In 1966, an MIT computer scientist named Joseph Weizenbaum built a chatbot called ELIZA. It was simple. It mostly turned your own words back into questions. Say "I'm worried about work" and it might reply, "Why are you worried about work?"

People loved it. Some of them confided in it. Weizenbaum's own secretary, who knew exactly how it worked, reportedly asked him to leave the room so she could talk to it in private. That instinct to treat a machine as if it understands us has a name now: the ELIZA effect.

Sixty years later, the machines really are far more capable. The instinct is the same, and I think it's the most common reason good prompts produce bad results.

What the habit looks like

When we talk to a colleague, we leave most things unsaid. They know the company, the client, last week's meeting and what "the usual format" means. We speak in shorthand because they share our context.

AI doesn't. But because it answers so fluently, we slip into talking to it the same way. We write a tidy prompt, get back something confident and generic, and decide the tool isn't very good. The prompt wasn't the problem. The missing context was.

The mindset shift

I tell colleagues to picture a brilliant new hire on their first morning. They are fast, well read and eager. They have never heard of your company, your customers or your standards. Anything you don't tell them, they will guess.

That one change in how you picture the tool fixes more than any prompt template I've seen. You stop asking, "What's the magic wording?" and start asking, "What would a smart stranger need to know to do this well?"

What to hand over

  • Who it's for. The reader, what they already know and what they care about.
  • What good looks like. An example you liked, or the house style you follow.
  • The facts it can use. Paste them in. Don't make it guess numbers or names.
  • What to avoid. Claims you can't support, words your brand never uses, topics that are off limits.
  • What happens next. Where the output goes and who checks it.

Build it once, reuse it forever

Typing all that context every time is tiring, which is why people skip it. The fix is to write it down once. Most AI tools now let you save instructions and reference files in a project or agent. I keep a style guide, a fact file and a set of examples in each project I use, so every new request starts with the context already loaded.

The output got better overnight. Not because I found better words, but because I stopped expecting the tool to read my mind.

I made a two-minute video on this idea. It's the featured video on my home page.