The answer that is correct and still not useful

There is a kind of answer that passes every test except the one that matters. It is true. It is precise. It is probably verifiable. And it leaves you with nothing you can act on, which is why it fails even though it is not wrong.

The failure is usually a missing step. A correct answer names what something is, or states a fact, but it does not connect that fact to the thing the person actually needed to do. The information is all there, and the usefulness is not.

Let me build an example that is entirely fiction, with every fact the useful answer will use stated up front, so nothing is invented halfway through. This is not a recorded interaction and not a real forecast.

A person drives to work across a bridge, and they usually reach that bridge at about 8:45 in the morning. A notice says the bridge will be closed tomorrow from 8 to 10. Their shift starts at 9. There is another route that takes twenty minutes longer. They ask: “Will the bridge closure affect my commute?”

A correct answer: “The bridge will be closed from 8 to 10 tomorrow.” This is true and precise. It is also, for the person deciding how to get to work, not very useful. They still have to notice on their own that their usual crossing time falls inside the closure, and that they should leave earlier or take the longer route.

A useful answer says the same true thing and then connects it: “The bridge is closed from 8 to 10, and you usually cross at 8:45, so you would reach it during the closure. Either leave early enough to cross before 8, or take the other route and add twenty minutes.” Every fact in that answer was already in the fiction: the 8:45 crossing time, the 8-to-10 closure, the nine o’clock start, and the twenty-minute alternate route. Nothing was added. What changed is that the supplied facts were connected to the decision the person actually has to make.

There is an older piece of advice that points at the same thing. The UNC Writing Center’s handout on audience says a reader wants the writer to make the necessary connections from one thought to the next, and that when the writer leaves those connections out, the reader gets confused. That is advice about writing a paper, not about AI answers, and it is not evidence of how any model behaves. I am applying its point to a narrower case: an answer that states a fact but skips the step connecting it to the reader’s goal is an answer with a gap in it.

This is a distinction worth keeping for anyone who writes with a model. A model can state facts cleanly, and it can produce the first kind of answer, the true statement with no step attached. What it may miss is the step, because the step depends on knowing what you were trying to accomplish, and that is the part you have to supply.

The question to ask when you get a correct answer is not is this true. It is does this tell me what to do next. If it does not, the answer is incomplete, and the fix is to ask for the next step explicitly. A correct answer and a useful one are not the same thing, and the distance between them is the part you fill in by knowing what you wanted the answer for.