On October 2, 2026, the communication agency Reputation Age released a study in partnership with Bubbling, a French platform that evaluates how brands appear in conversational AI systems. The report, titled Semester Results as Seen by ChatGPT and Gemini, examined the accuracy of financial data retrieved by two major AI models—ChatGPT and Gemini—when asked about the performance of 39 companies listed in the CAC 40, a major European stock market index. The study found that 40 percent of the financial figures provided by the AI models were incorrect. The accuracy varied depending on the type of financial metric requested, with the net result being correct in 67 percent of cases, net debt in 66 percent, and EBITDA in 57 percent. The published growth rate, however, was correct in only 42 percent of responses. The study categorized the errors into two main types. The first involved outdated or fabricated figures, with the highest error rate observed for EBITDA at 36 percent. The second type occurred when the AI referenced an official figure from a company that did not match the specific question, most notably when distinguishing between organic growth and published growth or between quarterly and semester data. For example, 24 percent of responses on published growth incorrectly cited a different figure. The error rate varied per company, with 46 percent of checks on BNP Paribas and 43 percent on Crédit Agricole affected by these selection flaws. The accuracy of the AI models ranged from 81 percent for Capgemini to 32 percent for Crédit Agricole. The study only published results for 16 companies—seven with the highest accuracy and nine with the lowest. Even Capgemini, the best-performing company in the sample, showed how confusing financial data could be. Its communication document listed seven different growth rates, but only two directly related to the current semester, and no organic growth rate was provided. As a result, an AI asked about Capgemini's organic growth would find no direct answer, as the 11.3 percent growth rate included the impact of recent acquisitions. The study also noted that 78 percent of the sources used by the AI models were company websites, while only 0.08 percent came from traditional press outlets, which had largely blocked AI access. The citations were almost exclusively provided by ChatGPT, with Gemini rarely including links. This means the study primarily reflects the behavior of one of the two AI models. Additionally, the study did not specify the versions of the AI models used, nor did it detail how accuracy was measured or how many figures were checked per company. The lack of transparency in the methodology raises questions about the reliability of the findings.