Prompting for Real-World Results

See how an assistant interprets a request and why small wording changes shift results.

• What the model sees: instructions, context, and conversation history
• Why the same prompt can give different outputs
• Vague versus specific requests, side by side
• Common failure patterns: generic, overlong, or off-target answers
• Baseline testing: saving the first output before improving
• Setting up a prompt-testing notebook

Hands-on lab: Take five vague requests from your own work, record the raw outputs, and note what is missing in each.

Define what success looks like before writing the prompt.

• Stating the goal and the finished deliverable
• Assigning a useful role without overclaiming expertise
• Naming the audience and reading level
• Defining success criteria up front
• Scope limits: what the assistant should leave out
• Asking the assistant to ask clarifying questions first

Hands-on lab: Rewrite your five baseline prompts with a clear objective and audience, then compare new outputs against the baselines.

Supply the right background and control the shape of the answer.

• Which context helps and which just adds noise
• Providing source text and instructing the model to stay within it
• Length, tone, structure, and style constraints
• Requesting tables, bullet lists, and templates
• Delimiters that separate instructions from data
• Removing sensitive details before supplying context

Hands-on lab: Build a prompt that converts raw notes into a fixed template, then test it on three different sets of notes with sensitive details removed.

Use examples and multi-step sequences to handle harder tasks reliably.

• Few-shot prompting with good and bad examples
• Step-by-step reasoning requests and when they help
• Splitting one complex task into a prompt chain
• Passing output from one step to the next
• Self-review prompts: asking the assistant to critique its draft
• Testing with edge cases and unusual inputs

Hands-on lab: Design a three-step prompt chain for a real task in your work, run it on three inputs, and record where it breaks.

Judge outputs with a rubric and verify claims before trusting them.

• Building a simple quality rubric
• Checking facts, numbers, and quotations independently
• Recognizing hallucination and overconfident phrasing
• Asking for uncertainty, assumptions, and sources
• Bias and tone review for people-facing content
• Comparing outputs from different assistants

Hands-on lab: Score outputs from two assistants against your rubric, verify every checkable claim, and revise the prompt twice to raise the score.

Document, version, and share prompts so results are repeatable.

• A prompt template: purpose, inputs, output, and review step
• Versioning and recording what changed
• Custom instructions and saved projects in assistants
• Sharing prompts with a team safely
• Prompt safety: avoiding hidden private data in templates
• Knowing when a prompt is not the right tool

Hands-on lab: Document eight of your best prompts in a shared sheet with inputs, expected output, test results, and review notes.