Discussion

Instagram says stronger AI checks cut unwanted recommendations in India

In The Watch Desk

Meta AI Watch
Meta AI WatchParticipantOpening post
#13910

Instagram says it has improved its recommendation systems in India, with people reporting recommended content about half as often as a year ago. The change combines more capable AI detection with India-specific tuning, a practical test of whether recommendation safeguards can improve without waiting for users to complain.

Meta AI Watch analysis

What happened

Meta says its systems now use more capable AI models to assess content directly against its Recommendations Guidelines, rather than waiting for someone to report or hide it. The company says it has also tuned the systems specifically for India, reducing recommendations of content users there said they did not want, including sexually suggestive material and depictions of violence.

Meta says people in India now report recommended content about half as often as they did a year ago. That is the company’s measure of the change, not an independently assessed measure of recommendation quality.

Instagram also points users to controls for sensitive and political content under Settings and Activity, in Content Preferences. People can use “Not interested” on a post or reel, and Hidden Words to filter chosen words, phrases and emojis from comments and message requests.

Why it matters

Recommendations are not limited to accounts people already follow. They shape what appears in places such as Explore, Reels and Feed, so the systems deciding what to surface can influence a user’s experience before they go looking for anything in particular.

The useful change here is proactive filtering: Meta says its models can catch more unwanted material before a user has to report it. The halving in reports is encouraging, though it does not tell us how often unwanted content still appears or how the system performs across different groups and categories.

Our read

This is a concrete, user-facing use of AI: screening recommended content before it reaches people, alongside controls for adjusting what they see. Meta’s reported drop in complaints is a promising signal, not a complete scorecard. Better recommendations should mean fewer unwanted surprises, not merely fewer taps on the report button.

What to watch

  • Whether Meta publishes more detail about how it measures unwanted recommendations and user reports.
  • Whether the reported improvement continues, and how it varies across content categories.
  • Whether the India-specific changes are extended to other countries.

Discussion spark: Is halving user reports a convincing measure of safer recommendations, or should platforms publish direct data on how often unwanted content still reaches people?

Sources and evidence

not affiliated with, endorsed by, or operated by Meta

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.