How to let a speaker read the room without watching anyone

The mistake everyone remembers

In 2020 Zoom shipped per-participant attention tracking, and within months withdrew it under sustained criticism. That was the correct outcome. A badge reading "Priya is not paying attention" is wrong about Priya roughly as often as it is right, and the moment such a badge exists people start managing the badge instead of the meeting.

The episode left the whole category with a reflex: attention is radioactive, do not touch it. Which is a shame, because the underlying problem it was trying to solve is completely real.

The failure was attribution, not measurement

A speaker in a physical room glances up and knows within a second whether they still have the room. On a video call that sense is simply amputated. You talk into a wall of motionless rectangles for forty minutes with no idea whether anyone is still with you. For trainers, educators and anyone running a town hall, that is not a minor inconvenience — it is the entire job made impossible.

Bridge's position is that the thing people objected to was never the signal. It was naming individuals with it. Aggregate the same signal, strip the identity, and it answers the one question a speaker most needs and can least perceive — while telling them nothing they could ever use against a particular person.

What the speaker actually sees

While you are speaking — and only while you are speaking — you see a count. "Nine of twelve have Bridge in view." Never who. Never a name, never a ranking, never a per-person score.

That is the whole feature. It is closer to looking up at a lecture hall than to reading a report.

The guarantees, stated precisely

These are worth having in front of you in a security review, because they are enforced in code and covered by automated tests rather than promised in a policy document.

• The output is a count and a ratio. No per-person attention value exists anywhere in the system — not hidden, not private, not reachable by a determined caller.
• Below four contributing people it reports nothing at all. At three, a "2 of 3" reading would identify the third person to anyone who can see the other two.
• It is opt-in and off by default, per person.
• Anyone sharing can see at any time exactly what they are contributing and why, in the same words the room is counted in.
• It is never stored — and that is guaranteed by architecture, not policy. The readings ride the room's existing peer-to-peer relay and never reach a database, a log or an endpoint. There is no table they could accumulate in.
• A reading expires to "unknown", never to "away". Someone on a flaky connection is not silently reported as absent.
• Opting out is silence, not a recorded refusal. There is no "declined" marker for anyone to count.

Why "never stored" is the claim that matters

Most vendors, asked about sensitive telemetry, can only promise not to look at data they are already holding. That promise is exactly as durable as the current management, the current policy and the current subpoena.

Because Bridge's attention readings never leave the peer-to-peer relay, there is no retained data to govern. That is a structurally stronger position than a good privacy policy, and it is the difference between a feature that passes a works-council review and one that does not.

If your instinct is still "we are not comfortable with this"

That is the right instinct and we would rather you kept it. The objection is to surveillance of individuals — which Bridge does not do and structurally cannot. It is off unless somebody turns it on, it is blind below four people, nobody is ever named, and nothing is ever written down.

Frequently asked