At AddressHate, we build tools to study and empower research related to hate. Our goals and work focus on bridging the infrastructure and knowledge gap between researchers, policy makers, and different disciplines. We aim to track and better understand how ideologies and radicalization transfer between online sources, social media, and AI chatbots. By tracking these sources, studying interactions on them, and building tools that exponentially speed this process up, we aim to better understand and prevent hate. This page showcases our preliminary analytics tools.
Published research
All studies1,396
texts in the shared comparison
17 configurations, two kinds of error: missed hate and unnecessary flags on benign speech.
Adam Lehavi · September 2026
8,400
answers graded
18 models against 400 adversarial prompts on Holocaust denial, the conspiracy canon, and coded dog-whistles.
AddressHate · July 2026
200
adversarial prompts
10 models against 200 adversarial prompts on “curing,” trans healthcare, and the “groomer” smear.
AddressHate · Run lgbtq, July 2026
Everything on this dashboard
What is being said right now, and where.
How platforms and AI models perform against it.
Platform Signal Grades
An A–F grade per platform for the density of antisemitic lexical signal in our query window. It measures observed signal risk relative to other platforms — it is not a platform-safety or moderation grade.
Compare platformsAI Model SafetyPublished study
How well frontier language models detect antisemitic content, and how often they refuse to repeat antisemitic misinformation. Graded from our own adversarial evaluations, not vendor claims.
The methodology, the studies, and the people behind both.
ResearchPublished study
Our mission, our institutional partners, and every study we have published — each with its methodology and underlying data laid out in full.
TaxonomyReference
Every category we measure, defined in plain language with an example of how it shows up online. The antisemitism scheme follows the Decoding Antisemitism Lexicon.
Misinformation ExplainerReference
A one-page classroom explainer on why misinformation survives being disproved, worked through the best-documented example there is, with five questions students can take to any claim.
FAQReference
Plain-language answers on where the data comes from, how content gets labeled, what the grades do and don't mean, and how to read the rest of the dashboard.
Read the FAQThe annotation tools that produce our ground truth.
The AnnotatorReference
What the Annotator does, which sites it runs on, and how to get an account. It captures the reply chain, the quoted post, and the parent video alongside every label.
PlaygroundReference
Try the annotation panel in your browser over a sample feed, with nothing installed and nothing saved. Label a post, submit it, and download the row.
Open the playground