Yeah, my experience is that they really don't follow directions with precision and they seem to always identify themselves.
The quality/cost of service have dropped off imo, I used to rave about them circa ~2017 but the intelligence of the service really dropped off.
The idea is great, I still use them once in a blue moon, but the execution is in this uncanny valley of not being cheap enough for rote work (like compiling expense spreadsheets from texted receipts) and not intelligent enough for complex research tasks.