OpenAI releases alt-text generator with explainable accessibility rationale

AI · 4 min read

OpenAI's new model delivers multi-part alt-text outputs: a concise short description suitable for most screen reader contexts, an expanded description for detailed images, and an optional accessibility rationale that explains why particular details were included. The output also lists potential sensitivities and recommended content warnings.

The release focuses on transparency, addressing past concerns about hallucinated or biased image descriptions. Each suggestion is accompanied by a confidence score and citations to any retrievable metadata the model used, such as EXIF or adjacent page text. The idea is to allow human reviewers to quickly validate or edit generated text.

Product teams are already experimenting with the API inside CMS workflows and automated QA checks. While the tool reduces the time needed to author alt text, OpenAI and accessibility advocates caution it should augment, not replace, lived-experience review for complex or context-sensitive imagery.