This paper presents robust evidence indicating that the tone of financial reports from the US mining industry firms can predict certain mining commodity returns. We assess this predictive ability with different tests, including evaluations of mean squared prediction errors, correlations, mean directional accuracy, and trading strategies. We compared our results with several benchmarks proposed in the literature, including the random walk, outperforming most statistically and economically. Our findings are related to the literature on firms' strategic disclosure choices, suggesting that firms may strategically tailor their disclosure decisions to influence investor expectations regarding future performance. Additionally, we conduct a series of placebo tests, revealing significant outcomes: (1) the tone of financial reports from other industries does not exhibit predictive power for mining industry commodity returns, (2) this predictive ability is not overshadowed when controlling for uncertainty measures, such as VIX, OVX, realized volatility and Google Trends commodity price searches, and (3) the tone of the mining industry's financial report does not predict returns for unrelated mining commodities. To our knowledge, our work is the first paper to employ financial report tone through text mining of nearly 60,000 financial reports to forecast commodity returns. Our findings substantially contribute to the commodity forecasting literature, offering valuable insights for portfolio managers and professionals seeking to enhance their forecasting capabilities.