If you start by opening up the SDS for the Sprint developer and the Marathon developer, you'll notice distinct differences right away. These may or may not work out to be relevant in the working-strength developer - it's hard to be sure based on the SDS. Yet, Ai is adamant, and I wouldn't go that far for sure. Compounding reasons are that the quantities have pretty massive bandwidth in the Sprint SDS and are entirely missing in the Marathon SDS, which makes something like the number given by ChatGPT for hydroquinone content highly doubtful and it does not substantiate the conclusion that it's apparently very similar/the same. This is ignoring the fact that ingredients may be present that are not listed.
Furthermore, that both products may be manufactured by the same party doesn't make them the same or even comparable.
There's numerous small errors like e.g. 'sulfite buffer' or describing phenodine as "metol-like" (in what sense?). I find the prosaic blubber you quoted from AI in post #3 tendentious codswallop that probably quite accurately reflect inane banter about photography on the forums, which just emphasizes that ChatGPT is subject to GIGO (=Garbage In, Garbage Out). There's more inane filler all over the output that you have to weed through to get to the relevant stuff; it's not necessarily incorrect (although you'd have to check every bit to be sure), but it doesn't add much to the argument either.
Do I think that these developers are conceptually similar? Sure. But there's plenty of stuff out there that's 'equally similar', so what does this even mean? It's just a generic borate-buffered PQ developer - that's pretty much all we need to know and it would have taken ChatGPT exaclty one line of output and a whole lot less number crunching to get there.
So the problem I have with output like the stuff you posted is mainly that the inconsistencies, factual errors and haphazard causation sheds doubt on the entirety of the answer, which means you basically have to verify everything to see if it holds up. In the end, you may very well spend more time having to prompt ChatGPT and then doing the desk research all over again as opposed to just pulling up to SDS's, put them side by side and form your own impression.
btw, the 'Sorry' is there because I don't want to discredit your work and kind of feel bad about taking the p*** over AI once again...but boy, does ChatGPT never cease to troll bigtime, LOL!