Want better results from ChatGPT? Start with two steps.

  1. Be clear about the job.
  2. Tell it what the answer should look like.

That’s it.

A lot of advice about prompting still makes this sound more complicated than it is. Use the magic phrase. Assign the perfect role. Add more context. Add more adjectives. Hope it all comes together.

Sometimes that helps. Usually the bigger win is simpler.

Fuzzy Prompts Are Bad Prompts

Fuzzy prompts sound like this:

Summarize this. Help me think through this. Write something about this topic.

Those are not really instructions. They’re openings.

The model has to guess what matters, how detailed to be, what format you need, and what “good” even means in the first place.

So it guesses. And the guess is usually generic, bloated, and mildly annoying.

A useful principle here: reduce the number of places the model can make a mistake.

These systems are probabilistic. A little non-determinism is part of the deal. If you leave too much open, you are basically inviting the model to improvise in the exact places you wanted precision.

A Better Default

A much better starting point is this:

Task: [ultimate goal]
Context: [relevant details only]
Return: [exact format and length]
Optimize for: [clarity, precision, completeness, speed]
If information is uncertain, say so clearly.

I like this because it forces the important parts into the open.

Part of the value of using AI well is that it pushes you to think more clearly and more structurally. A lot of the outcome is decided before the model starts generating. The planning and context do more work than most people realize.

Ask yourself:

  • What is the job?
  • What context actually changes the answer?
  • What shape do I need back?
  • What matters most here?

Once those are clear, the output usually gets better fast.

The Part More People Should Borrow

The newer GPT 5.4 guidance sharpens this in a useful way.

It puts more emphasis on output contracts.

That’s just a practical way of saying: tell the model what a good answer must look like.

Not just:

Summarize this.

More like:

Return exactly 5 bullets covering:
- main point
- supporting evidence
- risks
- open questions
- recommended next step
Keep it under 120 words. No preamble.

Think of it like giving the model a canvas with boundaries.

That one change solves a surprising number of problems. It cuts filler. It reduces drift. It makes the answer easier to scan. It also makes the output easier to judge because the model either followed the brief or it didn’t.

For most people using ChatGPT at work, this is more valuable than another pile of prompting tricks.

The next level after that is templates and schemas. Once you know the shapes of answers you keep asking for, reuse them.

If the Output Is Bad, Fix This First

Before you rewrite the whole prompt, check three things.

  1. Was the task actually clear?
  2. Did you give context that changes the answer?
  3. Did you define the output shape?

If not, start there.

Most of the time, the problem is not that the model needs more creativity.

It’s that it needs a cleaner brief.

One practical trick: revise your prompt, then paste the earlier version back in and label it clearly.

Old Prompt:
[paste prior version here]

That costs a few extra tokens, but it can help the model anchor to the progression of your thinking instead of half-reconstructing it from the latest revision.

The Point

Define the job. Define the output. Tighten from there.

That will get you farther than most ChatGPT advice on the internet.