Abstract
Due to an increased usage of AI agents in online behavioural research, several detection methods have been developed to ensure that data responses are from human participants. However, vision-enabled AI agents (e.g., ChatGPT, Perplexity Comet) are still able to complete studies that utilize traditional detection methods. Thus, the following study used the Cognitive Trap Framework introduced by Affonso (2026), which consists of six questions that are easy for human participants to complete but difficult for AI agents to process. A total of 410 participants on the Project Implicit platform completed the six "cognitive trap" questions. Participants failed an average of 0.78 questions, with only 5.61% of the sample failing three or more of the questions. Results indicate that the use of AI agents on the Project Implicit platform may be low at this time.