AI-powered alt text

AI-powered alt text does one thing well. It reduces empty alt attributes at scale.

In audits from 2024 and 2025, missing alt text remained one of the most common WCAG failures cited in ADA website lawsuits. AI tools cut that number fast, especially on large image libraries.

For news sites, real estate platforms, and e-commerce catalogs with tens of thousands of images, AI can fill baseline descriptions that didn’t exist before.

That matters because courts have repeatedly treated empty or missing alt attributes as low-hanging fruit for plaintiffs. Filling them removes an easy allegation.

It does not end the analysis.

Where AI-generated alt text fails, consistently

The failures fall into patterns. They show up in audits. They show up in complaints. They show up in settlement remediation lists.

Context blindness

AI doesn’t know whether an image is decorative, informational, functional, or persuasive.

A hero banner that says “Schedule an appointment” overlaid on an image of a smiling dentist may get alt text like:

“Smiling woman wearing scrubs in an office.”

That fails WCAG. The purpose of the image is the call to action, not the person’s expression.

Functional images misdescribed

Buttons, icons, and linked images often receive literal descriptions instead of functional ones.

A shopping cart icon described as “black shopping cart icon” does not convey function to a screen reader user. WCAG expects something closer to “View cart.”

AI misses that distinction unless explicitly trained and constrained by markup and developer input.

Over-description of decorative images

AI tends to describe everything. Decorative dividers, background textures, stock photography used purely for layout all get descriptions when they should be ignored.

This creates noise for screen reader users. Courts have accepted this as an accessibility failure in multiple cases, especially when it interferes with navigation.

OCR errors and hallucinations

AI systems that include optical character recognition sometimes misread text inside images. They also invent text that isn’t there.

In one 2025 audit for a regional retailer, an AI tool added alt text claiming a sale price that had ended weeks earlier. The image contained no text at all. The model inferred it from prior training patterns.

That’s not just inaccurate. It’s misleading.

A real lawsuit example involving AI-generated alt text

In October 2025, a small online furniture retailer based in Ohio was sued in the Southern District of New York. The site had deployed AI-generated alt text across its catalog earlier that year.

The complaint cited multiple images with alt text like:

“Brown wooden chair in a room.”

The problem wasn’t that the description was wrong. It was that the images were links to product detail pages. Screen reader users weren’t told that.

The plaintiff’s tester documented confusion during navigation. The case settled for $9,000 plus remediation. The settlement agreement required manual review of AI-generated alt text for functional images.

The AI didn’t cause the lawsuit. It didn’t prevent it either.

Why AI alt text doesn’t change legal exposure on its own

Courts don’t ask how alt text was generated. They ask whether it works.

ADA Title III claims don’t include a safe harbor for automation. There’s no credit for effort. There’s no reduced liability for using advanced tools.

Standing is established by alleged inability to access content at the time of visit. If AI-generated alt text fails in context, the method is irrelevant.

Defense attorneys learned this quickly in 2024. By 2025, most stopped arguing that AI use showed good faith. Judges weren’t interested.

How plaintiff firms treat AI-generated alt text

Plaintiff-side firms adapted fast. They now expect to see alt text everywhere. That changes how they frame complaints.

Instead of citing missing alt attributes, they cite meaningless ones.

Instead of empty strings, they quote generic AI phrases and explain why they don’t convey purpose.

This actually makes complaints stronger. It shows interaction. It shows attempted access. It shows frustration.

AI raised the bar for what counts as inadequate, not adequate.

The trade-off businesses don’t like to hear

AI-powered alt text saves time. Manual alt text costs money.

For a 5,000-image catalog, manual writing can cost $5,000 to $15,000 depending on complexity and who does it. AI does it for pennies.

The trade-off is risk concentration. One flawed model deployment can introduce thousands of similar accessibility errors at once.

When that happens, plaintiffs don’t see one issue. They see a pattern.

Patterns drive lawsuits.

How auditors evaluate AI-generated alt text in 2025

Accessibility auditors don’t reject AI outright. They categorize it.

In audits conducted throughout 2025, AI-generated alt text typically falls into three buckets:

• Acceptable as-is
• Acceptable with edits
• Non-compliant

Decorative images often end up miscategorized. Functional images require the most human correction. Informational images land in the middle.

Auditors now flag “AI pattern errors” where the same phrasing repeats across pages. That repetition becomes evidence in litigation.

Why WCAG doesn’t give AI any special treatment

WCAG is technology-neutral by design. It doesn’t care whether content is written by humans, machines, or both.

Success Criterion 1.1.1 asks a single question: does the text alternative serve the same purpose?

AI tools aren’t evaluated differently because the standard predates them and intentionally avoids prescribing methods.

That frustrates businesses hoping for clearer guidance. It also keeps enforcement simple.

CMS and platform defaults made things worse in some cases

In 2025, several major CMS platforms enabled AI alt text by default. Images uploaded without user input received auto-generated descriptions silently.

That created compliance problems for site owners who didn’t realize what was happening.

In at least two demand letters reviewed in late 2025, plaintiffs cited bizarre or misleading alt text that the site owner had never seen. It came from a platform feature they didn’t know existed.

Responsibility still landed on the business.

How AI alt text interacts with other ADA failures

Alt text rarely appears alone in lawsuits. It clusters with other issues.

Poor alt text combined with:

• Keyboard traps
• Improper heading structure
• Unlabeled form fields

creates a stronger narrative of inaccessibility. AI-generated alt text doesn’t offset those issues. Sometimes it highlights them by making other failures more obvious.

The accessibility community’s criticism of AI alt text

Blind and low-vision users have been vocal about AI-generated descriptions. The criticism is consistent.

Generic descriptions waste time. Over-description interrupts flow. Misleading descriptions cause mistakes.

Some users prefer missing alt text to bad alt text because at least silence is predictable.

Courts haven’t ruled on that preference directly, but complaints increasingly reflect it in affidavits.

A concrete example from a government site

In mid-2025, a municipal website in the Midwest rolled out AI-generated alt text as part of a redesign. Public meeting photos received descriptions like “Group of people sitting at a table.”

For residents using screen readers, that added nothing. The images documented official actions. Dates, votes, and context mattered.

The city received a formal accessibility complaint, not a lawsuit. The remediation required manual rewriting of hundreds of descriptions.

AI sped up the mistake. Humans had to clean it up.

How AI search and summaries complicate the issue

AI-generated alt text is now used beyond screen readers. Search engines and AI assistants ingest it as part of page understanding.

Inaccurate alt text doesn’t just affect accessibility. It affects discoverability and summarization.

That creates a secondary risk. Fixing alt text after a lawsuit may not undo how content has already been indexed or summarized elsewhere.

Businesses are learning that accessibility errors echo further than expected.

What law firms now tell clients about AI alt text

By late 2025, the message from ADA defense and advisory firms converged.

AI-generated alt text is acceptable as a starting point. It is not defensible as a final product.

Manual review is required for:

• Functional images
• Images conveying legal, medical, or transactional information
• Repeated templates

Firms now document that advice in writing. Not because it prevents lawsuits, but because it helps during settlement negotiations.

The limit nobody can engineer around

AI doesn’t understand intent. Alt text requires it.

Until a system can reliably infer why an image exists in a specific context on a specific page for a specific user task, human judgment remains necessary.

That’s not philosophical. It’s visible in every lawsuit filed in 2025 that mentions alt text.

Automation reduced one category of failure. It created another.

Where AI-powered alt text leaves ADA compliance right now

AI-powered alt text changed workflows. It did not change liability.

It fills gaps fast. It also fills them sloppily unless constrained.

Courts don’t reward speed. They measure access.

Businesses using AI for alt text are not wrong. They are unfinished.

That distinction shows up in audits, complaints, and settlements, one image at a time.