Most prompt advice is the same for every model: be clear, give background, say the format, show an example. A few details differ. Promptor has a “Works with” choice so its tips fit the AI you are writing for, and All keeps to what holds for every one.
What every AI shares
- A clear goal, with background and the audience.
- The format and size, as numbers.
- Pasted text marked off from the instructions.
- What to do, not only what not to do.
- An example when the format keeps coming out wrong.
- No keys or personal data.
Claude
Anthropic’s guide recommends wrapping each kind of content in its own XML tag, such as instructions, context and input, with consistent names. For very long documents it recommends putting the document above the question. It also says to be explicit about the format and the constraints, and gives the colleague test: if a newcomer would be confused by the prompt, so will Claude.
ChatGPT
OpenAI’s guide says to put the instructions first and separate them from the text with ### or triple quotes, to show the output format with an example, and to replace vague wording with specifics. For code, it suggests a leading word such as import or SELECT to nudge the model toward a language. It also recommends using the latest model, since newer models are easier to prompt.
Copilot
For GitHub Copilot, the same general advice applies. Standing rules, such as how your team writes code, belong in the project instructions (.github/copilot-instructions.md or AGENTS.md) rather than in every prompt. See Copilot instructions and path instructions.
Reasoning models
A reasoning model works through a problem before it answers. OpenAI’s guide notes that prompting these models differs from prompting other GPT models. In practice, give the goal and the constraints and let it decide how to think, rather than adding “think step by step”. For a standard model, Atlassian’s tip to ask it to explain its process can help you see how it got there.
Settings outside the prompt
OpenAI’s guide also covers settings such as temperature, maximum length and stop sequences. A low temperature suits factual work such as extraction, and the maximum length cuts the answer off rather than shortening it. These are API settings: chat tools and coding assistants usually do not expose them.
Where they disagree
Instructions first, or a long document first? OpenAI says instructions first; Anthropic says a very long document goes first. Say what not to do, or only what to do? Atlassian values both, while OpenAI and Anthropic prefer what to do. Promptor follows each source where it applies and says which one in its Why?.