You remember the summary, not the meeting

Oct 2

Six people on the call. You said the launch date was too tight. Two others agreed. Nothing was decided. The next morning an AI summary is in your inbox. Five tidy bullets. One says the launch date is confirmed. A week later someone asks what was decided. You answer with the bullets.

Researchers at Georgetown University wanted to know if a summary can change what people remember. They showed 331 people a short animated film where a car hits a pedestrian. A day or two later, everyone read a summary.

With an accurate one, people got 84 percent of the details right. With a misleading one, 45 percent. Telling them an AI wrote it made no difference. I use AI most of my working day. I still think that last finding should worry us.

Hippo files the tidy version

Your brain has an archivist. I usually call him Hippo. He stores what happens to you and files it so you can find it again.

His file from that call is a messy one. Half sentences. A tone of voice. Who went silent when the date came up.

A memory isn’t locked once it’s filed. Every time you take it out, Hippo files it again. Whatever is in front of you at that moment can end up in the file too.

So you open the summary. Hippo takes out the file from the call to compare. The summary is short and clear. His own notes are a mess. Hippo likes order. The tidy version goes in. After a while he can’t tell which lines were his.

Corty did read the label. She is the thinking part behind your forehead. She knows a machine wrote those bullets. But Hippo doesn’t file the label. He files what the bullets say. I think that’s why the warning didn’t help.

Six memories become one
You weren’t the only one who read it. All six of you did.

Before the summary there were six slightly different memories of the same call. That was useful. One person remembered the objection. Another remembered who promised what. When you compared them, you got close to what happened.

Now there are six copies of one version. And nobody checked what went into it. The researchers also let ChatGPT and Gemini summarise the films. Almost every summary left out the collision. The study used a film, not a meeting. I still recognise the pattern.

In my book I write about what false information does to the brain. A claim we know is wrong keeps shaping what we remember. We can’t just delete it. The brain needs information it can trust. Without it we start to doubt everything. That doubt takes capacity from the work.

Leaving things out is enough. What isn’t in the bullets slowly stops having happened.

Write your own three lines first
Before you open the next summary, write three lines of your own. What was decided. What was left open. Who disagreed. It takes two minutes. Now your version exists outside your head, where a summary can’t tidy it.

Then open the summary and compare. Whatever is in your lines and missing from the bullets is what the team will forget.

So put it back. Reply to everyone who got the summary with one or two lines. “The launch date is still open. Three of us had doubts.” Now the correction is in the same place as the bullets. A week later, ask someone what was decided. If they mention the open date, your lines went in.

I’m not going to stop using AI. I don’t think you should either. But the tidy version is the one that stays. Think of the last decision your team made. Do you remember the call, or the bullets? And who read the bullets before they became the memory?

/ Maria