Yes. Some of the cheap vendors are already struggling because they cannot compete with the new wave of even cheaper data (‘silicon samples’).

These panels already exist:
amerispeak.norc.org (@norc.org)
uasdata.usc.edu/index.php (@dornsife.usc.edu)
ssrs.com/ssrs-solutio... (@ssrsresearch.bsky.social)
www.lissdata.nl (NL)
www.gesis.org/en/gesis-pan... (D @gesis.org)
and many more...

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.

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.

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..

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.