The Prompt Matters

The Setup

You're a pediatrician. A 12-year-old boy with familial hypercholesterolemia (FH) is in your office. His LDL is 190 mg/dL despite six months of dietary modifications. His father had a myocardial infarction at age 42. You're considering statin therapy and want to check the evidence.

You open a clinical AI tool to help inform your decision.

Prompt A: The Vague Approach

What's Wrong With This Prompt?

Let's break down why this prompt is likely to produce unhelpful—or potentially misleading—output.

Problem 1: No clinical context

The AI has no idea who's asking or why. Are you a:

• Pediatrician considering treatment for a specific patient?
• Medical student studying pharmacology?
• Parent who read something online?
• Researcher reviewing the literature?

Each of these would warrant a completely different response. Without context, the AI has to guess—and it will default to a generic, encyclopedia-style overview that may not address your actual need.

Problem 2: No specific question

"What about statins in kids?" could mean dozens of things:

• Are they safe?
• Are they effective?
• Which ones are approved?
• At what age can they be started?
• What are the indications?
• What are the side effects?
• How do you monitor?
• What do guidelines say?

The AI will try to cover everything superficially rather than anything deeply. You'll get breadth without the depth you need to make a clinical decision.

Problem 3: No patient specifics

Statins in a child with heterozygous FH is a very different question than statins in a child with obesity-related dyslipidemia. Without knowing your patient's diagnosis, labs, family history, and what's already been tried, the AI can't tailor its response to your situation.

Problem 4: Increased hallucination risk

Here's a critical point from Module 2: vague prompts increase hallucination risk. When a question is ambiguous, the AI has more "degrees of freedom" in generating a response. It's more likely to:

• Fill gaps with plausible-sounding but potentially inaccurate information
• Blend information from different contexts inappropriately
• Generate confident-sounding statements about things it's uncertain about

A specific question constrains the response space. A vague question lets the AI wander.

Problem 5: No way to verify relevance

Even if the AI produces accurate general information, how will you know if it applies to your patient? You'll have to do the mental work of filtering and applying—work the AI could have done if you'd given it the information to work with.

What You're Likely to Get

With a prompt like "What about statins in kids?", expect:

• A general overview of pediatric statin use
• Broad statements about safety and efficacy
• Possibly outdated or non-specific guideline references
• No guidance on YOUR patient's specific situation
• No clear decision support
• Information you'll need to significantly filter and supplement

You've essentially asked the AI to write a review article when what you needed was a focused consultation.

Prompt B: The Specific Approach

What Makes This Prompt Effective?

Element 1: Professional identity establishes context

"I'm a pediatrician" tells the AI:

• Use clinical language appropriate for a physician
• Assume baseline medical knowledge
• Frame the response as peer-to-peer clinical guidance
• Focus on practical decision-making, not patient education

Element 2: Specific patient details focus the response

• "12-year-old boy"—age and sex matter for pediatric prescribing
• "familial hypercholesterolemia"—this is the specific indication, not generic dyslipidemia
• "LDL is 190 mg/dL"—quantifies the severity
• "despite 6 months of dietary modifications"—establishes that first-line therapy has been tried
• "father had an MI at age 42"—provides family history that influences risk assessment

Each detail constrains the response toward relevance. The AI isn't guessing what scenario you're asking about—you've told it.

Element 3: Explicit question structure

The prompt doesn't just ask "what should I do?" It breaks down the specific information needed:

• LDL threshold for treatment initiation
• Which statins have pediatric safety data
• Monitoring recommendations

This structure guides the AI to organize its response around your actual decision points rather than generating a generic overview.

Element 4: Evidence framing

"What does current evidence say" signals that you want evidence-based guidance, not opinion. This primes the AI to reference guidelines, studies, and established recommendations rather than generating general statements.

What You're Likely to Get

With this specific prompt, expect:

• Targeted information about statin therapy specifically in pediatric FH
• Reference to relevant guidelines (AAP, AHA, NLA recommendations for FH)
• Specific LDL thresholds for treatment initiation in this population
• Information about which statins (likely atorvastatin, rosuvastatin, pravastatin) have pediatric indication or safety data
• Monitoring parameters (baseline and follow-up LFTs, CK if symptomatic, lipid panels)
• Contextually relevant information you can actually apply to your patient

The response should function more like a focused literature summary or specialist consultation than an encyclopedia entry.

Reduced Hallucination Risk

The specific prompt reduces hallucination risk in several ways:

Narrower scope = less room for error. When you ask about pediatric FH specifically, the AI draws on a more focused knowledge domain. When you ask vaguely about "statins in kids," it has to synthesize across many contexts, increasing the chance of inappropriate blending.

Checkable claims. Specific questions tend to produce specific, verifiable answers. "Atorvastatin is FDA-approved for pediatric FH starting at age 10" is verifiable. "Statins can be used in children in some situations" is technically true but not useful or checkable.

Explicit uncertainty prompting. By asking "what does the evidence say," you're implicitly asking the AI to anchor its response in sources—making it more likely to acknowledge when evidence is limited rather than generating confident-sounding filler.

The Comparison at a Glance

The Takeaway

The same clinical question can produce dramatically different AI outputs depending on how you frame it. This isn't about knowing magic words or secret techniques. It's about the same principle that governs good clinical communication:

The habit to build: before submitting any prompt to a clinical AI tool, pause and ask yourself:

• Have I told it who I am and what context I'm working in?
• Have I provided the relevant patient details?
• Have I specified exactly what I need to know?
• Is my question focused enough that I'll be able to verify the answer?

Those few seconds of thought consistently transform vague, marginally useful AI responses into focused, clinically applicable guidance.

Connection to Module 2: Why This Matters for Safety

Remember from Module 2 that AI hallucinations aren't random—they're more likely under certain conditions. Vague prompts create those conditions:

• Ambiguity forces the AI to guess your intent, and guesses can be wrong
• Broad scope requires synthesis across contexts, increasing error risk
• Lack of constraints gives the AI more freedom to generate plausible-sounding but inaccurate content

Specific prompts are a safety practice, not just an efficiency practice. They reduce the attack surface for hallucination by constraining what the AI needs to produce.

This doesn't eliminate the need for verification—you should still check important claims regardless of how you prompted. But specific prompts make verification easier (because claims are more concrete) and reduce the volume of content you need to verify (because the response is more focused).

Your Challenge: Try It Yourself