Does prompt length matter?

Relevance matters more. Long prompts full of real context work; long prompts full of filler do not.

Should I include examples?

Yes. One or two examples of the desired output outperform paragraphs describing it.

Why does the same prompt give different results?

Generation is not deterministic. Constraining format and length reduces the variation that matters.

How do I stop invented facts?

Supply the source material and instruct the model to use only that, then verify against the source.

Do prompts transfer between models?

The structure does. Retest specific wording when you switch.

Is there one best prompt for writing?

No. The right prompt depends on audience, purpose and constraints.

PromptHub

Build and refine prompts that produce usable output.

PromptHub is a place to build, store and refine prompts. It structures a prompt into the four parts that matter — task, context, constraints and output format — and keeps the versions that worked so you are not rewriting them from memory.

Workspace

The PromptHub interface runs here. While it is being connected, the guidance below covers how to use it, what it handles well and where it falls short — the same information you would want before running a first job.

Read the steps Browse related guides

How to use PromptHub

1Pick a starting point from the library or begin with a blank structured prompt.
2Fill in the task, the context the model cannot infer, your constraints and the output format.
3Add one or two examples of the output you want.
4Run it, then change one element at a time when the result misses.
5Save the version that worked so you can reuse it.

Practical examples

Vague request turned specific

Input

"Improve this paragraph."

Result

"Rewrite this paragraph so a non-technical reader understands it. Keep both figures exactly as written, keep it under 90 words, and do not add new claims." The second version produces usable output on the first attempt.

Structured extraction

Input

A request to pull fields from pasted meeting notes, with the output format specified as JSON keys: decisions, owners, deadlines.

Result

Consistent JSON on every run, so the result can be pasted straight into a tracker rather than reformatted by hand.

Supported formats

Plain text prompts
Structured prompt templates
Saved prompt library entries

Features

Four-part prompt structure: task, context, constraints, format
Library of starting points by task type
Version history so working prompts are not lost
Notes on which model a prompt was tuned against

Limitations

Prompt structure transfers between models; specific phrasings and formatting quirks do not.
No prompt prevents a model from being confidently wrong on facts.
Prompts cannot supply information the model does not have; you must include it.
Multiple unrelated questions in one prompt get uneven answers — split them.

Privacy and data handling

Prompts you save are stored for your own reuse. Avoid embedding confidential material, personal data or credentials in a saved prompt.

The full statement is in our privacy policy.