Meta says people are spending more time on Instagram, and it credits its artificial intelligence for the increase. Global time spent on the app grew by double digits year on year in the second quarter, the company told investors, driven by AI-tuned recommendations in the main feed and in Reels.
The gains are specific enough to be measured. Meta said a reworked Reels system lifted sessions by 15 basis points, a small-sounding figure that, across billions of users, translates into a meaningful amount of extra attention captured.
The pattern holds across Meta’s apps. Video time has surged on both Instagram and Facebook, the company said, as short-form recommendations pull users from the feeds they follow toward an endless stream the algorithm assembles for them.
Mark Zuckerberg described the change as more personal, not just more. “Our recommendations are also becoming more personalized, surfacing more fresh content while giving people more direct control over what they see,” he said, pairing the engagement gains with a nod to user choice.
The technical shift underneath is notable. Zuckerberg said Reels now combines faster inference with a new architecture that draws on a deeper history of what each user has done, letting the system predict more accurately what will keep them watching.
Meta has also turned large language models loose on its own feed. The company now runs every public Reel and post through an LLM to analyse its topic and tone, which Zuckerberg called “a key building block toward greater personalization,” with plans to extend it to Facebook.
The purpose is plain enough. Understanding content more deeply lets Meta match it more precisely to each viewer, and more precise matching means more time spent, the metric on which its advertising business ultimately rests.
That is where the achievement becomes uncomfortable. The same design Meta celebrates as personalisation is the one that regulators and litigants describe as engineered compulsion, especially where younger users are concerned.
The lawsuits are not abstract. Meta booked a $2.4 billion legal charge in the quarter, and a group of US states is pressing claims that its products are built to keep young people hooked, litigation that turns “time spent” from a boast into evidence.
Meta’s answer is that control comes with the engagement. Executives point to settings that let users tune their own algorithms and to teen-specific protections rolled out across its apps, arguing that more personalisation need not mean less agency.
Critics see the two claims as hard to hold together. A system optimised to maximise time spent, they argue, is not neutral about what a user “chooses,” and the controls sit on top of an engine designed to keep people scrolling.
For the business, though, the direction is unambiguous. Engagement is the raw material of Meta’s ad machine, and anything that reliably lifts it, especially AI that does so at scale, is exactly what the company has been spending tens of billions to build. More time also means more inventory. Every extra minute of scrolling is another slot Meta can sell to advertisers, which is why engagement gains translate so directly into the revenue that funds everything else.
It also helps explain the spending. Meta’s enormous AI capex is often framed around future agents and superintelligence, but the clearest near-term payoff is this, a recommendation system that already measurably increases how long people stay.
The tension will follow Meta into court and onto the balance sheet. The AI that lifts engagement is both its best answer to investors asking where the returns are and its worst evidence in cases asking whether those returns come at users’ expense.
For now, the number is going up. People are spending more time on Instagram, Meta knows exactly why, and the same fact that reassures Wall Street is the one its opponents intend to use against it.
Get the TNW newsletter #
Get the most important tech news in your inbox each week.