You can start making statistical conclusions from as few as 3 samples and from around 30 you’re only getting square root improvements in margin of error. 600+ respondents is more than enough if there’s no selection bias, which I’d trust YouGov to be able to handle.