Seven AI Prompting Techniques That Actually Work for Healthcare L&D

At Pivto, we've been helping healthcare L&D teams integrate AI tools into their workflows. Here are some patterns we've seen work consistently.

In early 2025, prompt engineering dominated the learning and development conversation. Conference sessions, webinars, and professional communities buzzed with tips and templates for getting the most out of generative AI tools.

But then the models got better, and prompting wasn't such a hot topic for the average user.

The reality about prompting in 2026 is this – give the model better context, not better instructions.

The 2026 models are smart. Really smart. You don't need a lot of complex prompting techniques. You don't need to be combing through Reddit forums looking for prompts to impress your co-workers. You mostly just need to give them great information, examples, and have a couple of powerful techniques to maximise your results.

After experimenting across ChatGPT, Gemini, and Claude, here are seven techniques that matter most right now:

1. Show, Don't Tell

This is the single most powerful prompting technique, and it's been true since the origins of GPT-3.

Instead of writing a paragraph explaining what you want, just show an example.

The weak way:

Write a scenario-based assessment question for nurse competency training. Make it realistic and clinically relevant. Include plausible distractors. The correct answer shouldn't be obvious. Align it to Bloom's application level.

That's a lot of instructional design terminology that the model will interpret inconsistently.

The better way:

Here are three clinical scenario questions from our medication administration training that scored highest on learner engagement:  [paste example 1] [paste example 2] [paste example 3]  Write a new scenario question about high-alert medication verification in the same style.

This works for clinical competency assessments, patient safety case studies, SOPs for medical device training, regulatory compliance modules—anything where you already have examples that perform well.

2. Feed Real Data, Get Real Insights

The prompt isn't where the magic happens. The data is.

If you ask Claude to "write a module on hand hygiene compliance," you'll get generic content. If you feed it your actual infection control audit findings, near-miss reports, or specific unit-level compliance data, you'll get content built for your organization's real challenges.

Spend more time gathering meaningful data—incident reports, regulatory audit findings, patient complaints, staff interview themes, quality metrics—and less time perfecting your prompt syntax.

3. Ask for Reasoning, Not Ratings

Research suggests models default to safe, middle-ground responses when asked to rate things numerically. You'll get a lot of 3s out of 5. Not particularly useful.

Open-ended questions yield richer, more actionable insights. Instead of "Rate this learning objective 1-5 for clarity," ask the model to respond from a specific learner's perspective with their honest reaction.

4. Build in Critique Cycles

First drafts from AI are serviceable. Second drafts are better. Third drafts, after structured self-review, can be genuinely strong.

The question "what's missing?" consistently surfaces ideas you hadn't considered. For healthcare content: "What would a risk manager flag?" or "What would a skeptical clinician push back on?"

5. Store Your Context in Memory

If you're using Claude, its memory feature allows you to build persistent context across conversations. ChatGPT and Gemini have similar capabilities—look for "memory" or "custom instructions" in settings.

Consider storing: regulatory frameworks, HIPAA considerations, accreditation standards, learner profiles, approval workflows, and health literacy requirements. No more re-explaining FDA requirements every time you start a new project.

6. Design for the Resistant Learner

Healthcare workers are drowning in required training. Your content competes with patient care demands and a dozen other mandatory modules.

Ask the model to anticipate resistance. Have it review from the perspective of an experienced clinician who's seen this training every year. What would make them tune out? What would make them think this might actually be worth their attention?

7. Validate Clinical Accuracy

Healthcare content carries higher stakes. A poorly worded medication dosing example can create real risk.

Build clinical review into your AI workflow. Ask for review against current guidelines, statements that could be misinterpreted, appropriate scope, and anything a plaintiff's attorney might cite after an adverse event.

This doesn't replace SME review, but it surfaces issues before that review—making the process faster and your drafts stronger.

What's working for you?

I'd love to hear which techniques resonate with your work, or if you've discovered approaches I haven't covered. Reply to this email—I read every response.

Until next week,

Kelley

Kelley Robson  ·  CEO, Pivto Better Learning

‍