AFAIU Reuters is one of the leaders in terms of news/wire agencies for generating things like that via machine learning and other algos.

It might be an automated choice, where the author or editor merely had to check a few boxes before publishing to get a remotely relevant clipart and caption, essentially. Saves a lot of time and money in an age where it's sort of a weird abstract form that needs to be fulfilled.

-> Hed

-> Byline - Date

-> (Dek)

-> Graphic -> Caption

-> Body Copy

That is, it's not to say the system is elegant, yet....

https://innovation.thomsonreuters.com/en/labs.html

edit: LOL. Just a few more clicks brought me to this article. I laughed.

https://blogs.thomsonreuters.com/answerson/art-artificial-in...