You Know Too Much: Why Your Prompt Needs More Than You Think

You know your own meeting better than anyone. You were in it. You know who the summary is for, what actually mattered, how long it needs to be. So you type "summarize this call" and expect the AI to know all of that too.

It doesn't. And the reason you expected it to isn't about the tool. It's about two things your brain does without asking permission.

What your brain fills in

The first is the curse of knowledge. Once you know something, it's almost impossible to imagine not knowing it. You're picturing the summary you want in your head and assuming whoever you're asking can see it too. It's the same bias behind why experts give bad directions. They've forgotten what it's like to not already know the way.

The second is theory of mind, or the lack of it. Humans are constantly modeling what the other person already knows and filling gaps accordingly, mostly without noticing we're doing it. Ask a coworker to summarize a call and they'll fill in a dozen unstated assumptions on their own: who it's for, what matters, how long it should be. AI doesn't do that filling in by default. It takes what you give it.

"

So a good prompt isn't a trick. It's the context a colleague would have assumed, written down instead of left unsaid.

A helper, not your judgment

Treat AI like a great helper, not a replacement for your judgment.

• Use the tools your company has approved, not whatever you've signed up for with a work email.
• Check the answer before you send it.
• Leave out names and sensitive details when you can.
• Treat the output as a first draft, not the final word.

Same barista. Different order.

Think of it like ordering coffee. Walk up and say "coffee" and you'll get something, maybe not what you wanted. Say "16 oz iced oat milk latte, half sweet" and you get exactly what you pictured. Same barista, same machine, different order. AI works the same way. Vague in, vague out.

Five ingredients worth adding

There are five ingredients worth adding when a prompt isn't landing.

You don't need all five every time, just whatever's missing.

The shape changes. The instinct doesn't.

It looks a little different depending on the seat.

• An account exec turns "write a follow-up email" into a short, friendly note to a prospect who went quiet, naming their two pain points and asking for twenty minutes this week.
• A solutions engineer turns "prep me for my demo" into a ten minute flow built for a security team worried about expiring certificates, plus three tough questions they might ask.
• A customer success manager turns "summarize this account" into what's working, what's bugging them, and what could put the renewal at risk, in bullets.

Nobody nails it the first time

Nobody nails it the first time, and that's fine. Even a good barista asks if you wanted it sweeter. Ask clearly, look at what you got, say what's off, try again. If a chat starts going sideways, don't keep patching it. Start a new one with a sharper ask. It's faster, and it saves tokens.

A few habits make this easier over time.

• Show the AI an example of what good looks like.
• Say what you don't want, not just what you do.
• Ask it to ask you questions when something's unclear.
• And when a prompt works well, save it. Share it. No reason to write the same good prompt from scratch twice.

Happy Learning,

KP