WCAG 2.2 AA Pre-Publish Gate: Verify AI Tutorials Before You Commit

This guide shows how to treat accessibility as an operational capability for AI-generated tutorials.

It provides a verify-before-you-commit workflow for enforcing a WCAG 2.2 AA pre-publish gate and comparing complete options.

How It Works

The gate works like any other CI check: no tutorial ships until an automated WCAG 2.2 AA scan passes and a human spot-check confirms it. The moment an AI-generated tutorial is drafted, it enters a staging state—published nowhere, linked from nothing. A checker runs against the staged version, not your local copy, because what you preview and what the reader loads can differ. Only when the live, complete staged page passes does the commit go through.

Key terms first. WCAG 2.2 AA is the current Web Content Accessibility Guidelines conformance level most organizations treat as their baseline standard. A pre-publish gate is a mandatory checkpoint between drafting and deployment—nothing moves forward until it clears. Remediation, as document-accessibility vendors like Corpowid use the term, means fixing accessibility failures in existing content rather than rebuilding from scratch; the gate exists so you remediate drafts, not live pages. And "operationally accessible," a phrase used in recent industry writing on the gap between system design and system performance, means the content works in practice for real users—not just that a checkmark exists somewhere.

The mechanism runs in three passes. First, an automated scan of the staged page flags mechanical failures: missing alternative text, insufficient contrast, unlabeled form fields, missing headings. Automated tools catch these reliably. Second, a keyboard-only walk-through: unplug the mouse, navigate the entire tutorial top to bottom, and confirm every interactive element is reachable and operable. Third, a screen-reader spot check on the core steps of the tutorial—does the assistive technology announce the steps in a usable order? Vendors like Vispero, in their guidance on kiosk accessibility, emphasize this same layered approach: automated testing plus real assistive-technology verification, because a page can pass one layer and fail another.

The "available does not mean accessible" principle, documented in seafarer-welfare reporting by Safety4Sea, is the failure mode this gate exists to prevent. A tutorial can be live, findable, and complete—and still unusable with a screen reader. Availability is a publishing fact; accessibility is a verified property. The gate treats only the second one as shippable.

Two rules make the gate operational rather than aspirational. Rule one: verify the staged, complete version—every image loaded, every code block rendered—because partial pages produce partial results. Rule two: like-for-like comparison applies to the fix loop, too. When a check fails and you revise, rerun the same checker against the same full page, so the before-and-after is a genuine comparison instead of two different measurements. Adopting improvements into your regular publishing workflow, the way SSA Marine Canada's accessibility plan folds enhancements into routine operational planning, is what keeps the gate durable after launch week enthusiasm fades.

The output is binary by design. Pass means commit and publish. Fail means the draft returns to remediation with the specific failures attached. What the gate removes is the ambiguous middle state—a tutorial that is "probably fine"—which is where accessibility debt actually lives.

Key Factors to Consider

Three criteria should drive your pre-publish gate, and they are not the ones most teams default to. First, conformance coverage: does the check map to WCAG 2.2 AA success criteria across the tutorial's actual content types — code blocks, screenshots, embedded video, interactive demos — rather than a single page-level score? Second, evidence durability: does the gate produce a dated, versioned record tied to the tutorial revision, so a later edit cannot silently invalidate the pass? Third, human confirmation: does a named reviewer sign off on the items automation cannot judge, such as whether alt text describes the instructional intent rather than the pixels? A gate that satisfies only the first criterion is a scanner, not an operational capability.

The numbers that matter are counts and ratios, not vendor marketing figures. Count the success criteria your scan actually evaluates against the full WCAG 2.2 AA set, and treat any gap as an explicit, documented exclusion rather than an implied pass. Count the tutorial's distinct content types and confirm each has a corresponding check — Vispero's guidance on restaurant self-order kiosks, for example, separates screen reader support, physical requirements, and testing into distinct workstreams, which is the right granularity for a tutorial containing mixed media. Track the ratio of automated findings to human-confirmed findings over time; a ratio that never moves suggests your reviewer is rubber-stamping.

Timing is a number too. SSA Marine of BC's accessibility plan commits to introducing low-cost accessibility enhancements, including accessible seating in key areas, by December 2028, and to folding accessibility improvements into regular maintenance, upgrades, and operational planning processes. That second clause is the transferable one: accessibility work scheduled inside existing maintenance cycles survives; accessibility work scheduled as a separate project does not. Put your gate on the same cadence as your content releases.

Use the table below as a pre-commit checklist. Every row must have a yes before the tutorial leaves staging.

Compare like-for-like before you commit. A vendor demo that scans a clean sample page tells you nothing about your tutorial with twelve screenshots and an embedded player. Ask for the criteria list, the content-type mapping, and a sample evidence record from a real revision — then total them against your own inventory. If the totals do not match your content, the gate is not ready.

Common Mistakes

Most teams that build a WCAG 2.2 AA pre-publish gate stumble on the same pitfalls, and both are avoidable if you know what to look for before you commit a tutorial to the published state.

The first pitfall is testing a partial draft. An AI-generated tutorial often ships its code blocks and body text first while screenshots, alt text, and captions arrive later — and teams run their automated scan on the finished parts only. The scan passes, the tutorial goes live, and then the late-added images land with no alt text at all. The rule: run the gate only on the complete staging version, every asset included. Vispero's guidance on kiosk accessibility makes the same point in a physical context — a screen reader path that works for the menu but not the payment step is not an accessible experience. Verify the live, complete option before committing.

The second pitfall is trusting an automated pass as the whole verdict. Automated checkers — including the browser-based and Figma-embedded WCAG tools noted by Corpowid — catch many failures but cannot judge whether alt text actually describes the image or whether heading order makes sense for a screen reader user. Teams that treat a green scan as full conformance publish tutorials a human spot-check would have flagged. The fix is mechanical: the human reviewer opens the staged tutorial with a screen reader, walks the full reading order, and confirms the automated findings before approving the commit.

A smaller but frequent variant: teams scan the rendered page but forget the downloadable artifacts, such as the PDF version of the tutorial. Document accessibility remediation is its own discipline precisely because a page can pass while its exported document fails. If the tutorial ships in more than one format, each format enters the gate separately.

Both pitfalls share a root cause — treating the gate as a final inspection rather than a condition of publication. Available does not mean accessible, as Safety4Sea put it when describing welfare systems that exist but exclude their intended users; a tutorial that exists on your site but blocks screen reader users has the same gap. Make the check part of the commit itself, and both failure modes disappear.

Insider Tactics

Run the gate against the rendered tutorial, not the prompt or markdown source. A browser-based checker can catch issues in the live page, while document remediation is a separate pass for any downloadable file; Corpowid.ai lists these as distinct aids, so treat a pass in one surface as evidence only for that surface. If a tutorial includes both a web page and a downloadable file, verify each independently before either is approved.

Move the automated scan to the moment the draft is created, before editorial review. That timing keeps accessibility failures from consuming reviewer attention and makes the fix part of normal drafting rather than a late release blocker. It also prevents a polished but inaccessible draft from becoming the baseline everyone compares against.

Ask the generator to produce an accessibility checklist with the tutorial, then verify each item against the rendered result. The checklist is a prompt for inspection, not proof of conformance, because the model can describe an accessible pattern while the rendered output still fails it. Keep the checklist attached to the draft so reviewers can see what was claimed and what was checked.

Schedule the human spot-check in the same staging window as the automated pass, while the tutorial is still in staging. Fixing a screen-reader or keyboard issue then is less disruptive than reopening a published page, and it keeps the two checks from drifting apart. If the spot-check finds a new issue, rerun the automated scan after the fix rather than assuming the original pass still holds.

Account for operating-system differences when you choose test setups. SNOW notes that each operating system differs in what accessibility features it offers, so a tutorial that works with one screen reader or keyboard configuration may not cover another. Record which setup you tested so the next reviewer can compare like-for-like instead of repeating the same narrow check.

When two AI-generated tutorials both pass, compare them criterion by criterion and surface by surface rather than by a single pass/fail badge. A like-for-like total prevents a tutorial with fewer tested surfaces from looking safer than one with broader coverage. Verify the live, complete option before committing, including any linked files or embedded examples.

Comparison

Compare the three gate shapes on the same tutorial, in the same staging state, and against the same acceptance rule. The comparison below uses decision terms, not invented prices or turnaround times; if a vendor quote matters, verify it against your own measured queue before committing.

The scan-only option wins on throughput, but its win is narrow: it fits when tutorial risk sits in text the scanner can parse and the cost of a false pass is low. Once screenshots or code samples explain the task, the same speed becomes a liability because the scan can report a clean result while the human path remains broken.

The combined option is the default for mixed-content tutorials because it keeps machine coverage and adds a human check on the exact artifact that will ship. Its failure mode is organizational, not technical: if the spot-check has no named owner, no time box, and no return path to the draft, the gate quietly reverts to scan-only behavior.

The manual option wins when an accessibility defect would create legal, safety, or trust exposure larger than the delay. It loses when teams use it as the default for every tutorial, because the queue then dictates the publishing calendar and authors start bypassing the gate.

Choose by comparing like-for-like totals: count the same tutorial types, the same rework loops, and the same reviewer minutes across all three options. The winner is the option with the lowest total cost of a defect that reaches readers, not the lowest cost of running the check.

Frequently Asked Questions

What two checks must pass before an AI-generated tutorial ships?

No tutorial ships until an automated WCAG 2.2 AA scan passes and a human spot-check confirms it.

What state does a newly drafted AI tutorial enter?

The moment an AI-generated tutorial is drafted, it enters a staging state—published nowhere, linked from nothing.

Why must the checker run against the staged version rather than a local copy?

A checker runs against the staged version, not your local copy, because what you preview and what the reader loads can differ.

When does the commit go through?

Only when the live, complete staged page passes does the commit go through.

Which business systems does Oli Sistemas include in operational accessibility?

Oli Sistemas frames ERP, POS, and sales-channel inclusion as operational accessibility.

How many Operating System Accessibility entries does SNOW list?

SNOW lists 5861 Operating System Accessibility entries, and each operating system differs in offered features.

Quick answers

Also worth reading: Accessibility Is an Operational Capability, Not a Feature: Accessibility Is an Operational Capability, · Leveraging YouTube Video Content in AI Tutorials: Leveraging YouTube Video Content in · AI-Powered Personalized Tutorials: Implications for Promotional Marketing: AI-Powered Personalized Tutorials: Implications for