Getting the Model to Show Its Work
How asking an AI to reason step by step improves accuracy on math, trade-off decisions, and multi-cause diagnosis.
Ask for the work, not just the answer
A math teacher asks students to show their work for a reason: it catches mistakes and proves understanding. Asking an AI model to reason step by step does the same thing: you get a more reliable answer, and you can see exactly where it might have gone wrong if something looks off.
- Calculations and multi-step math
- Trade-off decisions with more than one factor
- Diagnosing a problem with several possible causes
- Anything where a wrong intermediate step quietly breaks the final answer
Step 1: Monthly cost per person if split 3 ways = $2,340 ÷ 3 ÷ 12 = $65/month. Step 2: Remaining months in the year for the new member = 12 − 4 = 8 months. Step 3: Prorated cost = $65 × 8 = $520. The new member owes $520 for their first year.
Plan A: $40 × 12 = $480/year. Plan B: 3,000 × $0.02 = $60/month → $60 × 12 = $720/year. Plan A is cheaper by $240/year. Plan B only wins if monthly traffic drops below 2,000 visitors.
Match the trigger phrase to the task
When step-by-step goes wrong
Each step follows from the one before it, assumptions are stated rather than silently assumed, and the final answer is something you could verify by re-reading the steps, not something that appears out of nowhere at the end.