, and general emails, the answer is generally yes, with one caveat: never include patient names or identifying details. For clinical notes that involve session-specific content, a purpose-built AI Scribe tool with the right privacy protections in place is the better choice. For a more complete breakdown of what privacy compliance looks like for AI scribe tools, this article covers both HIPAA and PIPEDA requirements.

How long does it take to get an AI prompt working well?

A few sessions of testing is usually enough. The most reliable approach: adjust one variable at a time, save versions that work, and build a small library of prompts you can reuse. Most practitioners who put in that initial effort find they spend significantly less time editing AI output within the first week or two. If you’re looking for more advice to help you start using AI in your practice, this article in Front Desk Magazine breaks it down, step-by-step.

💡If you're already on Jane and want to put this into practice with AI Scribe, the prompt and template setup guide is a good place to start. Here's how to set it up.