How We’re Keeping Reddit Real and Safe in the AI Era

menu

This is a search field with an auto-suggest feature attached.

There are no suggestions because the search field is empty.

How We’re Keeping Reddit Real and Safe in the AI Era

Announcements

July 06, 2026

TLDR: We’ve upgraded our automated defenses to catch violating behavior before our communities ever see it. By deploying advanced AI tools, we’ve reduced user exposure to spam by 20%, revoked nearly 2M fake votes daily, and slashed our enforcement time on hateful or violent content from hours to under five seconds.

Every day, millions of people use Reddit to connect in ordinary, often meaningful ways. It’s those connections and conversations that make Reddit the most human place on the Internet. And it’s why, in the age of AI, spam, bot activity, and inauthentic content are top of mind for people who love Reddit (and humans).

These issues are actually not new for us. We’ve been fending off bots for 21 years. And before there was AI slop, there was… well, regular slop.

We’ve invested heavily over the last two decades to build and continually enhance safeguards to protect Reddit’s authenticity. As the tactics of bad actors have evolved, our systems have evolved to stop them.

Below, we’re sharing more about how we’re using automated tools to uplevel our ability to keep Reddit real and safe.

Improving our Automated Tools to Reduce Spam Exposure

To stay ahead of the curve, we’re using AI to expand our capability for detecting manipulated and spammy content through new signals and faster enforcement.

We look at signals right when an account is created to stop suspicious actors before they ever get the chance to post. For those that do, we leverage LLMs to catch the highly subtle, coordinated patterns of fake behavior and artificial hype that older systems once missed. Also, we recently announced that any fishy automated accounts will be asked to verify their humanity.

In recent months, these updated automated systems have been working at a massive scale, and we’ve seen some pretty incredible results. We are now:

Blocking 23 million spam views per day before they ever reach a human user.

Catching ~25K net new spammy posts and comments a day.

Reducing spam exposure for our users by ~20% from January to March 2026, relative to the prior three months and an additional 10–15% drop in overall spam account exposure.

Revoking nearly 2M inauthentic votes per day over the last three months.

Our goal is simple: limit the impact of bad actors and exposure to their content. We don’t wait for content to reach the community to decide if it’s harmful. Our most effective work happens before a post is ever seen by a human.

Reducing Exposure to Harmful Content

While protecting the authenticity of conversation on Reddit is critical, keeping our platform healthy also requires a sophisticated approach to combating harmful content like hate, harassment, coordinated abuse or violent threats. While the vast majority of content on Reddit follows our rules, we take our responsibility to action this content extremely seriously, and intentionally bias toward reducing real-world harm and limiting exposure quickly.

We’ve recently expanded our automated systems to support enforcement against hate and violence in all English text content on Reddit (with more languages rolling out soon), leading to critical improvements:

Enforcement in Seconds: The average time between detection and enforcement on harmful content containing hate or violence is down to under five seconds.

Expanded Enforcement: We have increased enforcement actions on hate and violent content by more than 200%.

Reduced Exposure: The faster, higher-volume enforcement has helped reduce exposure to potentially harmful content by more than 40%.

Higher Precision: We’ve decreased false positives (where legitimate, non-violating content is removed) by over 40%.

Our Ongoing Commitment

What makes Reddit unique also makes our work on safety and authenticity unique. Our communities vary widely in size, purpose, and culture, so we focus on building adaptable tools and systems rather than relying on a single, rigid approach for every community.

We have a layered approach to identify and remove spam and other violating content quickly and at scale:

Platform Moderation (Admins): We have sitewide policies that everyone must follow. These are enforced at scale by our internal Safety teams using a combination of automated tooling and human review. Our Safety team is one of Reddit’s largest, with dozens of experts dedicated to proactively preventing and removing violating content.

Community Moderation (Mods): Volunteer moderators enforce community and platform rules, leveraging Reddit’s AI-powered tools to detect low-quality contributors. Key defenses include Reputation Filter to catch spammers, Crowd Control to restrict low-trust accounts, and Ban-Evasion Detection to block returning bad actors.

Up/Downvoting...