AI fix suggestions
Zylyn can generate corrected markup for each finding, based on your page's actual elements rather than generic advice.
How to generate them#
Open a report and click Generate AI fix suggestions.
Cost: 2 credits per finding. Suggestions are cached, so findings that have been generated before are free — the button tells you before you click, e.g. "2 of 3 already available at no cost", and charges only for the rest.
Generation takes under a minute. Suggestions appear inside each finding as they complete.
The four statuses#
This is the part to understand, because the status tells you how much you can trust the suggestion.
| Status | What it means | What to do |
|---|---|---|
| Applied / ready | A safe, complete fix was produced and self-validated | Review, then apply |
| Needs your input | Complete except for a [PLACEHOLDER] | Replace the placeholder with your own wording |
| Dev required | No safe automatic patch exists; options are listed | Have a developer decide |
| Decide | Automated validation could not confirm the change | Review manually before applying |
A worked example of Needs your input — an ARIA progressbar with no accessible name:
Zylyn cannot know what your progress bar measures. It writes the correct attribute and leaves the label to you. Do not ship the placeholder.
A worked example of Dev required — a contrast failure:
The suggestion is specific and correct, but the colour is a shared design token — changing it affects every element using it. That is a decision, not a patch.
Always review before applying#
The portal says this in three places, and it means it. These are AI-generated suggestions, not verified patches. Specifically:
- A suggestion can be technically valid but wrong for your design (the contrast example above)
- Adding an
aria-labelwith the wrong wording is worse than no label — it misleads screen reader users confidently - Changing shared tokens or components has effects beyond the page you scanned
Treat a suggestion as a well-informed pull request from a contributor who has only seen one page of your site. Read it, understand it, then apply it.
What they're good for#
- Turning findings into something actionable for someone who isn't an accessibility specialist
- Speed on mechanical fixes — alt text, labels, ARIA attributes, heading levels
- Teaching. Reading the suggestions is a fast way to learn why something failed
What they're not good for#
- Structural problems (landmarks, focus order, page architecture)
- Anything touching shared components or design tokens
- Replacing manual testing — a fix that satisfies the scanner can still be unusable with a screen reader