AI

Practical AI Prompting That Actually Works

A grounded, repeatable approach to prompting large language models — the patterns I reach for every day, minus the magic incantations.

Ryan Lee2 min read

There's no secret prompt that unlocks a model's hidden genius. What works is unglamorous: clear context, explicit constraints, and a willingness to iterate. Here's the approach I use for real work.

Start with the goal, not the phrasing

Before typing anything, get specific about the outcome you want. A vague request produces vague output. The single biggest quality improvement comes from describing the finished result in concrete terms.

Give the model a role and a reader

Tell it who it is and who it's writing for. "Explain this to a smart beginner" produces very different output than "write this for a senior engineer." Context shapes tone, depth, and vocabulary.

Constrain the output

Models are eager to please, which means they'll happily ramble. Constraints fix that:

  • Set a length: "in three sentences" or "under 150 words"
  • Set a format: a table, a numbered list, JSON
  • Set a boundary: "don't include marketing language"

Constraints aren't limitations — they're how you get usable output on the first try.

Iterate in small steps

The best results come from a conversation, not a one-shot prompt. Generate something rough, then refine:

  1. Get a draft that's roughly right
  2. Fix the biggest problem
  3. Repeat until it's good enough

This is faster than crafting the "perfect" prompt up front, and you learn what the model actually needs.

When to start over

If three refinements haven't fixed a problem, the issue is usually in the framing, not the details. Clear the context and restate the goal from scratch.

Verify everything that matters

Models are confident even when they're wrong. Treat output as a fast first draft, not a source of truth. Anything factual, numerical, or consequential gets checked. This one habit separates useful AI work from embarrassing mistakes.

The real skill

Prompting well is mostly clear thinking made explicit. If you can describe exactly what you want to another person, you can get it from a model. The tool rewards clarity — so the practice is worth it far beyond AI.

#prompting#llm#workflow
Share

Related reading

Business2 min read

Why I'm Building AI Workbench in Public

The reasoning behind starting a long-term content platform that documents a real AI learning journey — not another hype-driven AI company.

Enjoyed this? Get the next one.

Join the newsletter for practical AI notes — one useful email at a time.