In my field, we typically just use plain bibtex files. With a proper bibtex file and a decent journal latex template, the template should do the vast majority of the work towards formatting the bibliography the way the journal wants.

Typically there are some remaining warts (both due to suboptimal templates and to limitations of bibtex itself and its community/ecosystem like the address vs. location conundrum, those who have found it will know what I'm referring to), but often they are insignificant enough to just deal with them in the typesetting stage, and when they aren't they don't require much manual work, to me it's not even near the top in the ranking of inconveniences of publishing a scientific paper. And I'm not sure this kind of software would solve such nuances anyway.

Some years ago I tried several of this kind of referencing software, including Mendeley and a couple others (not Zotero because at that point it was Firefox-only IIRC, which was a red flag for me, I'm particular about using the browser I like) and to be honest I found them more trouble than they were worth, although seeing the general positivity here towards Zotero I think I'll give it a try now.

Of course, fields that use Word instead of Latex are a totally different world where probably nothing of my comment will apply.

A RegEx-based parsing system in JavaScript helps with the reference formatting. EndNote makes references uneditable (unless you also have a copy of EndNote) and often cannot adjust to a particular journal's idiosyncrasies...but I can do that easily with home-built software.

Zotero may be great, and I cannot really make an assessment one way or the other. All I know is that hardly anyone in my industry uses it. It's really a nonstarter. Often, people need special permission to get access to open-source software and IT departments do not look kindly on installation of these tools.

Not to mention, I can't think of one piece of software or database I was using in 2003 that I can still access and use except for plain text files. This kind of software has no longevity. For these reasons, the academic publishing industry has not really standardized the process.

After a copy of RefMan broke, following the EndNote acquisition, I remember a time when I had to reverse engineer the database so I could still access the contents. It was a big tabular file and there was a much easier way to make it work with a simple browser-based tool and a few lines of code. I made it work and showed it to my boss, who summarily rejected the solution, even though it worked much better than the original (broken) software and broken database files for hundreds of prior projects. He emphasized that he wanted a paid solution. I still don't understand the logic. These are the kinds of nightmares that I associate with any kind of proprietary software solution. Often, in publishing, we need something that will work now and will still be accessible in twenty years. That's hard to do with free software.

I see, so you wrote your own solution which you have to maintain and extend for each different citation style… pros and cons to that obviously.

Why is it so important that a citation system works in 2003 and in 2021? Papers once published become historical artefacts essentially.

I am thinking of one particular journal that had a problem with historical reference libraries under RefMan that all broke after the EndNote migration. A library of thousands of purchased articles, including pivotal trials that are commonly referenced, essentially became inaccessible after RefMan was purchased by EndNote. I believe EndNote also remotely inactivated the software and engaged in some other chicanery. That experience turned me off of reference manager software. Maybe Zotero has gotten it right this time! I don't know. I just know that few things last in tech and least of all libraries of data housed in software that may or may not be updated or supported by future systems.