Home / GenAI Tutorial / Day 3
Day 3 — Being specific - the biggest lever
Every vague word in your prompt is a decision you have handed to a machine that cannot ask you what you meant.
Vague words are unanswered questions
Ask for a "good summary" and the model has to decide what good means. It will pick something reasonable and generic. Ask for "five bullets, under 15 words each, aimed at a CFO who has not seen the data" and there is nothing left to guess.
The words that cost you
The workbench flags these because each one hides a decision:
| Vague | Specific |
|---|---|
| a good summary | five bullets, under 20 words each |
| some insights | the three largest variances against budget |
| improve this formula | make it work when the range has blanks |
| make it better | reduce it to one line without a helper column |
| relevant columns | Region, Product, Amount |
| recent data | orders on or after 1 April 2025 |
Try it
Every word in that prompt is doing no work. Now rewrite it with numbers and names:
Numbers make an answer checkable
This is the real reason to be specific
"Under 20 words" is not fussiness. It is a condition you can verify in one glance. A prompt full of measurable constraints produces output you can accept or reject in seconds. A vague prompt produces output you have to read carefully and still are not sure about.
Specific about what you do NOT want
Negative constraints are underused and extremely effective:
"Do not explain what VLOOKUP is" saves a paragraph you would have deleted. Every "do not" you can name is a round trip you avoid.
The specificity test
Before you press enter, ask: could two competent people read this prompt and produce very different work?
If yes, the prompt is underspecified - and the model will pick one of those interpretations without telling you which. Add the detail that separates them.
Try these yourself
- Rewrite "make a nice dashboard" with three measurable constraints.
- List five vague words you use often, with a specific replacement for each.
- Write a prompt using at least two negative constraints.
- Explain why measurable constraints make output faster to check.
- Apply the specificity test to a prompt you wrote yesterday.