Researchers Find a Better Way to Spot Russian Bot Networks on Twitter

When researchers want to expose a foreign propaganda campaign, they often look for the same thing you would: repeated posts. If hundreds of accounts tweet the identical message, that looks like coordination. But a new paper argues that not all repetition means the same thing — and treating it that way can send investigators chasing ghosts.

The authors studied 187,000 English-language tweets from six separate Russian influence campaigns on Twitter. They split repeated posts into two buckets. "Generic" posts are low-information and reusable — think 'Good morning!' or 'Great game!' Anyone might post those, so they create false links between strangers. "Non-generic" posts are specific and unusual, the kind that practically only spread when someone is organizing it. Then they used an AI language model to label tweets at scale, with humans checking the work.

Here's where it gets interesting. When the researchers looked for repeats using exact word matching, generic posts barely showed up. But when they used smarter AI methods that compare meaning rather than wording, generic posts made up nearly 39% of the campaigns they found. In other words, the most modern detection tools were also the ones most likely to flag noise as conspiracy.

After filtering out the generic posts, the picture changed: the detected networks shrank and lost some of their biggest clusters, but what remained was denser — a smaller, tighter, more deliberately organized group. For you as a social media user, this means the tools platforms and journalists use to spot manipulation may be getting more accurate. That matters when a story about "thousands of fake accounts" shapes what you believe about an election, a protest, or a public health campaign. Fewer false positives mean the real campaigns get taken more seriously.