Thread by @yoyoel

Since Saturday, we’ve been focused on addressing the surge in hateful conduct on Twitter. We’ve made measurable progress, removing more than 1500 accounts and reducing impressions on this content to nearly zero. Here’s the latest on our work, and what’s next.

Our primary success measure for content moderation is impressions: how many times harmful content is seen by our users. The changes we’ve made have almost entirely eliminated impressions on this content in search and elsewhere across Twitter.

Impressions on this content typically are extremely low, platform-wide. We’re primarily dealing with a focused, short-term trolling campaign. The 1500 accounts we removed don’t correspond with 1500 people; many are repeat bad actors.

Impressions don’t tell the whole story. These issues aren’t new, and the people targeted by hateful conduct aren’t numbers or data points. We’re going to continue investing in policy and technology to make things better.

Many of you have said you’ve reported hateful conduct and received notices saying it’s not a violation. Here’s why, and what we’re doing to fix it:

To try to understand the context behind potentially harmful Tweets, we treat first-person and bystander reports differently.

First person: This hateful interaction is happening to or targeting me.

Bystander: This is happening to someone else.

Why? Because bystanders don’t always have full context, we have a higher bar for bystander reports in order to find a violation. As a result, many reports of Tweets that in fact do violate our policies end up marked as non-violative on first review.

We’re changing how we enforce these policies, but not the policies themselves, to address the gaps here.

You’ll hear more from me and our teams in the days to come as we make progress. Talk is cheap; expect the data that proves we’re making meaningful improvements.