Which tool to choose
Here the choice isn't of the tool but of the way of writing: the goal is precisely not to depend on a particular assistant. It's worth keeping prompts portable when you use more than one (one for work, a free one at home, one built into an app), or when you want to be free to switch without rewriting everything. If you'll always and only use the same assistant, you can also exploit its proprietary shortcuts; but a portable prompt costs you nothing extra and keeps your hands free.
How to do it
A portable prompt is a prompt that doesn't take for granted who reads it. The more explicit it is, the more it crosses different assistants unscathed, because it doesn't lean on behaviors that only one of them has.
- Open with the role and the task in the clear: "you are a proofreader, correct this text".
- Give the context inside the prompt, not entrusted to the assistant's memory: what the text is, who it's for.
- Indicate the form of the result concretely: list, table, length.
- Write the constraints as positive, clear instructions, not as tricks: "answer in Italian, maximum ten lines".
- Avoid references to features of a single product ("use mode X"): if they don't exist elsewhere, the prompt breaks.
The operational syntax, portable prompt:
Role: you are an assistant that summarizes technical documents for non-specialists.
Task: summarize the text below.
Constraints: maximum 120 words, no technical terms without explaining them, neutral tone.
Form: three sentences of synthesis, then a list of at most four key points.
Text:
[paste here]
To verify portability, try the same prompt on two different assistants. If both return something useful and similar in form, the prompt is solid; if only one understands it, there's something too specific in it to remove.
A concrete example
Roberto uses the work assistant at the office and a free one at home, and had built himself prompts that worked only on the first because they leaned on a particular feature of it. At home he got poor responses. He rewrites his three main prompts in explicit form: role, task, constraints, form, all written out. Now the same prompts work on both, and when the company changed assistant he didn't have to redo anything. Portability protected him from a change he hadn't chosen.
When it does NOT work (and how to fix it)
If the same request gives very different results between one AI and another
Almost always the prompt leaves room for interpretation and each assistant fills it in its own way. Tighten the constraints: add exact length, precise form, an example of the wanted result. The more you close the doors to interpretation, the more the responses converge.
If you used a special feature another assistant doesn't have
Proprietary tricks (particular modes, special commands) don't travel. Rewrite that part in normal language: instead of invoking a mode, describe in words the behavior you wanted. Plain language is the only thing all assistants understand.
If the prompt has become very long to cover all cases
A portable prompt doesn't have to anticipate every assistant: it just has to be clear. If it's bloated, cut the defensive instructions ("if you're assistant X do this") and keep a single clean version. Clarity makes the special clauses pointless.
A tip from someone who actually uses it
Treat your best prompts like a toolbox that has to work in any workshop. Write them imagining that tomorrow the assistant you use disappears: what would another one need to understand you right away? That version, explicit and without hidden shortcuts, is the one worth saving. It makes you independent of the tool of the moment.
Frequently asked questions
Do all assistants understand prompts in the same way?
Broadly yes for clear requests, but each has its tendencies: one more long-winded, one more terse. An explicit prompt reduces these differences because it leaves less room for each one's habits. It's precisely vagueness that makes the responses diverge.
Do I have to write prompts in English to make them work everywhere?
No. Recent assistants handle Italian well, and writing in your own language makes you write clearer prompts, which is what counts for portability. Language isn't the barrier; ambiguity is.
Will a prompt that works on one AI also work on the next version?
It's not guaranteed, and that's why it's worth writing them explicit. Prompts that lean on a particular behavior of a version can break at the first update; the ones that say everything clearly hold up to changes much better. Writing explicit is also a form of insurance for the future.