The Context We Leave Unsaid
What learning to give AI better context reveals about human communication, shared understanding, and explaining decisions at work.
When I started learning how to give AI better context, I expected better answers from a tool. I didn't expect it to change how I look at my own conversations.
With AI, a clear instruction isn't enough. It needs the background, the constraints and what you're actually trying to achieve. If you leave that in your head, the answer comes back confident and slightly wrong. So you learn to ask what the other side needs to know before it can help you.
Somewhere along the way, I started asking the same question about people.
We're taught to be clear, concise and get to the point. But choosing which context to include is part of communicating well too.
Think of a simple sentence like "I don't think we should go ahead with this." By the time you say it, you may have looked at the cost, spoken to a few people and found a problem nobody saw earlier. The person across the table hears none of that. They hear a rejection.
So you start explaining. To you, every detail matters because each one shaped the decision. To them, a simple no is turning into a long story, and a quieter question starts forming: why is he working so hard to justify this? You're trying to make your reasoning visible, and they're starting to doubt your intentions.
That's the tension. If you say too little, you're misunderstood. If you say more, the explanation itself can look like a defence.
While I was reading more about this, I came across a research paper by Herbert Clark and Susan Brennan called Grounding in Communication. They study how people actually reach understanding in conversation, and one of their examples comes from an old Marx Brothers film.
Groucho says to Chico:
"I want you to meet Mr. Jones."
He means it as an introduction: here is Mr. Jones, say hello. Chico hears the same words and takes them as an instruction to go and meet Mr. Jones somewhere, so he asks where he should meet him.
Nothing was wrong with the sentence. Chico heard every word. But the two of them walked away with completely different meanings, and if Chico hadn't asked that question, Groucho would never have known.
That's the point the paper makes. We usually assume we've been understood as long as nobody objects. In reality, understanding has to be confirmed by both people, through a reply, a question or a response that shows the meaning actually landed. Until then, we've only been heard.
Part of the problem is that we enter conversations at different points. I may have spent three days thinking about something you've spent three minutes hearing about. The reasoning feels obvious to me because I lived through it, while you only got the conclusion. Oddly, this gets worse with people close to us. Research shows we overestimate how well friends and partners understand us. With strangers we explain properly. With people we know, we assume.
Sometimes the context can't be shared at all. A manager may be protecting someone's privacy, or a decision may rest on information that isn't final. In those moments, share what you can:
"There are private circumstances I can't discuss. I can explain how we'll distribute the work, how long this will last and when we'll review it."
But there's a risk here too. Once we know some things can't be shared, it becomes easy to answer every question with "You don't have the full context." Sometimes that's true. Sometimes it's just a way to avoid a fair concern. People still live with the consequences of our decisions, even when they can't see every reason, so their questions deserve a real answer.
This is also where AI and people differ. An AI doesn't bring its own experience into the room. People do, and they may know something I've missed. So before a long explanation, I've started checking where the other person is starting from:
"Before I explain my decision, what have you heard so far?"
As a listener, I try to ask how a detail affects the decision before I dismiss it as an excuse.
More context won't guarantee agreement. But at least we'll be disagreeing about the same situation.
I started learning about context to work better with AI. It ended up showing me how much I leave unsaid with people. Now, before I expect someone to understand my next decision, I ask myself one question:
What could they reasonably understand from what I've actually told them?
I'm curious how others handle this. When you can't share the full reasoning behind a decision, what do you tell your team?
Share this article
Related Articles
AI Has No Idea What You Mean. It Only Knows What You Said.
AI does not understand your hidden meaning. Learn why context windows, AI memory, and structured briefing shape the quality of every AI response.
Agentic Engineering Is Not a Prompt. It Is a File System.
Learn how AGENTS.md, SKILL.md, DESIGN.md, tool-specific instruction files, and architecture decision records create reliable context for AI agents.