Prompting - Best Practices · Digital Roadmap AI Academy

Prompting - Best Practices

These are practical, high-impact tips to make your prompting sharper, more efficient, and more future-proof. Each point calls out a common blind spot and gives you an actionable step you can apply immediately. They come from hands-on use across multiple models — not theory — so you can avoid wasted time, confusion, and frustration.

Save this list, revisit it, and watch your results improve over time.

Break prompts into reusable modules when they're quite big: short, focused blocks you can swap or rearrange without rewriting everything.
Use flexible wording, avoid model-specific quirks, and test prompts on multiple models while keeping a fallback version.
Explore prompts that instruct the AI to improve or analyze other prompts, building a meta-level skill set.
Record the model version in your notes and maintain a versioned prompt library for each model.
Review at least the sections on prompt formatting, capabilities, and limitations to uncover hidden features.
Decide the exact format you want (table, markdown, JSON) before prompting to ensure clean, reusable results.
Refine in steps, asking the AI to improve its own output until it meets your target standard.
Split large tasks into separate threads or requests to keep each focused and avoid token overload.
Skim the docs to learn about functions, system prompts, and structured outputs — even if you don’t code.
Complete at least one learning path to gain tested best practices straight from OpenAI.
Search for prompts and solutions others have shared to shortcut your own problem-solving.
When asking for edits or improvements, also request an explanation in a table showing why, what, and how.
Turn specific answers into generalized templates or frameworks so they work in multiple contexts.
Run edge cases, tricky inputs, and unusual instructions to discover weaknesses early.
Add “Explain your reasoning step-by-step” to spot logic gaps and strengthen your results.

Bonus Tip:

You can even use GitHub or another version control system to track changes over time for your prompts and maintain a well-organized versioned prompt library.

Resources:

EDIT: And, if you wanna test your prompts like 4o vs 5:

Just go to https://gptblindvoting.vercel.app/ (green button, as framed in the pic below)

P.S: The right button is to do an anonmyized test without showing model names it's using so you can learn about which model you would choose the most.

3 comments