AI enhancements explained: how AI improves your WCAGScan report
What the AI adds to a scan report, what it is never allowed to do, and how to use its suggestions safely.
The short version
| Who finds the problems? | The rule-based scanner. Your score, grade and issue list come from the scanner alone |
|---|---|
| What does AI add? | Two things: a suggested fix for the most important issues, and a review of wording that rules cannot judge |
| Where is it? | In the “AI Suggests” tab of the interactive report and in an appendix to the PDF report |
| Can AI change my site? | No. It has no access to your website. A person reviews and applies every change |
| Can AI be wrong? | Yes. Treat every suggestion as a draft to review and test, not as a finished fix |
The problem AI solves
A traditional accessibility report tells you what failed: “Text contrast 2.54:1 is below the 4.5:1 minimum” on 14 pages. That is accurate, but it leaves the hard part to you. What color should the text be? Which line of code do I change? How do I know the fix worked?
WCAGScan’s AI enhancements answer those questions inside the report. The scanner does the testing. AI does the explaining.
Part 1: Suggested fixes
After the scan finishes, WCAGScan groups identical problems together. A button with the same fault on 40 pages is one group, because it is almost always one template to fix. The highest-priority groups, Level A first, then Level AA, most frequent first, are sent to the AI with the scanner’s finding and the snippet of code that failed.
For each one the report shows:
- What is wrong, in plain English. Which element fails, which WCAG success criterion it fails, and who is affected.
- Corrected code. Your original snippet next to a suggested version, with a Copy button. The suggestion keeps your tags and attributes and changes only what the fix requires.
- How to check it. A note on how to verify the fix, or when a different technique would be better.
- Brackets where only you know the answer. AI cannot see what a photo shows or what your organization is called, so it writes [describe the photo] instead of inventing words.
Part 2: Wording review
Some barriers are about meaning, which rule-based checks cannot judge. A rule can tell that an image has alt text. It cannot tell that the alt text “IMG_2041.jpg” is useless. AI reads the titles, headings, link text and image descriptions on a sample of your pages and flags:
- 2.4.2 Page Titled: titles that are vague or the same on many pages
- 2.4.6 Headings and Labels: headings that say nothing about their section
- 2.4.4 Link Purpose: links such as “Go”, “Click here” or a bare web address
- 1.1.1 Non-text Content: alt text that is a file name or just “image”
Each finding quotes the text, explains the problem and offers better wording.
How this improves the report
- From a list of faults to a list of fixes. A web developer can act on the report the day it arrives.
- Readable by non-specialists. A principal, clerk or business owner can understand what is wrong without knowing WCAG.
- Fix once, not 40 times. Grouping shows which problems come from a shared template.
- The right order. Level A problems, the ones that block people completely, come first.
- Catches what rules miss. The wording review covers criteria that normally need a person.
What AI is never allowed to do
- Change your score, grade or issue list. Those come only from the rule-based scan, so two scans of the same page give the same score.
- Touch your website. It has no access to your site, your browser or your passwords.
- Recommend overlays. It never suggests scripts or widgets that alter your pages for visitors.
- Claim compliance. A suggestion is not a certificate and is not legal advice.
How WCAGScan checks the AI’s work
AI can make mistakes, so WCAGScan checks what it can before a suggestion reaches your report:
- Colors are re-measured. When AI proposes new colors for a contrast problem, WCAGScan recalculates the contrast ratio with the same math the scanner uses and tells you whether the suggestion passes.
- Quotes are verified. A wording finding is dropped unless the text it quotes really appears on that page.
- Native HTML first. Suggestions prefer standard HTML over ARIA, the approach WCAG guidance recommends.
Limits you should know
AI suggestions are advisory. Review each one, apply it in a test copy of your site where you can, and re-scan to confirm the issue is gone.
- It is a sample, not everything. Fix suggestions cover the highest-priority issue groups, and the wording review covers a sample of pages. On a large site with many different problems, some issues will not have a suggestion. The full issue list is always in the main report.
- It sees a snippet, not your whole site. A suggestion may need adjusting to fit your templates or style sheets.
- It cannot test. Only a re-scan, and for some criteria a person, can confirm a fix worked.
Your data
AI sees only what the scan already collected from your public pages: the failing code snippets, the scanner’s findings, and each reviewed page’s title, headings, link text and alt text. Nothing else from your site is sent, and login details are never shared.
How to use the AI section
- Open the interactive report and select the “AI Suggests” tab.
- Start at the top. The list is already in priority order.
- Give each suggestion to whoever maintains the site. Copy the corrected code and fill in anything in [brackets].
- Fix the template, not each page.
- Run a new scan to confirm the issue count has dropped.
Where WCAGScan helps
AI suggestions sit on top of the Deep Scan engine, which tests every page in a real browser; see the guide to Deep Scan. For examples of the AI output, see the AI features page. Run a free scan or see pricing for what each plan includes.