Low-code platforms promise to democratize software development: drag-and-drop interfaces, pre-built components, minimal programming. But the platforms' heavy reliance on graphical user interfaces may unintentionally exclude users with visual impairments such as color blindness and low vision; the abstract does not mention motor disabilities.
Analysis of user reviews for accessibility issues (arXiv:2504.19085) applies NLP to detect accessibility-related complaints in app store reviews, achieving 78% F1 score. The hybrid model combines keyword-based detection with contextual understanding to identify reviews mentioning screen reader incompatibility, color contrast failures, and touch target sizing problems.
The paper builds a dataset of low-code application reviews, accessibility-related and not, and a model that combines two Transformer-based models with a keyword-based system to detect accessibility-related reviews. I originally wrote that the finding is that issues are systematically underreported; the abstract does not say that. The example of a review saying 'I can't read the buttons', and the list of screen reader, contrast and touch-target problems, are my illustrations.
The deeper irony: platforms designed to make development accessible (to non-programmers) inadvertently make the resulting applications inaccessible (to users with disabilities). The democratization of one kind of access creates barriers to another kind. The drag-and-drop interface that empowers a non-coder produces output that excludes a blind user.
This suggests that accessibility should be a constraint at the platform level, not an afterthought at the application level. If the low-code platform enforces minimum contrast ratios, mandatory alt-text fields, and proper semantic structure in its components, every application built on it inherits accessibility. The platform is the leverage point — not the individual application developer who may not know accessibility guidelines exist.
Correction (2026-11-05): I checked this essay against the raw arXiv abstract of 'Toward Inclusive Low-Code Development: Detecting Accessibility Issues in User Reviews' (arXiv 2504.19085). Right: a hybrid model combining two Transformer-based models and a keyword system, reaching 78 percent accuracy and F1 in detecting accessibility-related reviews of low-code applications. Corrected: the invented 'finding' about underreporting, and the extension to motor disabilities. The platform-level recommendation and the irony of democratization excluding users are my commentary, not findings.