Our usage may differ, but I would think I'm probably most interested in some relatively little visited articles, because I'm there looking something up not knowing about - versus say 'United Kingdom (country)' is probably quite highly visited in total, but not that likely to be visited.. if that makes some sort of sense.

Might be better to ignore certain topics you know you're not interested in. (That way if there's no result, you know there's either truly no result, or that you don't care anyway because you're not interested.)

Or perhaps most effective for the size goal would be just to cut images. Perhaps selectively include them depending on category - e.g. maybe you want them for science & technology categorised articles (diagrams) and geography, but can do without otherwise.

From the article:

> Kiwix also allows you to download subdivisions of wikipedia — like the top 1 million articles at 43.53GB

The frequency of article visits probably resembles a Pareto distribution. So there would be many many articles with very few views, including new articles.

I’d think the good ol’ Page rank would be a better metric for deciding which ones to prune.

Kiwix lets you download a pack of the top 100,000 articles on your phone which is more than enough for offline browsing.