Include links with no text as "non-descriptive" in SEO audit

[patrickhulce]

Summary
Split out from #6483 (comment) per @rviscomi 's request. We should consider including all links that don't have any text content in our non-descriptive check. A link that doesn't have any text at all is just as non-descriptive as these other links :)

I think the main concern is that this will start including a bunch of things we don't foresee.

[rviscomi]

Thanks for filing this @patrickhulce! Will investigate.

[rviscomi]

Assigning to Matt to start implementing and testing on popular sites. Let's see what the new failure rate is before releasing it.

[mattzeunert]

Ran on 100 sites from here.

	Before	After
Failure rate	38%	81%
Average item count	0.75	10.5
Median item count	0	3

Right now the AnchorElements artifact uses innerText as the link text, which means the text is empty if the link is hidden. I've used textContent for now. If we don't want to count invisible elements we can detect that in the gatherer.

[rviscomi]

Hmm I think we should probably ignore hidden text to be consistent with the users' experience.

I'm concerned about the big spike in the failure rate though. Can you remind me when a warning gets triggered as opposed to a failure? Looked at the audit source code but didn't see a distinction. It might be worthwhile to warn about empty anchors rather than failing the audit. Not sure.

[mattzeunert]

Sorry, I meant items rather than warnings. Having one failing item in the audit result will cause the audit to fail.

Most of the time the images contain an image instead of text, sometimes with an aria-label.

[rviscomi]

Maybe we should be ignoring a > img cases? If they were to cause the audit to fail, what would the recommended fix even be: replace the image with text? put text next to the image? Not sure if img[alt] attributes are used instead, but I do know that aria-label attributes are not. I'll consult the WTAs.

[rviscomi]

Martin (@AVGP) suggested that if the anchor contains an image, we fall back to its alt text. Could we see how the failure rate changes with that implementation?

[mattzeunert]

This is if we include count image alt attributes as text:

	Before	After
Failure rate	37%	70%
Average item count	0.74	7.0
Median item count	0	1.5

Actually both of those numbers are inflated because I removed the visibility check. This is the result if we ignore invisible elements again:

	Before	After
Failure rate	27%	63%
Average item count	0.57	3.0
Median item count	0	1

[paulirish]

I'm pretty scared to make this policy change. We'd have to look at all the possible failures and determine if adding text is a reasonable solution for them all.

Would this be an improvement for accessibility or SEO? And why?

It seems like no one is strongly advocating for it right now, so let's shelve it?

[AVGP]

To be frank the high failure rate makes me nervous, too. The benefits are that it's a consequent follow up for the audit on meaningful link text and for accessibility and SEO its quite important to have context for images, i.e. alt text except for decorative images and those probably should not be the only affordance for whatever destination the link is providing... so I'm advocating for it, but I think I would like to better understand where the big chunk of failures are, maybe we need to educate the community better first. I will try to find some time in the upcoming weeks to take a look at the cause of these issues. sounds good? Paul Irish <notifications@github.com> schrieb am Mo., 25. Feb. 2019, 21:38:

…

I'm pretty scared to make this policy change. We'd have to look at all the possible failures and determine if adding text is a reasonable solution for them all. Would this be an improvement for accessibility or SEO? And why? It seems like no one is strongly advocating for it right now, so let's shelve it? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#6495 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAWmRmQbEM_2QLEbohLpLDIY87JPJANeks5vREncgaJpZM4YRfQj> .

[rviscomi]

SGTM. This is a low priority change so we have time to vet it properly.

[mattzeunert]

Here's the output from running the script.

Quite a few of them have an aria-label or a title, but I guess those don't count for SEO?

[connorjclark]

@AVGP @umaar is this still a priority in your opinions?

[AVGP]

We figured it isn't that pressing of an issue. If in doubt, I'd skip this.

[brendankenny]

Great, thanks! We can always revisit if it turns out we do have a blindspot here.