Social media posts can serve as an early warning system for rising unemployment, according to a new study published in PNAS Nexus. Researchers led by Sam Fraiberger developed an artificial intelligence model capable of detecting signs of joblessness in online conversations, enabling predictions of official unemployment claims up to two weeks before government data is released. The study analyzed data from 31.5 million Twitter users between 2020 and 2022, using a transformer-based AI model called JoblessBERT. This model was specifically trained to identify unemployment-related posts—even those containing slang, abbreviations, or misspellings like “I needa job!”—which traditional keyword-based methods often miss. To address the limitations of Twitter’s user demographics, which do not fully represent the general population, the researchers applied demographic adjustments to ensure more accurate national, state, and city-level forecasts. The results showed that JoblessBERT detected nearly three times more unemployment disclosures than earlier rule-based systems, while maintaining high precision. The model significantly outperformed conventional forecasting methods, reducing prediction errors by 54.3% compared to the industry consensus. Its ability to detect rapid shifts in labor market conditions proved especially valuable during the early stages of the COVID-19 pandemic. In March 2020, the AI system flagged a sharp increase in unemployment-related posts days before official claims data was published, offering a timely indicator of economic distress. The findings highlight the potential of combining AI with real-time social media data to enhance traditional economic monitoring. By providing faster, more granular insights, such models could support policymakers in responding quickly to economic downturns, particularly during crises when timely data is critical.