Previous research has highlighted the benefits of real-time automated feedback in enhancing originality in divergent thinking tasks. In this preregistered study, we sought to replicate these findings, investigating whether improvements in creative ideation persist after feedback is discontinued, and assess the impact on evaluation accuracy. A total of 230 participants were given three divergent thinking tasks (Alternate Uses tests), with or without semantic distance feedback in the first two trials. The third task was always performed without feedback. Participants were then asked to rate the originality of the ideas they produced in this last trial. Their evaluations were compared against originality scores calculated based on semantic distance and Large Language Models (LLM) for converging evidence. The results aligned with previous findings, showing that feedback was effective in improving overall levels of originality across the first two trials. Importantly, this effect carried over to the third trial after feedback was discontinued. However, feedback did not enhance evaluation accuracy, as participants in both conditions achieved relatively high levels of accuracy in rating the originality of their own ideas. We offer possible explanations for this unexpected result and discuss the study’s findings in the broader context of metacognition.