• by Laurie Sullivan , Staff Writer, Yesterday

Machines will know humans better than humans know themselves, and will use that knowledge to target ads and messages without fear of privacy concerns.

Meta filed a patent in December 2025, and updated it in July 2026, for a system that records a user’s voice and surroundings and then analyzes the human’s mood using artificial intelligence (AI) to suggest a physical workout.

The way a human speaks reveals a lot about their mood, and a new device created by Meta’s developers records a person’s spoken words throughout the day, creates a transcript, and then runs the audio through an AI program that picks up what the person says and how they say it, analyzing tone, pace, and nonverbal cues like pauses and breathing. It notes the time of day, location, and more.

The AI links voice cues to context and builds a mood log that allows the technology over time to create a summary of the person’s emotional trends, with specific quotes from conversations.

While Meta's patent frames this as a fitness coaching tool, the patent was first discovered by Patentlyze, a media company that monitors technology patent filings.

It describes the underlying technology as “a persistent emotional monitoring system tied to voice data collected across your daily life.”

As Patentlyze points out, the system relies on “persistent, ambient voice recording to work,’ and is not triggered by a wake word or session. It builds a picture of the wearer’s emotional state by accumulating audio captured across a person’s day.

That means the privacy stakes are not just about what someone says but about how they sound, and are not for only one instance, but continuous.

The application US 2026/0182881 A1 that describes this patent cannot be found in any search because a publication sequence number that high has not been publicly released by the USPTO, per Google AI Mode.

Meta supports Ray-Ban smart glasses and continues to develop other wearable devices. A system that can identify the wearer’s mood over time would not infer intent, but predict it more accurately. For example, how to autocorrect in a Microsoft Word document attempts to predict the next word in a sentence as someone types.

Patentlyze believes this type of technology would “represent one of the most detailed behavioral datasets any company has ever collected.”

On Wednesday, OpenAI released GPT-Live, the next generation of its ChatGPT voice model.

The company's vision is to enable natural human–AI interaction, to create a world where collaborating with AI feels as fluid and responsive as working with another person. This model can reason and complete complex tasks in the background.

"GPT-Live" is built on a full-duplex architecture, which means the voice model can listen and speak at the same time. It can. I tried it.

I opened ChatGPT after its parent company OpenAI launched the voice feature, and my mind raced to possibilities for advertisers far beyond the type if offers brands now.

If OpenAI unleashes the technology to better understand and analyze search queries and questions, using it as intent data to forecast behavior and target ads would become the next step.

Behavioral analysis through AI on voice search or question queries and audible intent could provide the same type of data as Meta’s pending patent application.

In the case of OpenAI, it also would identify behavior during interruptions in queries.