From Automation to User Empowerment: Investigating the Role of a Semi-automatic Tool in Social Media Accessibility
Leticia Seixas Pereira, Jose Coelho, Andre Rodrigues, Joao Guerreiro, Tiago Guerreiro, Carlos Duarte · 2024 · ACM Transactions on Accessible Computing · doi:10.1145/3647643
Summary
This paper presents SONAAR (Social Networks Accessible Authoring), a semi-automatic tool designed to improve social media accessibility by combining AI-powered image recognition with user-generated descriptions. The research addresses a persistent problem in digital accessibility: images shared on social media platforms frequently lack alternative text, creating barriers for people who use screen readers. Rather than relying solely on automated solutions or placing all responsibility on content creators, SONAAR takes a hybrid approach that leverages crowdsourcing alongside machine learning. The authors conducted three complementary studies to understand barriers and evaluate their solution. Study 1 combined a survey (258 responses) with semi-structured interviews (20 participants) to identify why accessible content creation remains rare on social media. Five key barriers emerged: lack of awareness about the importance of alt text, lack of know-how about how to add descriptions, the cost of additional effort required, lack of standardization across platforms, and inaccessibility of the platforms themselves. Many participants, including those with visual impairments, were unaware that social media platforms even support alternative text. SONAAR operates through both a Chrome browser extension and an Android application. When users encounter images without descriptions, they can request suggestions from a database of user-generated descriptions or receive AI-generated concepts from the Clarifai image recognition service. Users can also contribute their own descriptions to build the crowdsourced database. Study 2 analyzed usage logs showing 197 requests from 17 unique users, with 88% of requests coming through the browser extension and the majority of users having visual impairments.
Key findings
The research revealed that visually impaired users rarely encounter images with alternative text on social media, and most content creators do not engage in accessible practices despite platform support. Even participants who were aware of alt text options found the process too time-consuming or did not know where to find the feature. Study 3 evaluated the impact of providing educational documentation alongside SONAAR, with 24 participants completing tasks while reviewing materials about accessibility practices and how people with disabilities experience the web. Most participants rated the documentation as clear, comprehensive, and helpful. Importantly, the majority indicated increased motivation to adopt accessible practices after reviewing the materials, suggesting that awareness-building can shift attitudes toward accessibility. The log analysis showed higher engagement from visually impaired users compared to sighted users, with the current user base consisting primarily of people who benefit directly from improved image descriptions. This indicates that tools designed for content consumers may attract more sustained engagement than those targeting content creators. The browser extension saw significantly more usage than the mobile app, partly due to its ability to scan entire pages automatically rather than requiring manual image selection. Participants provided positive feedback about SONAAR overall, though they noted limitations including restricted language support (English and Portuguese only) and occasional interface difficulties with Facebook.
Relevance
This research offers valuable insights for organizations seeking to improve social media accessibility through technological intervention. The hybrid approach combining automation with human contribution addresses the quality limitations of purely AI-generated descriptions while reducing the burden on individual content creators. The five barriers identified provide a useful framework for accessibility advocates and trainers. Simply providing tools is insufficient; users need awareness of why accessibility matters, practical training on how to implement it, and streamlined workflows that minimize additional effort. The finding that documentation improved motivation suggests that pairing accessibility tools with educational content may increase adoption. For practitioners building accessible social media strategies, the research highlights that platform support for alt text is meaningless if users cannot find or do not understand the feature. Organizations should consider internal training and clear guidance alongside any technical solutions. The predominance of visually impaired users in the SONAAR user base also raises questions about how to engage sighted content creators in accessibility practices, since they are ultimately responsible for generating descriptions at scale.
Tags: social media accessibility · alternative text · image descriptions · crowdsourcing · assistive technology · user empowerment · automation
Standards referenced: WCAG