What percentage of images on SE have a meaningful alt-text?

Question

As a follow-up question of Stop image upload completion from stealing cursor focus, I wonder what percentage of images on SE have a meaningful alt-text.

Edit: by meaningful, I mean useful.

Answer 1

Here is a SEDE query which analyses the alt text of the all images included as Markdown in every post. No doubt it's possible to extend this to images included as HTML (<img alt="...), but those are relatively rare. Anyway, it gives a general idea of which alt texts are used and how often. They are sorted by frequency, and I expect the most meaningful to be rather unique, so at the bottom of the list.

There are 61k of those images on Meta Stack Exchange, so about a third has the boilerplate 'enter image description here' (a lot less than I expected). 'alt text' was the default in early years, as @Sonic noted. No alt text (the blank entry) is also quite common (12% of the occurrences) - probably mostly from users manually entering the Markdown with an already uploaded or hotlinked image. The usernames seem to be coming from automated Town Hall transcripts.

There are about 23k unique alt texts on Meta Stack Exchange. When you run the query on other sites, please be aware that SEDE can only return 50k rows - the total # of images will be correct though. Anyway, it'll definitely time out on Stack Overflow - feel free to fork it to include a date range or something, or run it with access mere mortals daren't dream about.

Answer 2

I took a slightly modified version of Glorfindel's query and ran it where you can't: against Stack Overflow. It took a long time, as you can imagine, and yielded that people are generally much better about this here on meta than on Stack Overflow - perhaps because the audience here is of generally more experience on the network, or that more accessibility issues are discussed (and so set better examples), or because Stack Overflow's image population is largely weighted toward newer users asking questions to solve immediate problems and not as interested as answerers in providing information for future readers.

Here is a partial result; I stopped after we got to single-instance:

I casually scanned through the first, oh, thousand or so entries, and found very few that seemed like meaningful text.

And especially because "meaningful" is itself subjective, and the tail is really, really long, I think it will be very hard to come up with any useful number or percentage.

I don't know what we can now do with that information, especially at scale, other than make a subjective determination that it is bad enough we need to implement some kind of feature to force users to add descriptive text.

I think like the "explaining down-votes" debate, you could argue that lazy people who want to watch the world burn will work around the system by entering garbage, adding or removing a single character, putting in a poop emoji, and what have you.

As mentioned in a comment, it is still hard to draw a line indicating accessibility problems, because lack of alt text on its own doesn't indicate (and would be impossible to determine programmatically, again at scale) whether there is helpful context above/below the image, that simply doesn't happen to be repeated in the alt tag. While it can't hurt to repeat it there, it could be an accessibility issue in the opposite way (since the same information might be repeated), and may make the work of maintainers harder, since future edits will need to keep the alt text in sync in two places.

Honestly I think positive reinforcement is the best policy here, and will admit that I am already much more aware of my lack of attention to detail in this area. Places where changes won't jeopardize site functionality, like the staging ground, post wizards, the tour, and FAQs, as well as our own behaviors when helping other users, can emphasize the usefulness of including this information.