like I said the way to approach this is to try to identify with the person you're pitching to and address their needs/goals.

The most obvious thing is to familiarize yourself with their research because they have research goals. But honestly you don't need to go overboard on this - no one that's not their student yet would be expected to know the research in depth. Pick 3 of their most recent papers, read the abstracts and conclusions and come up with vaguely interesting questions. Again they don't need to be deep they just need to be better than "what learning rate did you use" and more sophisticated than "why did you pick Adam over rmsprop". Obviously extra points if have your own perspective on something (even if it's not earth shatteringly novel). For example my new group works on distributed ML and on my first call I said that I'd always wondered about distributed inference (ie one test sample forward pass distributed across multiple nodes). I didn't have an answer or solution to the problem but it was something they hadn't considered.

The better thing though to discuss is new potential funding directions that you would enable them to avail themselves of. Figuring this out is admittedly tough if you're an undergrad because you don't know what you're good enough at that you could get funding. Again for example in my case I have some computer vision experience and so now my group can plausibly apply for grants at the intersection of distributed systems and CV. I didn't quite discover this intentionally but early conversations broached it organically and I recognized the opportunity to include it in my "pitch".

But even if you don't have some kind of skills alignment like that, the most important thing to stress is that you can independently start and complete projects and you can write/present academic work. That means you can speak in jargon, you know how to display results (plots, comparisons benchmarks, etc), and most importantly you can concisely explain what you've done. I had an unpublished lit review that I sent over as a writing sample.

This is getting a little long so I'll just say something everyone should always keep in mind: you're trying to build a rapport with a person. No matter how smart, successful, elite they are in their area they're not robots they're fleshly people first. Being friendly and funny and getting to know them outside of their research helps immensely (during the meeting not before - that's creepy!)

edit: sorry also another thing you can do is prep by speaking with a grad student and asking them good questions about the research, group, goals, PI, etc. this'll get you comfortable speaking in terms that the PI is certain to relate to.