Prolific has a lot of bot users, and people who use LLMs to formulate their replies to surveys. Here's a nice paper on it: https://arxiv.org/abs/2508.01390

This isn't specific to prolific. Rather, it's likely endemic to all paid research platforms, and will be a challenge forever.

We would like to have some hope of identifying these users so that we have the option of excluding them from our analysis. Doing it via analysis of their text alone is difficult-to-impossible. To that end, we'd like to record behavioral data (keystroke timing, copy/paste actions, text field change actions, etc), which will help us determine if there is a human behind the keyboard, and if they're pulling their text-based responses from elsewhere.

Using this data well is also a challenge, especially since we need to account properly for people who use assistive devices, speech-to-text, etc. However, the first step is to have the data at all.

I have a draft of some code for this ready, and will make a PR shortly. This is just the accompanying issue.