Skip to main content

How to detect and prevent the use of Large Language Models in studies

In this guide, you’ll learn how to detect and prevent the use of Learning Language Modules (LLMs) in your studies, should you wish for that limitation to apply. Please note that how vulnerable your study is to LLM usage will vary depending on the study you're conducting.

How to detect the use of LLMs

While there's no foolproof method to definitively detect LLM usage in a study, there are key indicators to watch for:

  • Length vs. time: Analyse the length of a participant's response in relation to the time they took to complete it. If the response was too long and was finished in an unusually short time frame, they might have used an LLM.
  • Check answers: People who use LLMs often use copied content in their responses. Check their answers for signs of this, such as the start of an answer being "Sure, I can help you with XYZ.”
  • Keystrokes and backspaces: Use software to track how many keystrokes and backspaces the participant uses during your study. If these counts are very low or close to zero compared to the text produced, they might have used an LLM. You can also use some software to check if a participant has used the copy and paste functions.

2 proactive ways to prevent participants from using LLMs in a study

Take these two simple steps to prevent LLM usage before it becomes an issue:

  • Directly ask participants not to use LLMs: Many participants might use LLMs simply because they don’t know it’s prohibited. If the use of LLMs isn’t allowed in your study, clearly communicate it.
  • Disable copy and paste: Disable the copy and paste function so participants can’t easily copy content from your study. Also, consider converting all text into images to deter copy and pasting.

Follow these best practices to detect and prevent the use of LLMs. If you have any questions or need further help, please don't hesitate to reach out to our support team.

I need further help

 Click here to contact us

Was this article helpful?
powered by Typeform