ASM built AI-generated alt text into its XML proofing workflow, working backwards from publication so authors can review and edit image descriptions late in the process (where compliance is realistic) rather than being instructed up front. After a year collecting author corrections as a check on quality, ASM found AI generated alt text as good as or better — at far greater accuracy — than its subject-matter-expert writers, and rolled it in. David Haber has since prototyped an audio-proofing tool (built in ~22 minutes with Claude) that lets authors proof alt text by ear using a 'sound XML' markup — e.g. forcing 'Mb' to be read as 'megabases' or inserting pauses — treating alt text as a listening format and a step toward multimodal publication.
Benefits & outcomes
AI alt text measured more accurate than SME-written, at lower cost.
Late-workflow editing plus correction-data collection improved quality and surfaced where authors misunderstood the purpose.
Audio proofing reframes alt text from a compliance chore into a consumable product authors want to get right.
Notes & quotes
Presented by David Haber (Publishing Operations Director, ASM) at PurePub.ai's "Doing things differently" panel (20 May 2026). Prototyping tool: Claude. Illustrates the conference's recurring 'vibe coding as a narrative tool' theme — building a working proof of concept to scope with a vendor, rather than writing lengthy requirements docs.
Publisher / company: American Society for Microbiology (ASM)
Date of mention: 20/05/2026
Source: https://journals.asm.org/