Publisher media library: stopping alt drift across a newsroom CMS

High-volume publishers ship dozens of images a day. Without a written standard, alt text becomes whatever the CMS field happens to contain — captions pasted wholesale, filenames, or blanks. Accessibility leads get questionnaire pressure; SEO leads notice thin Image search presence on explainers that should rank for years.

This composite newsroom ran WordPress with a central desk, freelance contributors, and a photo desk that uploaded crops faster than anyone reviewed metadata. New publishes averaged roughly 40% useful alt on informative featured images — and that number was a generous sample, because many chron posts used decorative crops that had been keyword-stuffed instead of left empty.

The first two weeks focused on stopping new debt. A one-page standard clarified informative vs decorative, banned filename alt, and required a human glance at AI drafts before save. Producers saw the review queue as part of the publish checklist — not a separate “accessibility project” they could defer to next quarter.

The photo desk agreed that chron templates with purely decorative lead crops would use empty alt by default. That single policy removed a recurring source of keyword stuffing that had been making coverage metrics look better than they were.

Once standards existed, the desk enabled OpptiAI’s generate → review → save loop on new featured images. AI drafts could see the crop and headline context; producers edited names, locations, and action verbs before save. Average approval time settled around 20–40 seconds per image — fast enough for a busy desk, slow enough to catch wrong attributions.

Reuse was the harder edge case. When the same wire photo appeared in three stories, alt text had to match each URL’s angle — not copy the first draft forever. The standard said: re-check alt when re-attaching an asset to a new post, even if the file already lived in the library.

Only after new-publish coverage stabilised did the team bulk-remediate money-section archives. Homepage heroes, section hubs, and evergreen explainers came first — roughly 2,100 images in scope. The chron firehose of same-day posts stayed out of Phase 1; trying to perfect ten years of breaking news would have buried the desk.

Bulk jobs ran in batches of a few hundred with the same review gate. Editors prioritised explainers that still earned organic traffic and Image search impressions, not posts that had already fallen off the index.

The SEO lead sampled 50 new publishes per week across top sections — not “every attachment ever uploaded.” Useful alt meant human-readable, specific to the image, and empty only when decorative. Filename strings and caption paste counted as failures.

By week six, sampled new-publish coverage sat near 88%. Archive money-section useful alt moved materially on re-audit, though the team did not claim a site-wide percentage that included a decade of chron posts.

Image search traffic did not triple because explainers got better alt text. Some evergreen URLs saw cleaner Image search snippets over time, but breaking news remained a different game — freshness and relevance dominate, not alt strings alone.

The newsroom also did not remediate every historical attachment. That was intentional. Governance plus money-section cleanup closed the gaps procurement and public audits would actually see, without pretending a six-week project fixed a fifteen-year library.

If you run a WordPress publisher, mirror the sequence: write standards, fix new publishes first, bulk-clean money sections, sample weekly so drift is visible. The publishers playbook goes deeper on roles, audit cadence, and reuse rules.

Pair operational work with the WCAG image guide for decorative policy language — and use free public-page audits on section hubs before anyone argues the gap is “just a few images.”