Prompts that travel: how to write once for every model

If you switch models often, you’ve seen it: a prompt that works beautifully in one model produces something bland — or bizarre — in the next. The good news is that prompts that travel well share a handful of habits.

1. Say who it’s for

“Explain this” is ambiguous. “Explain this to a new engineer in their first week” is not. Audience is the single most portable piece of context you can give.

2. Show the shape of the answer

Models differ in their default formatting. Some love bullet points, some love prose. If you care, say so: “three short paragraphs”, “a table with these columns”, “under 100 words”.

3. Material first, question last

When you paste a long document, put it before your instructions. Most models handle “here’s the material, here’s what I want” better than the reverse, especially with long inputs.

4. Ask for assumptions

“Before answering, list the assumptions you’re making” works across nearly every model and surfaces misunderstandings before they cost you a rewrite.

5. Skip the magic words

Secret phrases that worked on one model six months ago rarely transfer. Clear, specific instructions always do.

A good prompt reads like a good brief to a smart freelancer: context, goal, constraints, format.

Test it in Compare

The quickest way to know whether a prompt travels is to run it across three models at once. If all three get it right, your prompt is clear. If one stumbles, the stumble usually points at exactly what was ambiguous.