ChatGPT prompt improver
Paste a rough prompt, get a model-ready version that follows ChatGPT's conventions. Free. No sign-in.
no sign-in · nothing stored · free
What good ChatGPT prompts look like
ChatGPT can answer a loose request, but an answer is easier to use when the request defines the job. A practical prompt says what decision or deliverable is needed, supplies only the context that changes the answer, and makes the expected shape visible. This does not require a long template. For most work, a direct task, a few constraints, and a short output contract are enough. The aim is to remove consequential guesses while leaving the model room to do the work.
Name the deliverable and its reader
Start with the thing you need: a decision memo, debugging plan, customer email, comparison, or explanation. Then identify who will use it and why. "Draft a renewal email for a customer whose contract expires Friday" gives ChatGPT a clearer job than "write an email about renewal." Add background only when it affects the result, such as the relationship, deadline, policy, or evidence available.
Use a role when it changes the standard
A role is useful when it selects a real perspective or professional standard. Asking for a review from the perspective of a UK employment solicitor or a staff accessibility engineer changes what should be checked. Asking ChatGPT to be a genius does not. State the relevant jurisdiction, discipline, or audience instead of decorating the prompt with praise. For regulated or high-stakes work, request issues and questions to verify rather than presenting the output as professional advice.
Turn multi-part work into an ordered sequence
If the task has dependencies, write them in order. Ask ChatGPT to extract facts before it compares options, or to ask for missing inputs before it recommends a plan. Numbering also makes review easier because each requested step has a visible place in the response. Keep independent tasks separate. A prompt that asks for research, a strategy, a launch plan, and finished copy in one pass usually hides decisions that should happen between those stages.
Separate source material from instructions
When you paste notes, a transcript, or a document, mark its boundaries and say how it may be used. A simple "Source material:" block is often enough. Tell ChatGPT whether it may use outside knowledge, whether every claim must come from the supplied text, and how to handle gaps. This prevents background material from being mistaken for a command and makes unsupported additions easier to spot during review.
Finish with an output contract
Specify the shape you will actually use: five bullets, a table with named columns, valid JSON, or a 150-word email. Add tone only in observable terms, such as "plain language, short sentences, no sales language." Include exclusions that matter, not a generic list of virtues. If you need citations, calculations, or uncertainty labels, request them explicitly. A concise format contract saves the cleanup that follows an otherwise competent but inconvenient answer.
Make unknowns visible
A prompt should not force certainty where the input is incomplete. Tell ChatGPT to ask up to three questions before answering when missing facts could change the recommendation. For research or analysis, ask it to distinguish provided facts, assumptions, and open questions. For a draft, use bracketed placeholders instead of invented names, figures, or dates. This preserves momentum without allowing a polished response to conceal unsupported details.
Before and after
The rewrite keeps the original goal but turns an advice request into a two-stage conversation. ChatGPT must gather the facts that determine the negotiation, then return a plan with a usable script. The relationship constraint changes the tone, while the instruction not to invent market data makes a common source of false confidence visible.