When Every Word Becomes Training Data, People Start Editing Themselves

The Economist's Bartleby column has put its finger on something most leaders have not thought about yet. In You're not sending an email. You're training a model, the column explains how AI note-takers and recorded workplace conversations are becoming raw material for AI models, not just tools for productivity.

The example is striking. According to the column, Google bid $10 million for the data of Spirit Airlines, the bankrupt carrier, including 100 million emails and 500 million Teams messages. The deal is being challenged in court, including by the union representing Spirit's flight attendants. The column contrasts American practice with European law, where data collected for one purpose generally cannot be reused for another.

Then it closes with a prediction. If employees assume their words will train smarter models, they will become more guarded, and the workplace will become "more professional, and more dull."

I think that last line is not a footnote. It is the story.

Guardedness Is the Real Cost

Most coverage of workplace data focuses on privacy and legal risk. Both matter. But the deeper cost is behavioral. When people know that every word may be recorded, analyzed and reused indefinitely, they change what they say.

They stop thinking out loud. They hold back the half-formed idea that might have become the breakthrough. They avoid the candid comment that might have saved a project. They choose safe language over honest language. The meeting still happens, the transcript still gets generated, but the most valuable part of the conversation never takes place.

In my work, I call this the edit: the parts of themselves people learn to leave out in order to fit the room. Our research at GLLG found that 87% of managers report their individuality is suppressed at work. That was before every meeting became a permanent record. Recording does not create the edit, but it can scale it dramatically.

There is an irony here. Organizations are collecting more of their employees' words than ever, hoping to capture knowledge. If people respond by editing themselves, the organization ends up with more data about less of its people. It trains its AI on a cautious, self-censored version of the workforce.

Standards That Protect Candor

None of this means leaders should stop using AI tools. Meeting summaries, action items and searchable records are genuinely useful. Standardization is an asset, and clear rules about data are exactly the kind of standard organizations need.

The problem is that most organizations have not set those rules out loud. According to Littler's 2026 survey of more than 600 legal, HR and business leaders, only 17% say their organizations communicate proactively and consistently with employees concerned about AI's consequences. When people do not know the rules, they assume the worst, and they edit accordingly.

In my work on Leadership in the Age of Personalization, one of the five indicator questions is: Who do you let them be? Data policy now answers that question as much as culture does. A team that knows which conversations are recorded, why, and how the information will be used can choose to speak freely. A team left guessing will default to caution.

Standards and personalization are not opposites here. Clear standards create the safety that lets people be themselves.

Protect the Room Where Truth Gets Told

As I wrote in Why People Wait Until They Leave to Tell the Truth, the information required to make the right decisions almost always exists inside the building. The conviction to put that information on the table is what is missing. Conviction needs a room where it is safe to say what you believe before the outcome is certain.

If every room is recorded, leaders have to create some that are not. That means deliberate spaces, from one-on-ones to brainstorming sessions to candid debriefs, where people know they are speaking to colleagues, not to a dataset.

As I explored in Stop Reading Personality Types. Read Levels of Suppression., what looks like a personality trait is often a response to the environment. A quieter, more guarded workforce will not be a sign that people changed. It will be a sign that the room did.

Three Questions for Leaders Deploying AI Note-Takers

Do our people know what we record, why we record it and how it will be used? If the answer requires a policy document no one has read, the answer is no.

Where can people think out loud without a transcript? Name those spaces and protect them.

Are we measuring what we lose? Watch whether meetings become shorter, safer and less creative after AI tools arrive. Silence is data too.

The Economist is right that AI may make the workplace more professional and more dull. Leaders can still prevent the second part. The goal is not to capture every word. It is to keep the conversations worth capturing.

Want to explore these ideas further? Learn more about my work on leadership, identity and conviction at www.theglennllopis.com.