One person spends the meeting looking directly into the camera. They nod at the right moments, speak without hesitation, and never seem distracted.
Another keeps glancing down. She checks a notebook, pauses before answering, and looks away while someone explains a complicated dependency. Near the end, she catches the contradiction nobody else noticed: the launch date assumes a security review that has not been scheduled.
Who was more present?
Video meetings have always encouraged us to answer that question with visual shortcuts. Eye contact looks like attention. A smooth voice sounds like confidence. A camera-ready posture reads as respect. Those cues were never especially reliable, and meeting software is beginning to make them even less so.
That is not necessarily bad. Technology can reduce barriers, help people communicate across accents, and make remote conversations feel less awkward. But teams need to stop treating a polished signal as evidence of thoughtful participation.
Meeting presence is becoming editable
At Zoomtopia in October, Zoom announced several forthcoming meeting features that illustrate the shift. Pose Correction is designed to make a participant appear camera-forward while they glance away. Touch up my Voice is intended to smooth how an accent comes across in real time. Zoom says both are expected in December 2026, while a Live Meeting AI Agent that can nudge engagement, pacing, and clarity is expected in early 2027. These are announced capabilities, not features every customer can use today. Zoom's announcement
Each feature addresses a recognizable problem. Looking at notes can be mistaken for disengagement. A listener may struggle with an unfamiliar accent even when the speaker's argument is precise. A solo consultant may benefit from a private prompt to slow down.
The mistake would be to conclude that the improved surface reveals a better meeting underneath. A corrected gaze can make attention easier to perceive, but it cannot establish that someone understood the discussion. A smoother voice can reduce listening effort, but it cannot make an unsupported claim more credible.
This matters most wherever impressions carry consequences: interviews, performance conversations, sales calls, executive reviews, and hybrid meetings where some people already have more visibility than others.
Microsoft Research's hybrid-meeting guidance notes that remote participants face imbalances in how they are perceived, their access to conversational cues, and the distribution of power. It also says that it can be difficult to know how engaged a remote person is. Microsoft Research
AI-mediated presence does not create that ambiguity. It makes the ambiguity impossible to ignore.
Separate transmission quality from contribution quality
A clear microphone, readable captions, stable framing, translation, and voice processing can all improve the transmission of a person's contribution. They help an idea reach the room.
Contribution quality is a different question. Did the person understand the decision? Did they add relevant evidence? Did they notice a conflict? Can they explain what happens next?
Managers often collapse these two categories because transmission signals are immediate. A confident response feels easier to evaluate than a careful question. Direct camera gaze is visible throughout the call, while the value of someone checking a source may appear only once.
Use meeting technology to improve access and clarity. Judge participation from what enters the shared work.
Replace presence theater with four kinds of evidence
Do not solve this problem by inventing a new engagement score. Counting words, camera time, nods, chat messages, or eye movements simply replaces one shallow proxy with several.
Instead, design the meeting so useful participation leaves ordinary evidence. Four signals cover most collaborative work.
Understanding: Can participants restate the current question, decision, or constraint in their own words? This need not become a classroom exercise. Before a consequential choice, ask one person to summarize the proposal and another to name its main condition.
Friction: Did anyone introduce evidence that changed the conversation? A useful objection, source conflict, boundary, or counterexample often demonstrates more attention than ten minutes of visible agreement.
Movement: Did the meeting become more specific because of the contribution? Good participation might narrow the scope, clarify a customer need, identify an owner, or turn a vague risk into a testable question.
Handoff: Can the people responsible for the next step state what they accepted, what they need, and when the group will check again? A person who understands the handoff can usually describe the deliverable and its conditions without reading a polished recap word for word.
These signals are not a performance rubric. They are properties of the work. They help the group see whether shared understanding exists without asking anyone to perform attentiveness for a camera.
Put three checkpoints into the conversation
The easiest way to gather better evidence is to create moments where understanding can surface.
At the beginning, state what kind of participation the meeting needs. A decision review may need challenges and tradeoffs. A briefing may need clarifying questions. A working session may need people to edit the same artifact. "Be engaged" tells participants nothing. "We need to leave with one launch option and the unresolved risk beside it" gives them a job.
In the middle, pause at the point of highest uncertainty. Ask: "What assumption are we relying on here?" or "What would make this plan fail?" Give people a quiet minute to inspect notes, search a source, or write before answering. Looking away from the camera is often exactly what serious thinking requires.
At the end, confirm the shared record. Read back the decision, surviving concern, owner, and next checkpoint. Invite correction from anyone, including people who contributed through chat or need a little more processing time.
Three checkpoints are usually enough. If every comment must prove engagement, people will optimize for visible activity rather than good judgment.
Allow more than one credible way to participate
Some people think while speaking. Others need to write, inspect an artifact, or wait until two ideas connect. A participant may use captions, turn off video to preserve bandwidth, look toward a second screen, or type a precise correction instead of interrupting.
Set a few legitimate participation routes before the meeting:
- speak in the main conversation;
- add a question or source in chat;
- edit the shared artifact;
- respond to a named checkpoint after a short reflection period.
The organizer still has to notice those routes. If nobody monitors chat or brings written changes back into the discussion, offering alternatives is merely decorative.
For interviews and performance discussions, take extra care. Do not infer curiosity, confidence, honesty, or leadership from gaze and vocal polish. Ask consistent questions, record job-relevant evidence, and distinguish the content of an answer from the ease with which it arrived.
Keep coaching private and evaluation human
Real-time AI coaching may help a speaker notice that they are rushing or dominating the airtime. Treat those prompts as private aids, not measurements of character.
An engagement nudge should never become evidence that an employee was inattentive. The system may not know that the person is reading captions, checking the document under discussion, handling an accessibility need, or pausing to avoid an impulsive answer. Nor should a manager assume that someone who received no nudge was fully involved.
The same restraint applies to meeting analytics. Use them to ask better questions about the format: Did one person occupy most of the discussion? Did remote attendees have a route into the room? Did the meeting end with a confirmed decision? Do not use behavioral traces to manufacture certainty about an individual's internal state.
Design for proof of work, not proof of gaze
The more software improves the appearance of presence, the less useful appearance becomes as a management shortcut. That is an invitation to build fairer meetings.
Let people look at notes. Let technology make speech easier to hear. Then anchor the conversation in evidence: the question someone sharpened, the source they checked, the risk they surfaced, and the commitment they accurately carried forward.
Caspi supports that kind of meeting with real-time recap, suggested questions, contextual chat, proactive flags from connected tools, post-call action items, and persistent meeting memory. The useful role of a meeting copilot is not to certify who looked attentive. It is to help the team preserve what was understood, challenged, decided, and handed off.
A meeting does not need everyone to look present in the same way. It needs a reliable way to tell whether the work moved because they were there.