AI answers run long mostly not because the model knows too much, but because it does not know how much you need. Give it no length and no structure, and it takes the safest route: background, causes, steps and caveats all laid out in full, reinforced by a training habit of saying everything completely, until a one-sentence matter becomes a small essay. The fix is not switching models. It is fixing the length and the order inside your question, and the four moves below are ranked from strongest to weakest effect.
Move one: fix the length, then the order
Add a hard requirement right after your question, such as answer in three sentences, stay under two hundred words, give the conclusion first and add one line of reasoning. Once length is a number, the model is forced to make trade-offs, and filler is the first thing cut. Order matters just as much: lead with the conclusion and put details last, so the answer is no longer buried in the third paragraph. Simply saying be brief barely works, because its idea of brief and yours differ completely; numbers are what make it real.
Move two: state what you do not want
Half of a long answer is often the polite opening, a restatement of your question and a courteous closing. Write explicitly: no preamble, do not restate my question, do not end by asking whether I need more help. Remove those three and the reply usually shrinks by a third at once. For recurring situations, such as organising the same kind of information every day, save these requirements in the custom instructions or system settings so you stop repeating them in every prompt.
Move three: ask one thing at a time
Pack three questions into one prompt and you will get three sections, or nine. Split compound questions apart, ask a single point, then follow up layer by layer. The result is not only shorter; it also makes it far easier to spot exactly where an answer went wrong. Conversely, when you genuinely want a full, developed treatment, skip this method. Length itself is neither good nor bad, only fitting or not. For research, detail is a virtue; what needs changing is only the default.
Move four: ask for an outline first
When you are unsure how deep you need to go, ask for bullet headings only, one line each, nothing expanded. A glance tells you which items are actually worth reading, and you can then name one or two for expansion. That minimises total reading and avoids a full screen spent on parts you do not care about. One boundary deserves stating: for contracts, medical or legal content where errors are costly, never force brevity that strips out premises and risk warnings. Such answers deserve their length; check them first and only then use them, rather than trusting a compressed conclusion.
