AI Can Read Market Mood, But Can't Predict Your Stocks

Every day, investors and trading algorithms scan social media for mood. The idea is simple: if people sound scared, sell; if they sound excited, buy. Researchers put that idea to the test using 70,500 posts on X about securities class actions — lawsuits accusing a company of misleading investors — posted between 2002 and 2025, and matched them to actual stock price movements. They ran five sentiment tools through the exact same steps, including older dictionary-based programs and newer AI language models.

The surprising result: a tool that agrees with human readers about what a post means is not necessarily the tool that best predicts tomorrow's stock move. Reading tone and forecasting price turn out to be two different skills. Even stranger, whether a tool looked predictive often depended on how the researchers picked their data — a choice made at the desk, not a fact about the real world. That matters, because companies sell these tools on their "accuracy" scores.

The spam finding is just as striking. About 17.6% of the conversation was junk. And the sheer volume of posts predicted nothing at all: not how much the stock fell, and not how large the eventual legal settlement turned out to be. Loud does not mean important. A thousand angry posts carried no more information than a handful.

So what should you do with this? If you or your financial advisor lean on social media mood as a signal, treat it gently. Sentiment tools are reasonably good at describing what people are saying. They are much weaker at telling you what happens next. The things that actually move a stock — earnings, real lawsuits, regulation, competition — still matter far more than how many people are talking online.