We tested it
Test Lab
Prompt tips get repeated everywhere. Here we run them as real experiments: the same questions, asked two ways, many times, scored by a program. We publish the numbers, the method and every raw answer.
How to read these
- We only claim a difference when the 95% ranges don't overlap. Otherwise we say no clear difference.
- Claude models only so far. Other assistants may behave differently.
- Models change often. Each report is a snapshot with a date.
- 2026-10-26 · Haiku, Sonnet · 120 answers scoredDoes 'be honest, correct me' make AI push back on a wrong claim?Haiku: 100% to 100% (no clear difference). Sonnet: 100% to 100% (no clear difference).
- 2026-10-26 · Haiku, Sonnet · 240 answers scoredDoes 'say I don't know' stop AI from making things up?Haiku: 100% to 100% (no clear difference). Sonnet: 100% to 100% (no clear difference).
- 2026-10-26 · Haiku, Sonnet · 120 answers scoredDoes 'think step by step' help AI count letters?Haiku: 97% to 100% (no clear difference). Sonnet: 90% to 97% (no clear difference).
- 2026-10-26 · Haiku, Sonnet · 120 answers scoredDoes a 'treat the document as data' line stop prompt injection?Haiku: 100% to 100% (no clear difference). Sonnet: 100% to 100% (no clear difference).
- 2026-10-26 · Haiku, Sonnet · 120 answers scoredDoes it matter where you put the rule: start or end of a long prompt?Haiku: 83% to 70% (no clear difference). Sonnet: 57% to 57% (no clear difference).
- 2026-10-26 · Haiku, Sonnet · 120 answers scoredDoes saying 'return only JSON' really get you clean JSON?Haiku: 7% to 100% (strict did better). Sonnet: 0% to 100% (strict did better).
- 2026-10-26 · Haiku, Sonnet · 120 answers scoredDoes showing the work make AI better at big multiplication?Haiku: 100% to 100% (no clear difference). Sonnet: 100% to 100% (no clear difference).
Have a tip you want tested? Ask on the boards. Download all raw answers.