Details in the Prompt
A prompt is the text you send into the chat: a question, a request, or an assignment. There is no special language for it; a prompt is written in ordinary words.
Write a prompt the way you would talk to another person.
Missing conditions do not cancel the answer. It arrives anyway, it just does not quite belong to your task.
Whatever you leave unsaid, the model picks for you. The fewer details of yours the prompt carries, the more of those picks it makes, and the further the answer lands from what you need.
Practice shows which details were missing. You read the answer, find decisions somebody else made for you, and next time you state that condition up front.
Three pairs of prompts follow. Each one shows what the added details change.
A listing for a used bike
The answer fits any bike at all: no make, no model year, no price.
The model knows neither the bike, nor its condition, nor the buyer. It fills the empty spots with what it has seen in listings like this.
Now let's tell it what it cannot know.
The second prompt is longer by exactly the details the model did not have. The rest it wrote on its own.
A vague requirement and a checkable one
"Beautiful" and "professional" mean something different to everyone. You have one picture in your head, the model has another, and they match by accident.
The second prompt can be checked against the answer. You can see the length, you can see the exclamation marks, and the price is either first or it is not.
A laptop for your conditions
The answer came out averaged: with no conditions from you, the model answers for everyone at once. You cannot choose from it.
The last line of that answer is a set of questions you could have answered in the prompt.
In the second prompt, you showed up: your conditions, your limits, your desk with the distant outlet. The answer became advice for your case.
When starting a new chat is easier
You add the missing conditions as the conversation goes. Each one costs two more messages: you name what did not fit, and you read the new answer.
With every correction the chat gets longer. Measurements across models from different makers show that as the conversation grows, answers lose accuracy well before the length limit the service allows.
By that point you know more about the task: which conditions matter, what format you need, what you do not want. A new chat with those conditions in the first message costs less than another dozen correction messages in the old one.