On a newly created Instagram account, the first result for “synagogue” was an AI-generated video of a rabbi welcoming the Antichrist, the Nexus Project reports. It had drawn more than a million views. Researchers say that after they spent less than an hour watching antisemitic Reels, antisemitic posts occupied almost the whole homepage.
An antisemitism research group, the Nexus Project, details the findings in a new report arguing that Instagram’s recommendation system actively spreads anti-Jewish conspiracies and propaganda. The testing highlights two separate issues: searches using ordinary religious terms can surface hateful posts, and watching that material leads to recommendations for more.
To explore Instagram’s responses to searches about Jewish people, researchers tested several fresh accounts. Their exploratory work did not establish how frequently users across Instagram encounter these results. The first search result also warrants attention apart from the subsequent recommendations, since it appeared before the account had accumulated a long viewing history.
Much of what researchers encountered was generated with AI, according to the Nexus Project. The group argues that AI allows people to quickly and easily create images and videos portraying Jewish people as sinister. With less moderation, it says, this volume of material can spread alongside anti-Jewish conspiracies and other hateful posts.
“The scale and potential impact of this shift cannot be overstated,” the report says. “While social media companies like Meta were always lax in their enforcement of hate content, recent changes have transformed the landscape.”
Verified accounts supplied more than half of the antisemitic material on the account whose feed shifted within an hour. Describing what happened, the report says: “After less than an hour of watching a series of antisemitic Reels,” it says, “nearly our entire feed was devoted to antisemitic content with messages like ‘Jews started LGBTQ’ and a video telling viewers, ‘It’s time to wake up GOYIM.’” The word goyim refers to non-Jewish people and has also gained an association with antisemitic messaging.
The researchers found recommendations outside the feed, too. Instagram placed suggestions to explore “goyim brainwashing” and “Dancing Israelis,” both conspiracy topics, beneath antisemitic posts. Its “trending” suggestions also included “Erika,” a Nazi marching song used in thousands of Reels, the report states.
The search findings also prompted questions about the financial interests involved. Tests conducted separately and documented in the source article appeared to corroborate the Nexus Project’s findings. Searching “synagogue” through a decoy Instagram account brought up an AI-generated video among the earliest results: a large, satanic-looking figure stood inside a place of worship, covered in blood as Jewish men celebrated it.
For “Jew,” the results included numerous posts showing Jewish people as cartoonish distortions reminiscent of Nazi propaganda. The accounts posting them focused on hateful and conspiratorial material. Entering “Jews” in the search field produced Instagram’s suggested search, “jews controll [sic] everything.”
Looking up “Erika” produced extensive antisemitic results. Among them were AI-generated videos showing students and fans giving the Sieg Heil salute at a high school football game while a version of the Nazi anthem played. Links for downloading “Erika” ringtones appeared in the bios of some accounts promoting the song.
Combined, the antisemitic videos surfaced by those searches had millions of views. The source article’s reporters revisited that account a day later and saw still more hateful posts filling its feed.
Synthetic images are not the only method involved. According to the Nexus Project, many viral posts recut podcast excerpts and other footage so that Jewish people seem to have made or endorsed statements they never did. One example is the conspiracy claim that the Jewish community secretly controls the world.
The researchers also describe a financial incentive: creators and accounts market supplements, e-books, courses and subscriptions using antisemitic posts and conspiracy theories. Through advertising and paid verification checkmarks, the report argues, Meta could profit from this material as well.
Meta challenges the report. In the Nexus Project’s account, January 2025 marked a turning point for the spread of antisemitic content, when Meta scaled back moderation. The company dropped its partnership with third-party moderators in favor of a user-driven approach, describing the move as an effort to deliver “more speech and fewer mistakes”. Meta argued that moderation had gone “too far,” censoring excessive amounts of “harmless content.”
The source article argues that the change reduced Meta’s direct role in platform safety and left users with much of the responsibility for spotting and limiting misinformation, hate speech and abuse material. It also references earlier reporting about abusive posts that Meta did not remove after users reported them, including AI-generated images of coaches inappropriately touching young athletes and violent child abuse generated with AI.
Rejecting how the report portrays its approach, Meta said its “commitment and dedication to tackling antisemitism is unchanged.”
“We have not removed any antisemitism-specific protections and we continue to prohibit harmful stereotypes about Jewish people, Holocaust denial, white supremacist and dangerous organizations, and the use of the term Zionist as a proxy for Jewish or Israeli people in hate speech,” Meta said. The company cited its publicly available data on hateful-conduct enforcement and reports on platform integrity.
“This report contains both factual inaccuracies and misrepresentations of our standard that this type of violent and hateful material has no place on our platforms. In fact, since 2025 the prevalence of hateful conduct violations on our platforms, which we regularly report publicly, has remained consistent. Our policies prohibit content that attacks people based on their race, ethnicity, religion or any other protected characteristics, and we’ll continue taking action against any violating content we identify.”
These tests neither measure exposure across the platform nor demonstrate that the moderation rollback caused what researchers found. They do, however, leave a concrete question about enforcement and distribution: despite Meta’s stated policies, why did an ordinary search immediately return antisemitic content, with more of it filling the feed after researchers watched?



