AI Detection

The goal of the present project is to evaluate a set of items designed to detect survey responses generated by large language models (LLMs). We have previously shown that these items are able to label responses generated by LLMs as being generated by LLMs. However, to ensure that the items do not incorrectly label responses generated by humans as being generated by LLMs, we need to test the items in samples of humans. To that end, we will compare responses to the items from six human samples (e.g., CloudResearch Connect; Prolific) to responses to the items from six LLM samples (e.g., Gemini 2.5 Pro; OpenAI gpt-5.1).