How to talk to the AI: Prompting cheatsheet

Talking to an AI is a lot like giving instructions to a competent (but very literal) assistant.

If you say, “Get me three symptoms,” they might return anything. But if you say, “Find three symptoms my patient had in our session, summarize the key findings in three bullet points, and keep the tone professional,” you’ll get a result that’s much closer to what you want.

You don’t need technical knowledge to do this well. Clarity and specificity make all the difference.

Think of the AI as an intern; it gets the job done but needs specific instructions to give the best results. AI models are trained on large amounts of everyday language, so they respond best when you write to them the way you would speak to another person.

Prompt Engineering Best Practices

Below are a few simple tips to help you create effective prompts for Gemini models and ChatGPT models :

1. Be clear about what you need

Say precisely what you want the AI to do: summarize something, draft an email, rewrite text, or explain a concept. The more precise the request, the better the answer.

2. Provide the proper context

Add any details that matter: what you want the answer for, what your profession is, who the content is for, what the goal is, or what information should be highlighted. Context helps the system stay aligned with your intent.

3. Break complex tasks into smaller steps

If the request has several parts, split complex requests into multiple single tasks in a single message; this way, the model can follow instructions using sequential data. Example:

“1. Do X task

2. Based on X task in point 1, do Y task

3. Now do Z based on point 2.”

Referencing the previous steps will improve accuracy on the following steps for these kinds of tasks

4. Treat it like a conversation

You don’t have to get it perfect on the first try. Start with an initial request, check the answer, then refine or ask follow-up questions. Iteration usually leads to the best results.

In case you iterated multiple times in a single conversation, switch to a new conversation to have a clearer context for the AI so it does not mix up previous instructions

Templates to Understand AI Writing

In this section, we’ll be exploring how to use artificial intelligence in medicine in a way that delivers exactly what you’re using it for.

To use the platform as an AI search engine, try a prompt like this:

“I am researching [condition/topic] in [patient population]. Please provide:

A concise clinical overview (definition, epidemiology, pathophysiology).
Current guideline-based recommendations (include major societies and year).
First-line and second-line treatment options, including dosing ranges when appropriate.
Key clinical trials or high-quality evidence supporting recommendations.
Contraindications, major risks, and monitoring parameters.
Areas of controversy or evolving evidence.
Practical clinical pearls for real-world practice.

If evidence differs by age, comorbidities, or setting, please stratify accordingly.”

When using CompliantChatGPT to draft documents or for AI writing, use AI prompts like this one:

“I need help writing a [type of document: clinical note, prior authorization letter, appeal letter, academic abstract, discharge summary, patient education handout, research summary, policy memo, etc.].

Audience: [insurer, specialist colleague, patient (specify literacy level), journal, hospital administration, legal, etc.]

Tone: [formal, concise, persuasive, academic, patient-friendly, etc.]

Length: [brief / one page / 250 words / detailed]

Key clinical details to include:

• [Diagnosis]

• [Relevant history]

• [Labs/imaging]

• [Treatment provided or requested]

• [Clinical rationale]

• [Guideline support if applicable]

Please structure it clearly, use professional medical language, and ensure it is concise and persuasive where appropriate.”

Hope this helped you learn how to use ChatGPT effectively, as well as our Gemini models.