Why accessibility work stalls, and where AI fits
Under the Equality Act 2010, UK businesses must make reasonable adjustments so disabled people are not put at a substantial disadvantage, and in practice the WCAG 2.2 guidelines are the benchmark websites get measured against. Most small firms do not fail this on principle; they fail on volume. Three hundred product images without alt text, two years of video without captions and a returns policy written like a statute are a workload problem, not an attitude problem.
AI is genuinely good at exactly that kind of volume: producing first drafts of image descriptions, caption files and simplified copy in minutes rather than weeks. The working rule for everything below is the same: AI drafts, a human signs off. Raw machine output is not accessible output, and publishing it unchecked can be worse than publishing nothing, because it looks finished.
Alt text: fast drafts, careful edits
Claude and ChatGPT both describe uploaded images well, Microsoft 365 and Edge generate automatic alt text, and several WordPress plugins can batch-draft descriptions across an entire media library. That turns a week of tedium into an afternoon of editing.
Quality checks before you publish
- Rewrite for context: the same product photo needs different alt text on a product page ("black leather ankle boot with side zip") and in a blog post about your workshop
- Keep it concise, ideally under roughly 125 characters; screen reader users do not want an essay per image
- Never start with "image of" or "picture of": screen readers already announce that it is an image
- Mark purely decorative images with empty alt text (alt="") so they are skipped rather than described
- If an image contains important text, reproduce that text exactly, or better, move it out of the image entirely
Need a hand with this?
Our team delivers Accessibility & Compliance for UK businesses — with a free initial consultation, transparent fixed quotes and no lock-in contracts. Tell us what you're working on →
Captions and transcripts
Automatic speech recognition is now strong enough that YouTube's auto-captions, Whisper-based transcription tools and the built-in live captions in Teams and Zoom all produce a usable first pass. Descript is particularly practical for marketing teams because the transcript and the video edit are the same document: fix the words and the captions follow.
Quality checks before you publish
- Correct proper nouns, product names and industry jargon; these are exactly what auto-captioning gets wrong and exactly what your audience notices
- Add punctuation and sensible line breaks; a wall of lowercase text is technically a caption and practically useless
- Label speakers in interviews and panel recordings
- Watch the video once with the sound off; if the captions alone do not carry the meaning, they are not finished
- Publish the corrected transcript on the page too: it serves deaf users, skim-readers and your search rankings at the same time
Plain-English rewrites
Dense copy is an accessibility issue for people with dyslexia, people with cognitive disabilities and anyone reading in a second language. LLMs are strong rewriters when the instruction is specific: "Rewrite this returns policy at a UK reading age of about nine. Keep every condition and deadline intact, use short sentences and bullet points, and flag anything you were unsure how to simplify." A tool such as Hemingway Editor then grades the result, so you can verify readability actually improved rather than assuming it.
Quality checks before you publish
- Compare against the original line by line wherever wording carries legal weight: refund terms, guarantees, safety instructions
- Keep necessary technical terms, but define them on first use rather than deleting them
- Check structure as well as sentences: front-load the key point, use descriptive headings, break up long pages
- Test with one real reader from outside your business; five minutes watching them is worth an hour of guessing
What AI will not fix
Be sceptical of overlay widgets that promise instant compliance from one line of JavaScript. They do not repair the underlying code, they are widely criticised by the accessibility community, and disabled users regularly report that they make sites harder to use, not easier. Installing one does not discharge your Equality Act obligations either.
Keyboard navigation, focus order, colour contrast, form labels and heading structure are build-quality issues. AI coding assistants can help a developer fix them faster, but they must actually be fixed in the code. If your site fails a basic keyboard-only test, no amount of beautifully generated alt text compensates for a checkout nobody can tab through.
Key Takeaway
Let AI produce the first draft of every alt text, caption file and plain-English rewrite, then apply a human quality pass before publishing: rewrite alt text for page context, correct names and punctuation in captions, and confirm simplified copy still says what the original meant. Avoid overlay widgets that promise instant compliance; structural fixes such as keyboard navigation, focus order and colour contrast still need proper design and development work.
Make it routine, not a rescue project
- Draft alt text at the image upload step, not in a quarterly clean-up panic
- Caption every new video before publishing; retrofit the back catalogue in small monthly batches
- Run new page copy through a readability check as part of normal sign-off
- Schedule a quarterly keyboard-and-screen-reader spot check on your most important user journeys
An hour a week keeps you ahead of a problem that gets expensive when a complaint, a lost customer or a legal letter forces the issue. If you would like an accessibility review of your site with a prioritised fix list, our team can help.