Normsy
Reducing toxicity on social media without censorship: AI that proposes constructive replies, with people deciding.
An AI tool to reduce violence and toxicity on social media, built with Civic Health Project. It evolved from what was originally called Social Media Detoxifier.
1,021
Study participants
- Anthem Award 2024 — Best Use of AI
- Built with Civic Health Project
- With participation from NYU's Center for Social Media and Politics
The problem
The usual responses to toxicity online are to moderate, remove, or suspend. All of them act after the harm, and none change the tone of the conversation: they take content away without adding anything.
This project started from a different premise. The problem is not only detecting harmful content, but that constructive voices are usually absent from those conversations. If no one replies with judgment, all that remains is the toxic material.
What we built
The system identifies relevant conversations and uses language models to propose counterspeech: alternative replies that add context or defuse aggression, instead of removing content.
The decision is never automatic. Proposed replies go through human collaborators, who bring the context and judgment the model lacks. So the approach combines automatic classification and LLM generation, but keeps a person in the loop before anything is published.
From Social Media Detoxifier to Normsy
The project began as Social Media Detoxifier, with Civic Health Project, and under that name won the 2024 Anthem Award for Best Use of AI.
It later evolved into Normsy, extending the architecture with agentic memory, topic-based routing, and orchestration. It is a public product today.