BotSight

by NortonLifeLock Labs

Flag suspected bot accounts on Twitter in real-time and in-line using a custom-built state-of-the-art classifier

new BotSight is now available as an app for iOS

BotSight is available as a browser extension for all major browsers.

BotSight is available as a browser extension for all major browsers.

How It Works

BotSight annotates each Twitter handle with a bot-probability score directly on the Twitter website. This works across most of Twitter's views including Timeline, Search, Profile, Follower, and clicking on an individual Tweet.

The scores are generated by our state-of-the-art classifier in real-time. They are then placed directly onto the page, so you can tell if you're interacting with a real person or a bot.

Twitter screenshot with multiple mentions without annotations
Twitter screenshot with multiple mentions with annotations
In addition to annotating the Tweeter, we also scan the contents of the Tweet. In the screenshot above, BotSight also annotates the multiple mentioned handles in the Tweet's body.

In addition to annotating the Tweeter, we also scan the contents of the Tweet. In the screenshot above, BotSight also annotates the multiple mentioned handles in the Tweet's body.

Twitter screenshot with a quoted tweet without annotations
Twitter screenshot with a quoted tweet with annotations
BotSight can also annotate retweets and quoted tweets. It works in a variety of Twitter contexts including search, trending topics, and looking at a Tweet's replies.

BotSight can also annotate retweets and quoted tweets. It works in a variety of Twitter contexts including search, trending topics, and looking at a Tweet's replies.

Twitter follower page screenshot without annotations
Twitter follower page screenshot with annotations
BotSight can also annotate your followers and those you follow. It will also annotate descriptions if they mention other users or affiliations. However, we find this less useful than inline annotations.

While we do our best to make sure BotSight is accurate, it is still in active development. People may be misclassified as bots or vice-versa.

FAQ

  1. Does BotSight work on the Microsoft Edge browser?

    Not yet.

  2. Do you have plans for an Android app?

    If enough people are interested. If you are, please send an email to the address listed below.

  3. I'm seeing some human accounts with a score of 98% and others at 91%. What's the difference?

    Any account with a human score over 90% almost certainly belongs to a human.

  4. The percentages in the screenshots differ slightly from what I'm seeing.

    The model evaluates each account in real time, so as accounts tweet over time their percentages will naturally fluctuate.

  5. I found an account that's definitely a bot that has a green score. Or I found an account that's definitely not a bot but it has a red score. How can I report it?

    Please email DL-botsight [at] nortonlifelock [dot] com with a screenshot, the name of the account, and why you think the model is wrong. The model is improving every day and we will incorporate all feedback.

  6. Will BotSight eventually work on other social networks such as Facebook or Instagram?

    We hope so, but there is very little publicly available data provided by Facebook (which owns Instagram) about the bots that it has taken down. This makes it very hard for us (or anyone outside of Facebook) to build an effective bot-detection tool for Facebook or Instagram.

About the Authors

This extension was developed at NortonLifeLock Research Group (NRG), part of NortonLifeLock.

Privacy

Your privacy in very important to us. Please see our NortonLifeLock Global Privacy Statement, along with our NortonLifeLock Product Specific Privacy Statement for BotSight.