Which tool to choose
Vagueness and repetition are flaws common to all assistants, and they're corrected by how you ask more than by the choice of tool. That said, if an assistant systematically gives you bloated responses, try setting your conciseness preferences in its custom instructions: some start out more long-winded than others and a system rule calms them down. For the rest, the lever is the prompt: specific going in, dry coming out.
How to do it
A generic response is almost always the mirror of a generic question: the AI fills the void of context with sentences that fit everyone, that is, no one. Repetitions, on the other hand, are its way of occupying space when it has little to say.
- Give real context: who you are, who the response is for, for what purpose. "Explain marketing to me" is vague; "explain to a florist how to find customers in their neighborhood" is specific.
- Impose constraints that force choices: maximum length, number of points, a precise cut.
- Forbid the filler: no preambles, no recaps, no "it's important to note that".
- Give an example of the level of concreteness you want: one sentence of "yes, like this" orients more than a thousand instructions.
- After a bloated response, ask for a trimming pass instead of redoing everything.
The operational syntax, to trim a response:
Rewrite the previous response keeping only the useful information. Cut:
- the premises and the general introductions
- the sentences that repeat a concept already stated
- the adverbs and adjectives that don't change the meaning
Aim for half the words, without losing any concrete data.
After the trim, reread: if the short version says the same things as the long one, the long one was bloated and the short one is the right one. If instead you've lost something useful, put it back by hand: aggressive trimming has to be checked.
A concrete example
Francesca asks for advice for her social media page and receives a very long response that repeats three times "consistency is important" and gives suggestions valid for any business. Useless. She redoes the question with context: "I have a flower shop, I post twice a week, I want more foot traffic in the shop not more followers". The AI stops talking in general and proposes three concrete actions tied to the physical shop. Specificity going in removed vagueness coming out.
When it does NOT work (and how to fix it)
If the response stays generic even with the context
Probably the context is there but the constraint that forces a choice is missing. Add a hard limit: "give me exactly the three most important things, not a list of ten". When the AI has to choose few things, it's forced to give the specific ones instead of the obvious.
If it repeats the same concept with different words
It's filler to occupy space. Tell it explicitly: "don't rephrase the same point; every sentence must add something new". If it persists, ask for a shorter version: the forced reduction eliminates the repetitions because there's no more room for them.
If asking for conciseness makes you lose important information
You trimmed without distinguishing filler and substance. Be precise about what to remove: "cut the introductions and the repetitions, keep all the data and the operational steps". Targeted trimming hits the fluff and spares the facts; blind trimming hits at random.
A tip from someone who actually uses it
Vagueness is fought before the response, repetition after. Going in, invest in context: every detail you give about yourself and your purpose is one less response to throw away. Coming out, get into the habit of asking for a trimming pass on long responses: thirty seconds and you read half the words for the same information. They're two different gestures for two different flaws.
Frequently asked questions
Why does the AI tend to be long-winded?
It tends to pad because a long, articulated response "seems" more complete and useful, and the models are pushed toward what appears thorough. Without a length constraint, the instinct is to add, not to choose. The explicit limit brings the response back to the essential.
Does asking for short responses also make them less accurate?
No, if you distinguish what to cut. Removing premises and repetitions doesn't remove information; short, targeted responses are often more accurate because they force the AI to say only what it's sure of. Accuracy is lost only when you cut at random, not when you cut the filler.
Isn't a long, detailed response better than a short one anyway?
It's the misunderstanding that makes you read three times as much for the same content. Length isn't quality: a long response full of repetitions and obviousness is worth less than four lines that actually answer. Measure a response by how much useful information it contains per line, not by how long it is.