Unless you're using an ephemeral on-prem LLM that doesn't keep logs you should begin with an assumption that your AI chat transcripts will be available to your opponents in discovery. There isn't clear caselaw on this yet, but given that material you share with third parties other than your lawyers almost always is and sharing your strategy musing would probably be disastrous, you should plan accordingly.
Particularly, subpoena are quite powerful in the US and can be obtained quite easily and often without an adequate opportunity for the ultimate user to oppose them. The AI providers will comply with them as just a matter of process, and the first notice that an opponent is seeking your transcripts may be after they already have them. (And that's assuming your opposition doesn't have insiders at the AI company to begin with...)
That aside-- Having thoroughly dispatched an AI dependent opponent in court, uhh, well lets just say that I hope any future opponents read this article. ChatGPT outputs an astonishingly embarrassing amount of completely wrong pseudolegal nonsense and set him up for multiple case losing disasters.
There are also a number of advanced strategies available to an opponent who has realized you are using an LLM, as they now have oracle access to something similar to your own AI advice and can test their actions against the advice it will cause you to get from your AI. ... even so far as brute force searching rewrites of the language in their correspondence to increase the odds that you get disastrous advice.