Most “the AI gave a useless answer” moments come from prompts that leave too much to guess. These seven patterns fix most of them.
In this article
1. Give the context a colleague would need
❌ Write a formula to find the price.
✅ In Excel 365, sheet Orders has product codes in A2:A500. Sheet Prices has codes in A and prices in C. Write a formula for Orders!B2 that returns the price, or “Not found”.
2. Show an example of what you want
One or two examples (“few-shot”) beat a paragraph of description: Convert these to title case like the example. Input: “ASHA VERMA, DELHI” → Output: “Asha Verma, Delhi”.
3. Specify the output format
Answer as a table with columns Step, Formula, Why. Or Return only JSON with keys name, email, city. Formats make answers easy to paste and check.
4. Ask for reasoning on hard problems
Work through it step by step, then give the final answer on its own line. For multi-step calculations and logic this measurably reduces errors, and you can check each step.
5. State the constraints
No VBA. Must work in Excel 2016. Under 150 words. Audience: new joiners. Constraints stop the model choosing a “better” solution you cannot use.
6. Separate instructions from data
“`text
Summarise the customer email between the triple quotes in 3 bullet points.
“””
“`
Delimiters stop text inside the data (like “ignore the above”) from being mistaken for your instructions.
7. Let it ask questions first
Before answering, ask me up to 3 questions if anything important is unclear. Great for plans, documents and code that depends on your setup.