Making GIFs Accessible
Cole Gleason, Amy Pavel, Himalini Gururaj, Kris Kitani, Jeffrey Bigham · 2020 · Proceedings of the 22nd International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS 2020) · doi:10.1145/3373625.3417027
Summary
This Carnegie Mellon University study examines the accessibility of GIFs on social media, a visual medium that has become central to online conversation but remains largely inaccessible to people with vision impairments. The researchers conducted a multi-part investigation: first, a large-scale analysis of GIF usage on Twitter using approximately 108 million tweets collected continuously from February 26 to March 13, 2020, which contained 791,600 GIFs (0.7% of tweets). After filtering, 303,874 GIFs remained, of which only 126 (0.04%) contained alternative text — the ability to add alt text to GIFs on Twitter had launched just one month prior. The researchers identified 127,916 unique GIFs (42% of total), with 187 exceeding 100 uses, following a long-tail distribution suggesting that accessible descriptions for the most popular GIFs could be reused at scale. They then coded visual elements in 97 of the 1,000 most popular GIFs, finding that 87 were excerpted from longer videos, 77 were live-action, 85 contained at least one face (58 where the face was the focus), 53 showed characters performing actions, 37 contained visual indications of sound, and 14 contained text. Second, the researchers conducted formative interviews with 10 blind Twitter users (ages 20-54, average 36.2 years) about their experiences with GIFs online, followed by a second session where participants compared three accessible alternative formats for 15 representative GIFs: alternative text descriptions, the original source audio from the video the GIF was excerpted from, and spoken audio descriptions overlaid on source audio.
Key findings
The interviews revealed that inaccessible GIFs create real social consequences. One participant described ending an interpersonal relationship because they could never understand the reaction GIFs the other person sent. Five participants reported mostly ignoring GIFs because they are inaccessible, and four had experienced GIFs interrupting conversations. Workarounds were limited: four participants used surrounding tweet text to guess GIF content, one used Microsoft Seeing AI on screenshots, and three relied on friends to describe GIFs — though participants noted the social awkwardness of repeatedly asking. Regarding what information to include in descriptions, participants prioritized: who is in the GIF and what action they are performing; the character name, actor/actress, and source work for GIFs from TV/movies; any text or dialogue present (with care to distinguish overlaid text from original dialogue); and facial expressions, which were the focal point in about 60% of popular GIFs. Participants reported a tension between objective description ("Oprah shrugs") and subjective interpretation ("Oprah shrugs as if to say I told you so"). For format preferences, all participants viewed alternative text as a minimum accessibility requirement — it works with screen readers and Braille displays, can be skimmed quickly, and does not vary in volume. However, six of ten participants preferred audio descriptions as the best way to experience GIFs for a richer, more emotive experience. Source audio alone was the least preferred format (8 of 10 rated it last) because it often did not describe the visual action, contained confusing background noise, or was discordant with the visual meaning. A follow-up sample in June 2020, after Twitter began auto-including GIF titles from aggregation sites like GIPHY, found 47.4% of GIFs had automatic alt text — but 99.3% of this was just the GIF title, which typically names the person but not the action.
Relevance
This paper addresses a growing gap in social media accessibility as visual communication increasingly displaces text. GIFs function as emotional shorthand in online conversation — they convey reactions, attitudes, and cultural references that are integral to social participation. For accessibility practitioners, the finding that only 0.04% of GIFs had user-provided alt text (and even Twitter's automatic titles describe who but not what) highlights both the scale of the problem and the opportunity for platform-level solutions. The recommendation that platforms create libraries of descriptive alt text and audio descriptions for popular GIFs is practical and scalable given the long-tail distribution of GIF usage. The study also extends the theory and practice of image description to animated content, identifying unique challenges: GIFs convey action across multiple frames, contain visual indications of sound, and carry emotional and cultural meaning that may require interpretive rather than purely objective description. For content creators, the paper provides practical guidance on what to include in GIF descriptions: characters, actions, expressions, source material, and text/dialogue. The extension to audio descriptions — where narration is layered over source audio — offers a model for richer accessible alternatives that could apply to short-form video content on platforms like TikTok and Instagram Reels.
Tags: GIF accessibility · alternative text · audio description · blind · low vision · social media · Twitter · image description · visual media
Standards referenced: WCAG 2.1