Synthetic Media, Real Accessibility Gaps: How to Audit AI-Generated Content for WCAG 2.2 Compliance
As brands flood their sites with AI-generated video, images, and audio, they're quietly accumulating serious accessibility debt. Here's a four-step framework for auditing and fixing synthetic media compliance, built from real client work at Mintec.
In our last two projects involving AI-generated media — an e-learning platform with synthetic video lessons and a retailer with AI-produced product catalogs — we found the same pattern: gorgeous content, terrible accessibility. AI-generated videos without captions. AI-produced images with empty alt attributes. Synthetic audio with no transcript.
Both teams were stunned. They'd invested heavily in AI content pipelines but never once considered what WCAG 2.2 had to say about AI-generated media. And they're not alone.
The Three-Way Collision
Three trends are converging in 2026 and creating a compliance blind spot:
Synthetic media is everywhere. The market hit $5B in 2024 and projections place it at $48B by 2033. AI-generated video now accounts for 34% of all synthetic content, per TechRT. Every week another brand rolls out AI-generated product videos, synthetic spokespeople, or AI-created imagery.
WCAG 2.2 is the baseline. The Web Content Accessibility Guidelines 2.2 became the de facto compliance standard in 2025-2026. Success criteria like 1.2.2 (Captions - Prerecorded), 1.1.1 (Non-text Content), and 1.2.1 (Audio-only and Video-only - Prerecorded) apply to every piece of synthetic media on a public website.
Accessibility lawsuits are accelerating. WebAIM reports a 300% increase in accessibility-related lawsuits between 2022 and 2026. The risk isn't theoretical — it's a growing liability for every brand publishing AI-generated content at scale.
Most teams treat these as separate concerns. They're not. Every AI-generated video without captions is a WCAG 1.2.2 failure. Every AI-generated image without descriptive alt text is a WCAG 1.1.1 failure. Every synthetic podcast episode without a transcript is a WCAG 1.2.1 failure.
Where Synthetic Media Breaks Accessibility: A Framework
Not all synthetic media violates accessibility in the same way. The failure pattern depends on the content type. Here's our classification from client work:
| Content Type | WCAG Criterion | Typical Failure | Why AI Makes It Worse |
|---|---|---|---|
| AI-generated video (synthetic spokespeople, animated explainers) | 1.2.2 Captions (Prerecorded), 1.2.5 Audio Description | No captions or auto-generated captions with <95% accuracy | AI video pipelines focus on visual quality — accessibility metadata is an afterthought |
| AI-generated images (product photos, illustrations, hero images) | 1.1.1 Non-text Content | Empty alt attribute or alt="AI-generated image" — not descriptive | Volume: an AI catalog pipeline can output 1,000+ images per day, each needing unique alt text |
| AI-generated audio (synthetic voiceovers, podcasts, narrations) | 1.2.1 Audio-only and Video-only | No transcript provided alongside the audio | Audio generation tools rarely produce transcripts by default |
| AI-generated media with dynamic content (real-time AI avatars, personalized video) | 2.2.2 Pause, Stop, Hide, 3.2.2 On Input | Moving content that users can't pause or control | Dynamic AI content inherits accessibility issues from both the media and the interactive layer |
The pattern is consistent: AI content platforms optimize for visual fidelity, rendering speed, and cost per asset. Accessibility metadata (captions, alt text, transcripts) ships as an optional feature rather than a requirement.
The Four-Step Audit Framework
Over the last 12 months, we've refined a process for auditing synthetic media accessibility. It's not complicated, but it's systematic.
Step 1: Inventory Every AI-Generated Asset
You can't fix what you haven't found. Most teams don't know how much of their site content is AI-generated. Start with a crawl:
- Use
axe-coreorLighthouseto flag images with missing or empty alt text — these are your synthetic image candidates - Review your media library for videos generated through AI tools (Synthesia, HeyGen, ElevenLabs, etc.)
- Audit your CDN and CMS for auto-generated audio files
In our e-learning client's case, we found 847 AI-generated video lessons — none with captions. The content team had no idea the AI tool produced video without subtitle tracks by default.
Step 2: Classify by WCAG Criterion
Map each asset type to the relevant WCAG 2.2 success criteria. Use the table above as your classification matrix. Prioritize by:
- Legal risk: Video content without captions (SC 1.2.2) carries higher litigation exposure than decorative images with null alt attributes
- User impact: Content that blocks core user tasks (product demos without audio descriptions, checkout images without alt text) gets priority over marketing banners
Step 3: Fix with the Right Tool for Each Type
One-size-fits-all accessibility fixes don't work for synthetic media. AI content needs AI-powered remediation:
- AI video → captions: Run synthetic videos through Whisper or Deepgram for automatic caption generation. Target >98% accuracy, then human-review the first 10 seconds and the last 10 seconds of every video — those are where AI transcription most often fails (intro music, credits, multiple speakers).
- AI images → alt text: Use vision-language models (GPT-4o, Claude 3.5, Gemini 2.5) to batch-generate alt text. Our workflow generates 5 alt text candidates per image in parallel, then selects the most descriptive one. Human-review the top 20% of images (heroes, product shots, instructional diagrams).
- AI audio → transcripts: Most AI voice generators can output SSML or word-level timestamps. Extract these as a base transcript and format as a web-accessible HTML document.
Step 4: Automate Compliance into the Pipeline
The most important step. If you're only auditing after publication, you're already behind. Synthetic media pipelines produce at a velocity that breaks manual QA.
Build accessibility checks into your content pipeline:
- After the AI content generator outputs an asset, run an automated accessibility check before it enters the CMS
- Reject assets that fail WCAG checks (no captions, no alt text, no transcript) with a clear error message to the content team
- Track compliance rate as a KPI — aim for 100% of synthetic media assets passing automated checks
We implemented this for a real estate client that uses AI to generate virtual tour videos. The pipeline now auto-generates captions and audio descriptions before any video hits the CDN. Compliance went from 12% to 100% in two weeks.
What We've Learned from Client Implementations
Some patterns we've seen repeat across projects:
The volume trap is real. A retailer using AI-generated product imagery told us "we'll write alt text for the important ones." Their "important" list grew from 50 images to 2,000 in three months. Batch automation isn't optional — it's the only way to keep up.
AI can fix AI's accessibility problems — but needs guardrails. The same models that generate synthetic media can also generate captions, alt text, and transcripts. The key is building the accessibility step into the generation pipeline rather than treating it as post-production. This cuts remediation cost by roughly 80% compared to retrofitting.
Don't forget interactive synthetic media. Dynamic AI avatars and real-time personalized video introduce WCAG 2.2 considerations beyond the media itself — pause controls (SC 2.2.2), keyboard accessibility (SC 2.1.1), and screen reader compatibility. These require custom development work, not just auto-generated metadata.
The Bottom Line
Synthetic media isn't going anywhere. The $48B projection for 2033 makes that clear. But the brands that adopt AI content pipelines without building accessibility compliance into the workflow are creating litigation risk at the same scale they're creating content.
The fix isn't complicated: audit what you have, classify it by WCAG criterion, remediate with AI-powered tools, and build compliance checks into the pipeline. The cost of doing it upfront is roughly 5-8% of your synthetic media production budget. The cost of a lawsuit or a bad accessibility audit is exponentially higher.
Our team at Mintec has built accessibility-compliant synthetic media pipelines for clients in education, retail, and real estate. If you're scaling AI content production and want to stay on the right side of WCAG 2.2, we should talk.
Related reading: Web Accessibility: Why Your Website Is Costing You Customers · Performance Budgets for Synthetic Media Sites · The Real Cost of Rich Media in 2026
Frequently Asked Questions
What WCAG requirements apply to AI-generated video?
WCAG 2.2 Success Criterion 1.2.2 (Captions - Prerecorded) requires captions for all prerecorded audio content in synchronized media, and SC 1.2.5 (Audio Description - Prerecorded) requires audio descriptions. These apply regardless of whether the video was AI-generated or traditionally produced.
Can AI tools generate WCAG-compliant alt text automatically?
Yes, modern vision-language models can generate descriptive alt text at scale, but they require human review. Research shows AI-generated alt text misses context-critical details 15-30% of the time. The best workflow is AI generation + human spot-checking for the top 20% of images.
How does synthetic media create unique accessibility risks compared to traditional media?
Scale. Human-produced media typically goes through editorial review where someone catches missing captions or alt text. AI content pipelines can produce hundreds of video variants per hour — all of them bypassing the editorial accessibility gate. The problem isn't that AI generates inaccessible content; it's that it generates it at a velocity that breaks traditional QA workflows.
