My observation with ChatGPT, for both programming and writing, is that it's a great way to get started, but - as you also illuminated - a polished final product is challenging. Also, it has a word token limit that it doesn't tell you about unless you ask. If you run into that limit, it gets really unwieldy without telling you why. That's where it seems especially likely to ignore requests for corrections or to reintroduce previously corrected errors.

To me (at least with the free version), the best work flow seems to be:

1. Get an outline and/or draft version from ChatGPT/Bard
2. Iterate until satisfied
   2.1. Polish it with human revisions
   2.2. Ask for advice from ChatGPT/Bard. This can take the form of generalized feedback about the revisions or delving into specific details.

To me, it seems like a very helpful assistant, but it's definitely not ready to be a sole content producer for high quality content. I don't think it really saves much time (which you also noticed), but I do think that the interactive process can produce a better result. (Kind of like a Jevons paradox for information productivity. ;-)

In general, I prefer ChatGPT over bard, but there have been cases where Bard gave me better results. With the 2021 information cutoff that you mentioned, Bard might do a better job on articles about current events.

Aside from writing articles, I do have to say that this is a pretty useful skill - but again - it doesn't always get it right, so supervision and critical evaluation is needed from the human side:

Me

ChatGPT