This issue was reported in a community talk page here (in French): https://fr.wikipedia.org/wiki/Wikip%C3%A9dia:Le_Bistro/28_octobre_2025#c-Handisport_marmotte-20251028031000-EBlackorby-WMF-20251027192500
Translation below:
Good evening, I’m experimenting with your model and my Voice Over tells me by second “file new image gif”. Usually VoiceOver is able to describe a picture or photo. This by speech synthesis. VoiceOver adds several seconds "text possible". My assumption your bidule is not accessible to blind and semi-sighted people. Handisport marmot (Talk) 28 October 2025 at 04:10 (CET)
Theory
iOS uses an on-device machine learning system to do computer vision and "see" images when VoiceOver is being used; users are given a verbal description of the content of the image. This is in addition to any alt text that may be present, as far as I understand. This feature is not a web standard (it's an added thing that iOS does), but it would be good to support these image descriptions if possible
My theory for why this is happening is that we have marked up some images in this feature as <button> elements (we use any provided alt text as labels in this case). We've done this when the images are interactive elements which trigger changes in UI state (as opposed to navigating to a new location). It seems likely to me that button markup is not processed for ML-powered image descriptions the way that other elements like <img> or <a> tags likely are.
Proposed fix
We'd need to test out some different markup changes and then test in VoiceOver in order to see if we can get the image description behavior working again.
However, changing this markup may have other consequences (we are using <button> elements here in the first place out of accessibility considerations too). We don't want to fix one problem while breaking something else at the same time. Styles may also need to be re-worked if there are significant markup changes.
We are about to launch an A/B test of this feature, so this kind of larger markup change is too risky to explore right now. We should come back to this after the A/B test has concluded, assuming we decide we want to continue development of this feature.