I have used Turnitin and Quillbot for AI detection, and I think they are not accurate. Unless there are very generic writings that may be repeated in many manuscripts, I do not think any tool will be able to distinguish between an actual writing and the LLM. This is due to the inherent generative architecture of the LLMS.
Furthermore, as the LLMs improve, detecting their usage for academic writing will become even more challenging. My observation is that the companies that are running these LLMs, like Anthropic, have substantial resources and are constantly coming up with cutting edge features, while the software usually used in academia is a decade behind or more.
Turnitin has yet to solve rudimentary UX and software engineering issues, let alone come up with advanced pattern recognition algorithms.