Which tool to choose
Saving doesn't depend on the tool but on how you write. That said, if you pay as you go (the per-token rates of developer tools), every word really counts; if you have a subscription with limited messages, it counts more not to waste the attempts. In both cases the lever is the same: the right prompt on the first try. Tokens (the units of text the AI counts in input and output) are saved by reducing the wasted rounds, not just by shortening the sentences.
How to do it
Every correction message you send because the first answer was wrong costs as much as the good answer. The real saving is getting the answer right on the first attempt.
- Open the prompt with the task, not with pleasantries. The AI doesn't take offense and you don't spend words.
- State the form and the length: "in five points," "maximum a hundred words," "only the table, no explanation."
- Give the necessary context all at once, instead of letting it emerge through back-and-forth.
- When you ask for a review, indicate precisely what to change, not "do it better."
- Change chat when you change topic: dragging a long conversation onto a new theme forces the AI to reread the whole history on every answer.
The operational syntax, a dense prompt:
Rewrite this email in a cordial but direct tone, maximum 80 words, keep the payment request and the deadline. Return only the email, with no comments.
[email text]
Compare with the wasteful way: "hi, I need a hand with an email, could you help me make it nicer? then maybe we'll shorten it." Four messages instead of one to get to the same point.
A concrete example
Luca uses the AI to generate product descriptions and kept hitting the daily message limit by mid-morning. He changes method: instead of one description per message with three corrections each, he sends a single prompt with the template filled in ("tone, length, what to include") and five products together. He gets five usable descriptions in one message, where before he made one in four. The limit now lasts him until evening.
When it does NOT work (and how to fix it)
If you shorten too much and the answer becomes wrong
The short prompt mustn't be poor: it must be dense. If you cut the necessary context, the AI guesses and you correct, spending more. Keep everything needed to nail the answer, cut only the pleasantries and the repetitions. The measure is the result on the first try, not the word count.
If the long chat has become slow and costly
A conversation that's been dragging on for hours makes the AI reread the whole history on every answer: you pay (in tokens or in waiting) even for the past. When a topic is closed, ask for a brief summary of what you decided, copy it, and open a new chat starting from that summary. Restart light.
If you use the same prompts every day and rewrite them by hand
Rewriting the same prompt every time is the real waste of time. Save the recurring prompts in a note or, if the tool allows it, as a quick command or a custom assistant. You change only the variable part.
A tip from someone who actually uses it
The most expensive message is the one you send to say "no, I meant something else." Before sending an important prompt, reread it and ask yourself: is there a way the AI can misunderstand this? Thirty seconds of rereading are worth three correction messages. That's where you save the most, not in trimming adjectives.
Frequently asked questions
What exactly are tokens and why do they matter?
A token is a piece of a word: the AI measures in tokens both what you write to it and what it answers. It matters because whoever pays as they go pays per token, and whoever has a subscription has a cap anyway. Long answers and endless chats consume tokens even when you don't notice.
Is it worth writing prompts in English to save tokens?
Marginally yes, English uses slightly fewer tokens for the same meaning, but the difference rarely justifies the effort of thinking in another language. If Italian makes you write clearer prompts, write in Italian: a clear prompt that works on the first try saves more than an English one you have to correct.
Does a longer prompt always mean a better answer?
No, it's the misconception that wastes the most. Beyond what's necessary, the excess instructions confuse the AI and dilute what matters. The best answer comes from the prompt that gives everything necessary and nothing else. Dense beats long.