Information and content you can trust

Preventing abuse

To keep people safe from abusive content, we use protections powered by machine learning. Gmail automatically blocks nearly 10 million spam emails from inboxes every minute, and Search has tools to prevent Autocomplete from suggesting potentially harmful queries. Automatic detection helps YouTube remove harmful content efficiently, effectively, and at scale — in Q2 of 2023, 93% of policy-violative videos removed from YouTube were first detected automatically. We also implement safety guardrails in our generative AI tools to minimize the risk of their being used to create harmful content.

In addition, each of our products is governed by a set of policies that outlines acceptable and unacceptable content and behaviors. Our policies are continuously honed and updated to address emerging risks. When it comes to our work in AI, we also rely on our AI Principles to guide product development and help us test and evaluate every AI application before it launches.

Detecting harmful content

As the tactics of bad actors evolve, we must work even harder to detect harmful content that gets onto our products. AI is helping us scale abuse detection across our platforms. AI-powered classifiers help quickly flag potentially harmful content for removal or escalation to a human reviewer. In 2022, automated enforcement helped us detect and block over 51.2 million ads with hate speech, violence, and harmful health claims. Additionally, large language models, a breakthrough type of AI, show the promise of exponentially reducing the time it takes to detect and evaluate harmful material, especially from new and emerging risks.

We also work with outside organizations that flag content they think may be harmful. Both Google and YouTube take feedback from hundreds of Priority Flaggers, organizations around the world with cultural and subject-matter expertise that escalate content to us for review.

Responding appropriately

We rely on both people and AI-driven technology to evaluate potential policy violations and respond appropriately to content that is flagged. When a piece of content violates our policies, we can restrict, remove, demonetize, or take account-level actions to reduce future abuse.

In 2022, Google Maps blocked or removed over 300 million pieces of fake content, 115 million policy-violating reviews, and 20 million attempts to create fake business profiles. In Q2 2023, YouTube removed over 14 million channels and 7 million videos for violating our Community Guidelines.

To evaluate context and nuance, while reducing the risk of over-removal, we rely on ~20,000 expertly trained reviewers to work in a variety of roles to enforce our policies, moderate content, and evaluate flagged content across Google’s products and services.

If a creator or publisher feels we’ve made a wrong call, they have the ability to appeal our decisions.