Why Most Prompts Fail
Most people write prompts like Google searches: short, keyword-heavy, hoping the tool fills blanks. Copilot generates a response shaped entirely by your input, vague prompts get confident guesses, often wrong or useless.
Lesson 1
Search habits do not transfer to Copilot.
A specific prompt narrows the pattern space. You steer; the model guesses less.
Three prompt failure modes
Core principles
- Failure Mode 1: Missing Goal: topic is not an outcome. Say what you want created (summary, recommendation, draft) and for whom.
- Failure Mode 2: Missing Context: who you are, audience, constraints, what success looks like, bridge the gap the model cannot read.
- Failure Mode 3: Missing Expectation: format, length, tone, structure, saves rounds of "make it shorter."
- Smart-new-hire test: would a sharp new hire who never worked at MillerKnoll know exactly what to produce? If not, your prompt is not finished.
Check yourself
According to the smart-new-hire test, when is a prompt finished?
The test asks: would someone competent but totally new to this organization know exactly what to produce? If they would have to guess at goal, context, or expected format, the prompt is not finished.
Do this in Copilot
Review your last three Copilot tasks; label each failure mode. Do not rewrite yet, Lesson 2 gives the fix.
This prompt likely fails all three modes. Rewrite it with a clear goal, MillerKnoll context, and expectations, then compare outputs.
Diagnose a weak prompt
Tell me about workplace ergonomics.
- Failure-mode diagnosis
- Stronger version starter: Summarize the three most evidence-backed ergonomic interventions for desk workers, in plain language, for a non-specialist facilities audience, in five bullets under 120 words.
Did you run this in Copilot? Mark complete when you have tried it.
RecordedNext lesson: Three Frameworks That Go Beyond GCSE →
Navigate: press j for next lesson, k for previous.