Ask the AI what it needs

Asking the AI what it needs means letting the assistant name the data, documents, or examples that would improve its answers before you guess. You ask what would make its output more useful, then supply that.

Asking the AI what it needs means letting the assistant name the data, documents, or examples that would improve its answers before you guess. Instead of dumping everything you have or supplying nothing, you put the assistant in charge of scoping its own inputs, then you fill the gaps it points to. Say you want help drafting replies to customer questions. You ask the assistant what it would need to answer well, and it lists three things: a few of your past replies for tone, your return policy, and the product names you sell. You hand over those three, and the next draft lands far closer than a cold attempt. The assistant often knows what is missing before you do.

Builder example

When you are setting up an assistant for a recurring task, the slowest path is loading material by trial and error until the answers stop being generic. Ask the assistant up front what inputs would raise its accuracy, and it usually returns a short, specific list: an example of the format you want, the rules it should follow, the records it should check first. For a meeting-summary helper, it might ask for one summary you already liked and the names of the people who usually attend, which removes two rounds of vague output.

Common confusion: Asking the AI what it needs is not the same as asking it to do more work on its own. The assistant still cannot fetch private files you have not shared; it can only point to the inputs that would help, after which you decide what to hand over.