I think we’re not sure how much it’d help with CV - remember, the CV is trained to recognize photos of an organism, not the actual organism itself. So what’s most helpful is to take photos that would be similar to ones people would take in the wild. If the animal is in a zoo and things like bars are in the foreground or the background is not natural, it might not help the CV provide the best suggestions for in situ photos of the same organism in the wild. Similarly, if the CV was trained on a bunch photos of pinned and spread specimens of a certain insect species, it might not be able to properly ID that species from in situ photos where the wings are folded and the background is not white. But, we don’t know that definitively, it would take some experimentation which is unfortunately timely and costly.
Photos of a plant in a botanical garden would probably be a much closer match to in situ wild photos of that plant.
That being said, in general I’d use iNat in ways that make you happy and (in IMO) focus mostly on wild organisms. But the latter isn’t a hard and fast rule, as long as the observations are properly marked.