What AI Does Not Know
The most common prompting failure has nothing to do with how you write. It has to do with what you leave out.
AI is supposedly on the verge of superintelligence, about to know everything and do everything. Then you ask it to write something. An email to your ex. An article for the company newsletter. And what comes back has that “slightly-too-clean” feel, competent and weightless at the same time.
Maybe AI isn’t all it’s made out to be.
The information you do not realise you are holding.
Here is how most people use AI. You want to ask AI to draft a case for changing how your company creates content. You think you should move from long-form videos to short three-minute clips. You know your audience will not sit through a 48-minute instructional video, but they will happily watch someone fillet a fish on TikTok. You know, the current creative team was there before social media and gravitates toward “creativity” over practicality. You know the last attempt to change direction was met with coordinated pushback, mostly from Dan, the 67-year-old Creative Director who has been there longer than anyone. You know the old methods are not working. And you know money is tight.
Then you type your prompt: “Write me a proposal to switch our content strategy to short-form video.”
And the model gives you something generic, and you are surprised.
You should not be. Everything that mattered, the fish, the audience, Dan the main man, the money, stayed in your head. None of it reached the prompt. Unless you put it in, it does not exist in the conversation.
This is the single biggest gap between people who get useful AI output and people who do not. What separates them is a habit: recognising how much context you are silently assuming the model has.
The shift that fixes it
Before you write the prompt, ask one question.
What does the AI need to know to give me what I actually want?
That single reframe changes how you approach every interaction. You stop writing requests and start writing briefs. You stop hoping the AI will guess what you mean and start telling it.
Run a quick audit on any prompt before you send it:
What is the situation behind this request?
What have I already tried?
Who is this for?
What does success look like?
What should it avoid?
If any of those are missing or vague, fix that before you hit enter. The thirty seconds it takes will save you twenty minutes of editing on the other side.
A worked example
Take a common scenario. You need to write an email to a client about a project delay.
The thin prompt: “Write me an email to my client about the project delay.”
The output will be polite, generic, apologetic, and unusable. It does not know your client, the cause of the delay, or the dynamics of your relationship.
The audited prompt: “Write an email to a long-term client (eight-year relationship) explaining a three-week delay on their website redesign. The cause is on our side: a key developer left mid-project. The client has been somewhat difficult to communicate with in the past, so the tone needs to be direct and accountable rather than grovelling. Keep it under 150 words. End by proposing a specific recovery plan rather than just apologising.”
The audited prompt contains nothing you did not already know. It just makes the silent context visible to the AI model.
The output will also be dramatically better.
The habit, not the technique
The reason this works is simple. The audit forces you to articulate what is already in your head.
This is also why CAST is a strong scaffold when putting together a prompt. Context, audience, structure, tone. Four questions designed to surface what you are silently assuming AI knows. Run them through before you send, and you will notice that the output improves noticeably.
Coming next
The next article builds on this idea from a different angle. Instead of describing what you want, show it. Examples often communicate what no amount of explanation can. We will look at why and how to use it.
The output you want is already in your head. Put it in the prompt.