I'm building a CMS that won't let you publish an inaccessible page
Windrow is a website builder with an automated publish gate. If a page fails WCAG 2.2 Level AA, it does not go live. Not a warning banner, not a score you can dismiss, a gate. I am building it because twenty years and 300-plus audits taught me that education by itself has never once fixed this.
There are 230 days until the April 26, 2027 Title II deadline. I am not going to pretend Windrow solves that for anyone this year. I am going to explain why I think the gate is the right idea anyway.
Why build another website builder?
Because I keep writing the same finding. You can force people to take required training before gaining access to the CMS. The training can point out the correct way of doing things. But the users still want to "design" things themselves.
I've concluded that editors aren't lazy and they are not indifferent. They are busy and they have opinions. They want to "design" the organization's website like it's their own. They are a communications coordinator with four other jobs, publishing at 4:50 on a Friday, and the tool in front of them says the page is fine.
The tool lets them fail. So they fail. Then we run an audit, hand them a spreadsheet, and call it a training problem.
It was never a training problem. It was a default settings problem.
What does the publish gate actually block?
The gate runs on every publish attempt. If a blocking check fails, the button does not work, and the editor gets a plain-language explanation and a link to the exact element.
| Blocked | Warned but publishable | Cannot be checked by any tool |
|---|---|---|
| Missing alt attribute on a content image | Alt text that looks auto-generated | Whether the alt text is true |
| Insufficient color contrast | Very long link text | Whether a heading describes the section |
| Skipped heading level | Table with many columns | Whether a task is completable by keyboard end to end |
| Form field with no programmatic label | Ambiguous link text like "read more" | Whether captions match the audio |
| Missing page language or title | Deeply nested lists | Whether the page makes sense out loud |
The gate catches roughly the same third that any automated checker catches. It does not make a site conformant. It makes the machine-checkable third impossible to get wrong, permanently, without anyone remembering to care.
That is a smaller claim than most accessibility products make. But it's the truth.
What are the other decisions?
Six, and each one is a place where I chose the constrained option.
- Contrast-safe palettes only. You pick a theme. Every foreground and background pair in it already passes. You cannot build a failing combination, because the failing combinations are not offered.
- AI alt text with required human verification. The model drafts. A person must read it and click confirm before publish. The draft saves time. The confirmation preserves the step that matters.
- Theme switching without content loss. Change the entire look without touching a word. Content and presentation stay separate, which is also what makes the accessibility guarantees hold across themes.
- Static HTML export. Your site, your files, whenever you want, in a format any host can serve. If you cannot leave a tool easily, you were never really using it.
- Auto-updating accessibility statement. Generated from your actual last scan, with the date and what was checked. Most published accessibility statements are aspirational fiction. This one is a receipt.
- WCAG 2.2 AA, not 2.1. The legal floor for Title II is 2.1 AA. I am building to 2.2 because target size, focus appearance, and dragging alternatives are real problems for real people, and because the floor tends to rise.
Pricing, so nobody has to email me and ask: free tier, $19 a month for an individual, $29 for nonprofit and education, $49 for an organization, $199 for an agency. The nonprofit and education tier is priced the way it is on purpose, since those are exactly the people staring at 2027 with no budget.
What am I probably getting wrong?
Three things keep me up, and I would rather hear about them now than after launch.
- The gate will make some people angry. There is a version of this where a communications director hits a hard block at 4:50 on a Friday, cannot publish an urgent notice, and hates me forever. I have gone back and forth on an override with an audit trail. Right now I lean no, because every override I have ever seen become available has become routine within a quarter.
- AI alt text may make things worse before better. A confirm button is a very thin barrier against clicking confirm. I may need friction there that people will resent.
- Small tools might be the wrong shape for institutions. Windrow is built so one person can run a site without me. Large organizations want federation, roles, workflow, and audit logs. Those are real needs and they are also how simple products become the platforms I have spent twenty years complaining about.
I do not have clean answers to any of the three. I would rather say that in public in September than discover it in a support inbox next spring.
If you build or manage websites for a school, a city, a clinic, or a nonprofit, I want one answer from you: what would actually make you switch? Not what would make you curious. What would make you move a live site. Join the Windrow mailing list and tell me. I read all of it, and I will write about what I hear.
Key points
- Windrow is an accessibility-first website builder whose publish gate blocks any page that fails machine-checkable WCAG 2.2 AA criteria, rather than warning and letting it through.
- The gate exists because inaccessible pages are a defaults problem, not a training problem: busy editors fail because the tool permits it.
- Automated checking catches roughly a third of WCAG issues, so the gate guarantees that third and makes no claim about the rest.
- Six supporting decisions: contrast-safe palettes, AI alt text with required human confirmation, theme switching without content loss, static HTML export, an auto-generated accessibility statement, and building to 2.2 AA rather than the 2.1 AA legal floor.
- Open questions I am asking readers about: whether the gate needs an override, whether AI alt text needs more friction, and whether a deliberately small tool can serve institutions without becoming a platform.
Let me know what you think!
See you next Monday.