<aside>

On this page:

</aside>

<aside>

Expanse Docs

🧠 Product Docs

• Expanse 101
• Key features

🤖 Prompting Tips

LLM Directory

Prompting Tips & Tricks

Prompt Library

💻 Use Cases

For Work

For Fun

📚 Resources

Prompt Engineering Resources

🍄 About Expanse

Meet the team

</aside>

Do we need to be polite to our future AI overlords?

Ever wondered if saying "please" to ChatGPT actually helps? Turns out, it depends…

Researchers at Wharton tested GPT-4o and GPT-4o-mini on tough PhD-level questions, asking each question 100 times with different prompts. They tried being polite ("Please answer"), demanding ("I order you to answer"), and using various formatting instructions.

<aside> 💡

This article is based off academic research from Ethan Mollick et al (Source)

</aside>

Surprising findings

• AI models are inconsistent! Sometimes they get the same question right, sometimes wrong.
• Being polite helped with some questions but hurt performance on others.
• Formatting instructions (telling the AI exactly how to structure its answer) consistently improved results.
• How you measure "success" matters enormously - are you happy if the AI gets it right most of the time, or do you need 100% accuracy?

What this means for you

Don't believe in one-size-fits-all prompt "hacks." Instead, test different approaches for your specific needs. If you're using AI for something important, ask multiple times and compare answers. And yes, clear formatting instructions generally help, but being extra polite isn't always the magic trick some claim it to be.

Bottom line: The best way to prompt AI depends entirely on what you're asking and how reliable you need the answer to be!

See also:

Context Management Guide