Microsoft Patents an AI System That Summarizes Meeting Content You Missed While Distracted

By Patentlyze Team · Updated May 16, 2026

We've all zoned out during a video call and quietly panicked about what we just missed. Microsoft is patenting a system that detects when that happens — and automatically sends you a summary of what slipped by.

What Microsoft's distraction-detection meeting AI actually does

Imagine you're on a Teams call and someone starts sharing a slide deck, but you're busy reading a Slack message on your other screen. According to this patent, Microsoft's system would notice that your attention drifted — and then automatically generate a text summary of whatever was being presented while you were looking away.

The system watches multiple streams happening in a meeting at once: the video feed, any shared screens, audio, and so on. It correlates where your attention is at any given moment against what content is being shown. If there's a mismatch — you're focused on one thing while something else is being presented — it flags that content as potentially missed.

A generative AI model (think the same kind of technology behind ChatGPT) then writes a quick summary of the missed content and delivers it to you, presumably as a quiet in-meeting notification. You stay in the loop without having to awkwardly ask someone to repeat themselves.

How attention signals trigger Microsoft's real-time summarizer

The patent describes a multi-stream monitoring system built around the concept of attention signals — data inputs that indicate where a meeting participant's focus is directed at any given moment. These signals could come from a variety of sources: gaze tracking via a webcam, mouse activity, active application windows, or other behavioral cues. The system uses these signals to infer that a user is engaged with one piece of content while a different piece of content is simultaneously being presented.

Here's the core loop the patent claims:

• The system continuously monitors multiple teleconferencing data streams (video, screen shares, audio, slides, etc.) in parallel.
• An attention signal is received indicating that a participant is focused on Stream A during a specific time window.
• The system identifies what was being shown in Stream B during that same window — content the user likely missed.
• That missed content is fed into a generative ML model (a large language model or multimodal model) with a prompt to summarize it.
• The generated summary is delivered to the distracted participant, presumably as a non-disruptive in-meeting notification.

The key technical innovation here is the time correlation step — matching the attention-absence window to specific segments of secondary content streams, then using generative AI to compress that content into something digestible on the fly.

What this means for the future of Teams meetings

If this ships in Microsoft Teams, it would mark a meaningful shift from AI as a post-meeting tool (like Copilot's existing meeting transcripts and summaries) to AI as a real-time co-pilot. Right now, you either pay attention or you rewatch the recording later. This system would blur that line — giving you a live safety net during the call itself.

There's also a broader signal here about attention tracking in productivity software. Building this kind of system requires monitoring user behavior at a fairly granular level — what you're looking at, when, and for how long. That raises real questions about how Microsoft would communicate that monitoring to participants, and whether enterprise IT admins would have controls over it. The patent doesn't address those guardrails, but they'll be the first thing enterprise customers ask about.

Editorial commentary on a publicly published patent application. Not legal advice.