Guides

Seven simple principles for better prompts

The quality of an answer usually tracks the quality of the request. These seven simple habits make your prompts in VIGPT sharper — and the results closer to something you can actually use.

VIGPT TeamVIGPT Team4 min read
A short, vague request beside the same request after it's been sharpened
On this page

The model can't read your mind. What you get back usually depends on what you put in the message — not a magic word, but a few simple habits in how you ask.

This post lines up seven principles. None of them is complicated, and you don't have to apply them all at once — even the first two or three make a difference.

1. Say what the output should do

Most weak prompts have a vague verb. "Write an email" doesn't tell the model what the email is supposed to achieve.

Instead, say what the output should do:

  • Instead of "write an email" → "write a short, polite email to my landlord saying this month's rent will be three days late, and apologizing for it."
  • Instead of "review this contract" → "in this contract, tell me which clauses work against me and which ones are worth negotiating."

A clear verb and a clear goal is half the work.

2. Give context

The model knows nothing about your situation unless you tell it. Who's the audience, where will the text be used, what are the constraints — those are what move an answer from "generic" to "right for your job."

An example:

"Write a caption for an Instagram post."

versus:

"Write a caption for an Instagram post. The product is an online coding course, the audience is first-year students, keep the tone friendly, and avoid heavy technical jargon."

The second is longer, but the model no longer has to guess.

3. Show an example

If you have a style or format in mind, don't describe it — show it. One good example is more precise than several sentences of explanation.

  • "Look at these three headlines I've written before and make ten new ones in the same style."
  • "Give the answer in exactly the structure I've used in this example: one summary sentence, then three points."

Showing an example works best when the tone or the format matters to you.

4. Set the format and the length

If you don't say what the answer should look like, the model decides for you — and you might get a long paragraph when you wanted a short list.

So say it:

  • "As a table, with two columns: pro and con."
  • "In 100 words or fewer."
  • "Give three options, one sentence each."
  • "Just the code, no explanation."

Setting the format makes the model's job easier and your review easier too.

5. Ask for one thing per message

Five requests in one message usually means five half-finished answers. It's better to break the work up and go step by step: get the text first, then ask for it shorter, then ask for a title.

You don't need to worry about losing the thread. The conversation continues and the model remembers what you've said, so each message moves the same topic forward without re-explaining.

6. Give it the material instead of describing it

If you have the material, don't retell or retype it — hand it over. That's both more accurate and faster.

  • Instead of describing a report, attach the file. In the message bar, the add photo and file button is exactly for this. A PDF is accepted up to 15 MB.
  • Instead of paraphrasing an error message, copy the actual text and paste it in.
  • If the answer needs current information — a price, news, a recent event — use web search so the answer is grounded in up-to-date data (part of the Standard plan and above).

Which model to use isn't your concern. When you attach a file, Auto knows the message needs a model that can read the document and picks accordingly.

7. Refine, don't restart

The first answer is rarely exactly what you want — and that's fine. Instead of clearing the conversation and starting over, say what to change:

  • "Make it shorter."
  • "Make the tone more formal."
  • "Cut this part and put an example there instead."

Each time you refine, the model gets closer. Your history and attached files stay put, even if you switch models mid-way. And if you want to see how a different model answers, you can compare the same question across a few of them.

In short

The seven principles, at a glance:

  • Say what the output should do.
  • Give context so the model doesn't guess.
  • Show an example.
  • Set the format and the length.
  • Ask for one thing per message.
  • Give it the material instead of describing it.
  • Refine, don't restart.

A good prompt isn't a secret incantation; it's just a clear request. The clearer you ask, the less you have to ask again.

Ready to try VIGPT?

Dozens of AI models in one place — start for free.