Florian Keusch
floriankeusch.bsky.social
Florian Keusch
@floriankeusch.bsky.social
Austrian in Mannheim; Prof. of Social Data Science & Methodology at U of Mannheim; survey methodologist; (mobile) web surveys; passive mobile data collection
Yes. Some of the cheap vendors are already struggling because they cannot compete with the new wave of even cheaper data (‘silicon samples’).
November 18, 2025 at 9:49 PM
I really see a potential threat for studies that recruite samples via social media platforms (but these samples were wacky even before LLMs) or some river sampling. Not for well managed (nonprobability) online panels and certainly not for high-quality probability panels.
November 18, 2025 at 9:30 PM
But how would this work in reality? Say I build multiple bots, then I would need to register for each of that bot in a panel. If a panel follows the ESOMAR guidelines then they do double-opt-in verification and might even require a bank account or at least a unique email.
November 18, 2025 at 9:30 PM
But there are a number of high quality probability based online panels where I think the influence of LLMs and AI-respondents will be very limited. These are not cheap for a reason..
November 18, 2025 at 8:27 PM
True for nonprobability online access panels that allow self-selection and river sampling-like approaches. Many of them had major quality issues even before LLMs.
November 18, 2025 at 8:27 PM