daBongo LMS AI Training Courses

Lesson Content

Why prompts matter – the mechanistic view.

AI language models generate text by predicting the most likely continuation of the input they receive. Vague input produces generic continuation. Specific, well-framed input constrains the probability distribution toward the kind of output you actually need. This is not a stylistic preference – it is how the underlying mechanism works.

When you give Claude "write a presentation," it generates the statistically most likely response to that kind of request – a generic presentation. When you give it a specific role, audience, purpose, structure, and constraints, it generates text in that constrained space, which happens to be much closer to what you actually needed.

Role-framing.

Asking AI to adopt a specific role or perspective shapes the frame through which it interprets your request and generates output. Compare:

• "Review this proposal." -> Generic review from no particular frame.
• "Review this proposal as a skeptical senior investor evaluating whether to fund it. Focus on financial viability, market assumptions, and execution risk." -> Review from a specific analytical frame with specified dimensions.

Role-framing does not make AI more accurate – it focuses the generative process toward a useful perspective.

Constraint-setting – narrowing the space.

Constraints are not limitations on Claude's capability. They are instructions that narrow the output space from "everything technically possible" to "what you actually need." Good constraints prevent the most common output failures:

• Length constraints prevent bloat
• Tone constraints prevent register mismatch
• Scope constraints prevent topic drift
• Negative constraints (do not include X, avoid Y) prevent predictable failure modes

Output specification.

Specify the structure of the output you need. AI defaults to prose with headers – which may not be what you need. Specify: table, bulleted list, numbered steps, Q&A format, two-column comparison, executive summary followed by detail. Output structure is a choice you can make explicitly.

Prompt failure signals.

Before sending a prompt, scan for these failure signals:

• The task verb is vague ("discuss," "help with," "look at") -> replace with specific action
• No format guidance included -> add length and structure
• The context is missing -> add role/audience/purpose
• The request would be ambiguous to any competent human -> it will be ambiguous to Claude too