Key takeaways up front
- Six repeatable patterns cover most business prompting
- Evaluation is the pattern most people skip
- Templates matter more than clever wording
Prompting is not a trick vocabulary. It is a small set of patterns you reuse, plus a habit of checking output against a standard you defined in advance.
1The six patterns
Learn these in order; each one fixes a specific failure mode.
- Role — tell the model whose judgement to imitate
- Context — supply the source material rather than trusting memory
- Constraints — length, format, tone, what to refuse
- Examples — two or three gold outputs beat a paragraph of instruction
- Evaluation — define what a good answer contains before you read it
- Iteration — change one variable at a time and keep a log
2Templates by function
A template is a prompt with slots. Build one per recurring task and store it where your team can reuse it.
- HR: JD-to-screening-rubric converter with bias constraints
- Sales: account research brief with source citation requirement
- Finance: variance commentary with a fixed section structure
3The mistakes that waste the most time
Most poor output traces back to missing context or an undefined success criterion — not to the model. Fix those two before rewriting phrasing.
What to do next
- Write the evaluation rubric before the prompt.
- Store prompts as reusable team templates.
- Change one variable per iteration and log the result.
#Prompt engineering#AI for HR#AI for sales