New report: Instagram algorithm serves antisemitic content to regular users

A newly published report alleges Instagram's recommendation algorithm surfaces antisemitic propaganda to ordinary users who never sought it out. The claim, circulated this week by Israel National News, points at Meta's recommendation engine — the system that decides which posts land in a user's main feed, Reels tab, and Explore grid — rather than at user search behavior.

What does the report actually claim?

The headline finding is that algorithmic distribution, not user intent, drives the exposure. Mainstream accounts with no apparent interest in extremist content are reportedly receiving antisemitic material through Instagram's automated surfaces. The implication matters: people aren't pulling this content toward themselves. The platform is pushing it their way.

The report frames this as a problem of amplification (boosting certain posts so more users see them) rather than curation (organizing content a user has chosen to follow). That distinction is technical, but it carries real weight in how regulators, advertisers, and Meta itself talk about responsibility for what shows up in feeds.

What does this change for practitioners?

For social media managers and media buyers, the finding — once corroborated — disrupts a basic working assumption. Brands typically treat Instagram as a relatively contained environment for paid placements and creator deals, layering brand-safety tools (software that flags or blocks ads from appearing next to harmful content) to filter risky adjacencies. An algorithm that volunteers extremist content to neutral users weakens that defense.

Marketers should pressure their verification partners — firms like DoubleVerify, Integral Ad Science, and HUMAN that score adjacency risk — on whether they screen algorithmic surfaces or only static posts. Many pre-bid filters (automated checks run before an ad auction to keep ads away from flagged content) operate at the publisher or individual-post level. They don't see the recommendation graph that put a user in front of a post in the first place.

Agency teams should also revisit category blocks (rosters of topics a brand refuses to appear alongside) and confirm "extremism" and "hate speech" are flagged at the keyword and topic-cluster level, not just at the surface-content level.

What remains unverified

The publicly available summary doesn't detail the report's authors, methodology, or sample size. Practitioners shouldn't treat the claim as settled until Meta or independent auditors publish corroborating data. Meta has historically argued that its recommendation system surfaces "engaging" content rather than "amplifies" harmful material — a distinction the company has defended in congressional testimony and oversight board cases.

The practical takeaway is procedural: brands running campaigns inside algorithm-driven feeds need to assume hostile content can reach neutral audiences, and price that risk into verification spend and crisis-response playbooks. The next test will come when Meta's own transparency disclosures — its quarterly content-moderation reports — land and outside researchers can benchmark them against the new report's data.