Of the 1,678 real requests people have submitted to Briefing Fox, the second most common obstacle in the whole dataset, named in 33 of them, is not confusion about what they want. It is this: they know exactly what they want and cannot get it into words the AI will act on. “Has the facts but not the wording,” is how one person put it. This article is for that group, not for the 65 people whose obstacle was “no idea what to ask for,” because those two problems have different fixes and most advice about AI conflates them.
You already have the information. The prompt is losing it.
The pattern in the data is blunt. The median finished brief in the Briefing Fox corpus runs 26 words. The median abandoned one runs 13. Same person, same underlying need, half the information reaching the AI. Something is getting lost between what someone knows and what they type, and it is not a vocabulary problem. It is a structure problem: nothing tells them which thirteen extra words matter.
A prompt puts the burden of structure on the person typing, in the ten seconds before they hit enter, with no checklist and no feedback until the output arrives wrong. A brief moves that structure earlier, into a set of specific questions asked before the AI starts, so the information that would otherwise get lost has somewhere to go.
What a brief actually is
Briefing is not an AI concept. It is how every field that cannot afford a wasted first draft already operates. An ad agency briefs its creative team before a campaign starts. A structural engineer briefs the build crew before anyone pours concrete. In both cases the brief exists because asking the executor to guess is more expensive than answering the questions up front.
Applied to AI, a brief answers four questions before the model starts generating anything:
– Who is the AI acting as for this task?
– What does it need to know about this specific situation that a generic version of the task would not include?
– What is it not allowed to do, get wrong, or leave out?
– What shape does the finished output need to take?
None of these are hard questions. They are just the ones a prompt never asks, so most people never answer them until the output comes back wrong and they are answering them anyway, one correction at a time.
The same request, briefed and unbriefed
Take a task nearly everyone has handed to AI at some point: write a performance review for someone on your team. As a prompt, that is the whole request. As a brief, it carries the four answers: the reviewer’s own role and how directly they manage the person, three or four specifics about what actually happened this cycle rather than a generic evaluation, the constraint that it needs to read as encouraging rather than clinical because the company culture expects that tone, and a format that fits the company’s actual review template rather than a generic five-paragraph structure.
Both go to the same model. The prompt produces something that reads like every performance review the model has ever seen. The brief produces something specific to one person’s actual year, in the shape the reader is expecting. The gap between them is not what the AI is capable of. It is entirely what it was given to work with.
Why this is worth fixing rather than living with
The finish rate makes the stakes concrete. Requests that arrive already specific finish 94.6% of the time in the Briefing Fox data. Requests that arrive vague finish 1.4% of the time. That gap is not about which people are better at AI. It is about which requests carried enough structure to survive contact with a model that fills every gap with its best statistical guess rather than asking what you meant.
If you already have the facts and just lack the wording, that gap is closeable in minutes, not by learning a new skill, but by answering the same four questions before you type rather than after the first disappointing draft. Briefing Fox exists to ask those four questions for you and assemble the answers into a brief, but the questions themselves work whether or not you ever open the tool.
Before your next request
Write down four short answers before you open an AI tool: the role it should act as, the one or two specifics about your situation a generic answer would miss, the one constraint it must not violate, and the exact shape you need back. That is the entire discipline. It costs a few minutes and it is the difference between the 1.4% and the 94.6%.
Try Briefing Fox free at briefingfox.com
A prompt asks and leaves the structure to chance. A brief answers, before the AI starts, the four questions that determine whether the output fits your situation: the role it should act as, the specific context a generic answer would miss, the constraints it must respect, and the format the output needs to take.
Because knowing what you want and getting it into the request are two different steps, and most people skip the second one under time pressure. In Briefing Fox’s own usage data, the median finished request runs 26 words and the median abandoned one runs 13: the same knowledge, but half of it never made it into the prompt.
The role the AI should act as, the specific facts about your situation a generic version of the task would miss, the constraint it is not allowed to violate, and the exact shape the output needs to take. Answering these before you type takes minutes and determines most of what comes back.
No, and that is a common confusion. Not knowing what to ask for and knowing but not having the wording are different problems with different fixes. This is written for the second group: people who already have the facts and lose them in an unstructured request.