It was just another online meeting. A familiar grid of faces, a shared document, the usual start-of-meeting chatter. Then, without fanfare, a new participant appeared in the chat log: [Amy’s AI Assistant] has joined the meeting.

No one introduced it.
No one explained what it was doing.
No one asked if we were okay with it being there.
And Amy wasn’t here.

Some participants didn’t even notice. Others glanced at the notification, paused, and then carried on. But for me, something shifted. I felt a tightness in my chest. My contribution, once at the tip of my tongue, suddenly felt risky. I didn’t know what this bot was recording. I didn’t know who invited it. And I wasn’t sure I wanted my voice, my words, or even my tone of voice added to some AI training dataset, or misquoted in a summary I’d never see.

I hesitated. And in that hesitation, something human was lost.

The cost of silent convenience

We’re getting used to AI tools quietly folding themselves into our workflows. Meeting transcription bots, summarisation assistants, AI note-takers. They promise smoother collaboration, less cognitive load, and more efficient documentation.

But when those tools enter shared spaces without context or consent, they don’t just support the conversation. They reshape it. What happened in that meeting wasn’t just a technical glitch or oversight. It was a micro-example of something bigger:

A culture that prioritises productivity for some, over the participation for all.

Here’s what that moment revealed to me:

• Lack of consent suppresses contribution. I wasn’t the only one who paused. I could see it in the hesitation of others. AI without warning creates uncertainty, and uncertainty breeds silence.
• Ambiguity about purpose leads to distrust. Was it recording audio? Is it screen capturing? Transcribing text? Who could access the files? When we don’t know what a tool is doing, we default to protecting ourselves — by retreating.
• Bias in transcription and voice tech excludes people. Colleagues with diverse accents are often misrepresented or dropped entirely by AI transcription tools. And when meeting summaries become the “official record,” those inaccuracies quietly erase contributions.

All of this adds up to a chilling effect on the very thing meetings are meant to cultivate: collaborative, creative, inclusive dialogue.

What happens when participation feels risky?

For educators, researchers, and design professionals alike, this isn’t just a UX quirk — it’s a profound shift in how shared spaces are governed. The presence of AI tools in meetings (1) Introduces power asymmetries. Some people know it’s there, others don’t. Some have access to the output, others don’t. (2) It changes how people self-censor, particularly those already navigating marginalisation in tech or academia. (3) It risks turning real-time collaboration into data extraction, where people become sources, not co-creators.

If you’re a student, do you question a tool your lecturer enables?
If you’re a researcher, are you confident your voice will be transcribed fairly?
If you’re a professional, how often do you get to opt out?

In moments like these, our digital spaces become less about co-presence and more about quiet surveillance. And unless we actively resist that drift, we’ll design a future where being recorded is default, and being heard is optional.

Toward consented, inclusive AI presence

So where do we go from here? This isn’t a call to throw out every AI assistant. But it is a call to design differently, and more humanely. Here are some questions  I’ve been thinking through, inspired by work in

Visibility and friction are features, not bugs

We often design AI tools to be seamless, invisible, ambient. But when something is recording, summarising, or processing human input, its presence should be visible.

Design idea: Give AI a clear identity in meetings — a name, a visible avatar, a “this is what I’m doing” message pinned to the screen. Let its presence be as disruptive as its function.

Yes, it breaks the illusion of smooth collaboration. But it also breaks the illusion that nothing is at stake.

Consent is an ongoing process

One-time consent buried in a settings menu isn’t enough. Real consent means people can opt in, opt out, or change their minds. And it should happen in context.

Ritual idea: Begin meetings with an AI roll call. “The AI assistant is here to transcribe this session. Here’s what it captures, who sees it, and where it goes. Is everyone okay with that?”

This isn’t just procedural. It builds shared awareness and trust.

Transcription bias needs active mitigation

Voice tech is still deeply flawed when it comes to understanding diverse speech patterns, accents, and dialects. And yet we’re leaning on it more than ever especially in international, multilingual, and hybrid teams.

Intervention idea: Tools should auto-flag low-confidence transcriptions and let users annotate or correct.

And meeting leads should treat transcripts as first drafts, not gospel.

Participation is a design outcome

If people don’t feel safe to contribute, that’s a failure of design.

We must treat psychological safety and equitable presence as core design goals, especially when tech enters social spaces.

This goes double for education, where the stakes of participation are often felt most acutely.

Final thought

What AI tool designers may think of as a helpful assistant, participants might experience as an ambient threat. Especially when the terms are unclear, the consent is assumed, and the system doesn’t see or hear everyone equally. As AI becomes part of our collaborative rituals, we don’t just need better tech. We need better practices, better questions, and better norms.

Over to you

Have you had an unexpected AI moment in a shared space? What do you think good “AI meeting etiquette” should look like?