Our Brandwatch-powered monitor tracks hate queries across X, Reddit, YouTube, Bluesky, Tumblr, Facebook, Instagram, the chan boards, and thousands of news sites and forums. This view shows how much is out there, which platforms carry it, and which coded signals are gaining ground.
Possible antisemitism content over the last 30 days. These numbers are based on keywords, and not human verified as hateful.
The terms and phrases with the most movement this week, compared with the week before.
Posts matched each day
How many posts matched our keyword searches each day. Pick a platform to see its posts on their own.
Coded signals over time
Daily volume of our configured lexical signals, rolled into the four families behind the platform Signal Risk Grades.
Where these numbers come from
Every day we collect public posts that match our antisemitism keyword searches, using Brandwatch, a social-listening service. It is the widest net we cast: everything that might be hateful, before anyone has checked it. It is where our work starts, not the finding. From this pool we bring posts into our Annotator, where trained reviewers label what is actually hateful, and we compare platforms on our Score Cards.
Loading the live window…
Quick answers to the questions people ask most about this page.
Researchers, policymakers, tech platforms, journalists and community groups: anyone who needs to understand online hate before they act on it, and who needs numbers they can trust.
Public posts that matched our keyword searches for this kind of hate, on the platforms listed on the page. A match means a post used words or phrases linked to hate. It does not mean a person has checked it and confirmed it is hateful. And it is not every post on the internet, only what our searches catch.
The grades show how often coded hateful language turned up on each platform in our search results, over the dates shown. They are not a rating of how safe a platform is overall, or of how well it removes harmful content.
The scores come from studies of AI models that AddressHate ran and published, plus, for antisemitism, a published detection test. Each study tested certain model versions in certain conditions. Before you compare two models, check which study and which version each score comes from. A model that did well in one test may not do as well on other content.
The numbers on this page update once a day. Score Cards built from our studies stay fixed to when the study was run, and any sample data is labelled as a sample. If a live source is down, we say so rather than show made-up numbers.
The Taxonomy page explains each category of hate we track. Each published report explains how it was done.