Context

How much background information AI needs before taking action

Context is the background information AI needs before it can give you a useful answer, and most of the time, not enough of it gets passed along.

When a human colleague starts a task, they bring years of implicit background: they know your clients, your constraints, your tone, the history of the project. AI starts from zero every time unless you bring that context with you. The quality of AI output is often less about the model and more about how much relevant background was in the prompt.

Context has different types. There is factual context (who the client is, what the product does), situational context (where we are in the project, what just changed), preference context (the tone we use, what the reader cares about), and constraint context (what we cannot say, what format they expect). Most people only pass the first type and wonder why the output misses the mark.

More context is not always better. Irrelevant background can dilute a prompt and push out the useful information. The skill is choosing the right context: specific, relevant, and just enough to close the gap between what the tool knows and what it needs to know.

Make it visiblePick one AI task you do regularly and write a context document for it: the relevant background, the audience, the constraints, and anything the tool should not do. Paste it at the start of your next prompt and note what changes.

Why AI needs this

Each part of your work matters to AI in a specific way. Some of it is context a tool needs before it can help, some of it is work a tool can take on, and some of it is judgment that should stay with you.

Project history and decisions

Without knowing what has already been decided, AI will suggest things you have already ruled out. A brief summary of where things stand saves that wasted step.

Who the output is for

The reader shapes the answer. A message to a technical team and the same message to a board need different language and framing. AI needs to know the audience.

What not to include

AI will include plausible content unless you tell it what to leave out. Constraints are context too, and they tend to be the most valuable kind.

The source material

When the task involves specific data, documents, or examples, passing those in directly almost always produces better output than asking AI to work from its general knowledge.

Questions to explore

Use these on your own or in a group. There are no right answers, only better conversations.

What background does a new person need to start working on this project well? How much of that do you pass to AI?
Which types of context do you regularly forget to include in your prompts?
How much context is too much? Where does passing more background stop helping?
Where does the useful context for your most common AI tasks actually live, and how easy is it to retrieve?
Which of your AI use cases would benefit most from having a standard context document to paste in?

Readiness traps

Assuming AI knows things it cannot know. It does not know your client, your history, or your preferences unless you tell it. Leaving that out produces confidently generic output.
Pasting in everything and hoping it helps. A dump of unorganised documents is not context. AI needs the relevant parts surfaced and clearly labelled to use them well.
Context that goes stale. A project brief from six months ago may mislead more than it helps if the situation has changed. Context needs to be current.