Image description: Person examining printed reports and documents at a table.

We analyzed 8,788 web accessibility lawsuits – here’s what actually gets companies sued

We analyzed 8,788 web accessibility lawsuits – here’s what actually gets companies sued

A federal court database of 113,120 specific accessibility complaints from real ADA Title III filings. Twelve recurring patterns drive over 70% of cases. Four law firms file 60% of lawsuits. The median case settles in 97 days. Here is what plaintiffs are actually suing over — line by line, page by page.

Findings · Case file 01 06 entries · derived from 113,120 issues across 8,788 cases

What 113,120 lawsuit complaints reveal

1. 01 3,449

   Filings have grown in four years — from 466 in 2021 to 3,449 in 2025

   2024 saw 2,156 cases. 2025 saw 3,449. Q1 2026 is on pace to match or exceed 2025. Federal accessibility litigation is no longer cyclical — it is structural.

2. 02 60%

   Just four law firm groups filed 60% of all 8,788 cases

   Gottlieb-related firms: 1,798 cases. Stein Saks: 1,562. Throndset Michenfelder: 1,007. Equal Access Law Group: 988. The litigation is concentrated, not distributed.

3. 03 97 days

   Half of cases close in under 90 days — the median is 97

   46.2% settle inside 90 days; 83.5% inside 180. The economic logic of these cases isn’t a verdict — it’s a quick settlement that avoids legal costs.

4. 04 17,693

   “Not announced by the screen reader” is the single most-cited complaint

   17,693 issues — 21.7% of all extracted complaints — describe a screen reader failure. Forms, cart icons, modals, and confirmation messages dominate.

5. 05 991

   Identical boilerplate appears in hundreds of complaints

   One sentence — “These barriers to access have denied Plaintiff full and equal access…” — appears verbatim in 991 separate filings. The complaints are templated.

6. 06 36%

   eCommerce / retail defendants account for 36% of all cases

   The “Add to Cart” button not announcing confirmation, the cart icon labeled only as “collapsed,” unlabeled checkout fields — these specific patterns recur across thousands of complaints.

Source113,120 issue descriptions extracted from 8,788 federal accessibility lawsuits filed between 2007 and April 2026, cross-referenced with case metadata from PACER federal court records.

---

How we analyzed the data

Most public discussion of accessibility lawsuits relies on yearly summaries from advocacy groups or marketing pieces from compliance vendors. We took a different approach: we analyzed the underlying complaint documents themselves. Specifically, we worked from a dataset built by extracting structured information from 8,788 federal court cases filed under ADA Title III and related statutes — every case in our database where the plaintiff alleged a website, app, or other digital service failed to meet accessibility requirements.

The dataset has two layers. The first layer is case metadata: case title, case number, plaintiff name, defendant name, plaintiff’s law firm, lead attorney, filing date, and closure date. This came from PACER federal court records — the public docket system used by every U.S. district court. The second layer is granular: 113,120 individual issue descriptions extracted from the underlying complaints, each tagged to a specific category (screen reader, keyboard navigation, forms, color contrast, and so on) and linked back to its source case.

The category taxonomy was built bottom-up from keyword analysis of the actual complaint texts. We did not start with WCAG success criteria and search for matches. We started with what plaintiffs actually wrote in their complaints, clustered the language, and let the categories emerge from the data. The result: 27 functional categories matching how lawsuits describe accessibility failures, not how compliance frameworks formally classify them.

Important caveats

This dataset has clear limitations and we want to state them upfront. First, it is federal cases only — many ADA accessibility cases are also filed in state courts (notably under California’s Unruh Act and New York City’s Human Rights Law) and those state filings are not in our data. Second, the dataset captures filed cases, not demand letters. The number of demand letters that resolve before any lawsuit is filed is widely believed to be several times larger than the filed-case count, but those don’t appear in court records. Third, our categorization is heuristic: a single complaint typically describes multiple barriers, and “General / Uncategorized” remains the largest bucket (38,672 issues) because plaintiff language often describes a barrier in narrative form before getting specific. Fourth, we cannot tell from PACER metadata whether a “closed” case settled, was dismissed, was won by the plaintiff, or was won by the defendant — only that the docket was closed.

What this dataset is good for is identifying recurring patterns: which page elements get cited, which complaints are templated, which firms are filing, how filings are trending, and how long cases last. It is not a substitute for legal advice on any specific case.

---

The volume problem: Filings are up

The most immediate finding is the trajectory. In 2021, there were 466 federal accessibility cases filed in our dataset. In 2025, there were 3,449. The first three full months of 2026 (Jan, Feb, Mar) saw 280, 273, and 282 filings respectively — implying an annualized rate of approximately 3,300 cases for 2026, holding roughly steady at the new elevated baseline.

The acceleration in late 2024 is particularly striking. Through August 2024, monthly filings averaged 130 cases. From September 2024 forward, the monthly average jumped to 277 cases — and has stayed at that level ever since. The single highest month in our dataset is July 2025 with 376 filings; that’s roughly 12 new accessibility lawsuits filed every business day across the U.S. federal court system.

Two contextual notes. First, our dataset captures federal cases pulled from PACER and may not include every accessibility-related state-court filing — California’s Unruh Act, New York’s Human Rights Law, and similar state statutes generate substantial additional litigation that is not federal. Industry trackers that report annual totals in the 4,000–5,000 range typically combine federal and major state-court filings. Second, the year-over-year growth in federal filings tracks with the broader trend reported by accessibility lawsuit trackers — the dataset’s trajectory is consistent with what other public reporting has shown.

---

Who is filing these lawsuits

If 8,788 cases were spread evenly across the U.S. legal community, we’d expect to see thousands of different attorneys filing one or two cases each. That is not what the data shows. Federal accessibility litigation is extraordinarily concentrated on a small number of plaintiff firms.

The top 10 plaintiff law firms account for 70% of all cases

Counts combine all variant spellings of the same firm name as recorded in PACER (e.g., “Stein Saks, PLLC” and “Stein Saks PLLC” combined).

Just these top 10 firm groups account for 70% of all 8,788 cases in our dataset (6,151 cases, after combining variant spellings). The top 5 firm groups alone account for 45.1%. By any measure of legal-services markets, this is an extraordinary concentration. For comparison, the 50 largest U.S. law firms by revenue collectively handle perhaps 8–10% of all federal litigation.

The plaintiff concentration is even tighter

The dataset reveals a clear two-tier structure of plaintiffs. On one side, 36 individuals filed 50 or more cases each, and 6 individuals filed 100 or more. These are commonly referred to as “serial plaintiffs” or “tester plaintiffs.” On the other side, 600 of the 1,071 unique plaintiffs in our dataset filed only one case each — these are individual disabled users who encountered a specific barrier on a specific website and pursued one claim.

The serial plaintiffs are not random. One complainant filed 256 cases between 2023 and 2026, with a sharp acceleration: 7 cases in 2023, 52 in 2024, 135 in 2025, and 62 in the first quarter of 2026 alone. Her defendants span Williams-Sonoma, Hanesbrands, Fossil Group, Oxford Industries, FullBeauty Brands, and dozens of other consumer brands. Another plaintiff filed 220 cases since 2021, with defendants ranging from Sherwin-Williams to small Florida real estate firms. The pattern is consistent: each serial plaintiff partners with one or two specific law firms and files in volume against a wide and varied set of defendants.

---

The 97-day settlement: Why cases don’t go to trial

Of the 8,788 cases in our dataset, 6,951 had both a filing date and a closure date — enough to compute case duration. The distribution is striking.

What this distribution reflects, in practice, is that ADA Title III website cases are built for fast settlement, not for trial. The economics work like this. A defendant company served with a complaint faces a choice: spend $50,000–$200,000+ on initial defense work (motion to dismiss, discovery, expert witnesses), or spend $5,000–$30,000 on a quick settlement that includes a remediation commitment plus the plaintiff’s attorney fees. For most defendants — especially mid-market eCommerce, restaurants, and service businesses — settlement is unambiguously cheaper, even when the underlying complaint is contestable on its merits.

“Pay to make it go away” is not a defect of this system. It is the system. The 97-day median is the visible signature of that economic logic.

This also explains why the dataset shows so many one-and-done plaintiffs (600 of 1,071) but a small group of repeat filers (36 with 50+ cases). The repeat filers are operating at industrial scale: identify 100+ websites with similar barriers, file 100+ complaints, settle each within 90 days, collect attorney fees from each settlement. The math only works at volume — and the volume is what produces the settlement-amenable defendants.

---

What gets sued: The top 12 issue categories

From the 113,120 individual issue descriptions, we measured how often each barrier category appears across unique cases — not raw issue counts (which would over-weight cases with longer complaints), but the share of cases in which the category appears at least once.

Each percentage is the share of the 7,543 cases for which we have categorized issues that mention the barrier at least once. Many cases mention multiple categories, so percentages sum to more than 100%.

The pattern is interpretable. Screen reader incompatibility is the master category: it shows up in over half of all cases, and most of the other categories (forms, navigation, modals, alt text, links) are specific manifestations of the same underlying problem — the website does not communicate properly with assistive technology. Keyboard navigation is the second master category: 37% of cases describe an inability to operate the site without a mouse, often combined with screen reader issues.

The smaller categories are no less important to specific industries. Color contrast appears in only 309 cases overall, but in ~10% of cases involving healthcare, financial services, and government websites where text-heavy content is core to the user task. Video/audio appears in 23% of cases overall but in close to 100% of cases involving educational institutions or media companies where video is the product itself.

---

The #1 complaint: “Not announced by the screen reader”

Of all the language in our dataset, the most-cited single complaint pattern is some variant of “the element is not announced by the screen reader” or “not labeled to integrate with the screen reader.” 17,693 separate issues use this framing — 21.7% of all extracted complaints.

What makes this category particularly diagnostic is that it tells you precisely where on a page the failure occurred. Plaintiffs (and their lawyers) are very specific about which element a screen reader fails to announce. Here are real examples directly from complaint texts in our dataset:

The diagnostic value here is enormous. Each complaint pinpoints a specific element with a specific failure. They are not abstract — they describe what a screen reader actually says. “Number link.” “Collapsed.” “Button button button.” These are the literal output of a screen reader hitting an element that lacks an accessible name. Anyone doing manual accessibility QA on a site can use these exact phrases as test cases: “Does our cart button announce as ‘cart’ or as ‘collapsed’? Does our quantity field allow the user to move past it after entering a value?”

The patterns within the patterns

Within the screen reader category, certain sub-patterns recur with striking consistency:

The last row is from an actual complaint in our dataset. The precision is the point: a plaintiff’s lawyer reading their screen reader transcript can copy the literal output into the complaint, and that becomes the basis for a count. Every page of the site that produces an obvious screen reader artifact is a potential count in a future complaint.

---

Forms: The quietest liability

Form-related complaints appear in 1,226 cases — about 16% of our dataset — but they are under-represented in the headline numbers because most form complaints are tagged under “screen reader” or “general” rather than as an explicit form category. The actual rate of form-related grievances across all complaints is closer to 30–40% based on keyword scan.

What plaintiffs cite, specifically:

The two real-world failures we kept seeing in checkout-related complaints:

The “0 check box not checked” announcement is a particularly common artifact: it happens when a custom checkbox component uses a div with an inline handler, the script wires aria-checked="false" but no associated label, and the screen reader falls back to announcing whatever DOM text it can find — which in this case was a “0” that referred to something else entirely.

---

Navigation menus: The hamburger tax

Global navigation appears in 2,732 cases (36%). The most common targets within this category are mobile hamburger menus, mega-menus, and dropdown submenus. The complaints are remarkably consistent across defendants:

What happens in practice: a sighted user clicks the hamburger icon, the menu animates open, and they see options. A screen reader user tabs to the same icon, hears “button” (no accessible name), activates it, and either nothing is announced (the menu opened but the screen reader doesn’t know it), or focus stays on the icon while the menu’s items are visually visible but not in the keyboard tab order.

This is a textbook WCAG 2.1 violation: SC 4.1.2 (Name, Role, Value) requires that user interface components have a programmatic role and accessible name; SC 2.1.1 (Keyboard) requires that all functionality be operable through a keyboard. A hamburger button without an accessible name fails both. The fix is two attributes: aria-label="Menu" and aria-expanded="true|false" updated on toggle. Every site we saw cited in our dataset for hamburger-menu issues was a site that had not done these two things.

---

Images and alt text: 42% of all cases

Image accessibility appears in 3,179 cases — 42% of our dataset. The complaints are not just about missing alt text. They are about wrong, redundant, or misleading alt text, often introduced by automated tools.

The recurring patterns:

One specific complaint pattern from our dataset is so consistent we’ll quote it directly:

This is the signature of a CMS image template that hasn’t been configured: the site’s marketing team is uploading a “hero” image to a slot called hero-3-graphic, the CMS is using the slot name as the fallback alt text, and the site has four such hero rotations on the homepage. Every blind visitor to the homepage hears the same useless string four times in succession. It’s a single CMS configuration away from being fixed — but it appears in complaints repeatedly because no one ever fixed it.

The “alt=’image of a blue and yellow sign’ for a logo” problem

A growing share of image-related complaints in 2024–2026 trace to AI-generated alt text from accessibility overlay tools. Our category data flags 241 complaints describing alt text that is either nonsensical or actively misleading — descriptions like “text,” “file,” “city,” or vague visual descriptions for elements that have specific, important meaning (a company logo, a product photo, a status icon). These complaints are particularly hard for defendants because the AI-generated alt text was added by an automated overlay tool that the company believed was making the site more accessible — and the complaint is that it made the site less accessible by introducing wrong information.

---

Modals, popups, and cookie banners: 21% of cases

Modal and popup issues appear in 1,616 cases (21%). The single most common phrase in this category is “not announced” (502 mentions) — meaning the modal opened, but the screen reader has no idea it opened, and focus stayed on the trigger button.

The pattern is so consistent across thousands of cases that it’s worth stating in full as the canonical “modal failure mode”:

1. User clicks a button (sighted user sees the modal open).
2. The modal is added to the DOM with no role="dialog" or aria-modal="true".
3. Focus is not moved into the modal.
4. The screen reader user is still stuck on the original button, and has no idea the page state changed.
5. They tab forward — the modal’s focus is not trapped, so they tab into the page background.
6. They hit Escape — nothing happens (the modal has no Escape handler).
7. The visual modal blocks them from accessing what’s behind it.
8. They give up.

The fix is well-documented and standard: role="dialog" on the modal container, aria-modal="true", move focus to the first focusable element inside the modal on open, trap focus within the modal while it’s open, return focus to the trigger on close, and listen for Escape to close. Yet 100+ complaints in our dataset describe a modal that cannot be closed — meaning the Escape handler doesn’t exist and the close button itself isn’t reachable.

---

The checkout flow: Where eCommerce loses cases

Checkout-specific complaints appear in 1,656 cases (22%). When you combine these with related categories — Add to Cart (718 cases), Shopping Cart / Basket Page (1,022), Payment (524), and Address Management (130) — the share of cases describing a broken purchase flow is much higher.

The complaint patterns are remarkably specific to the eCommerce funnel:

The economic damage of a checkout barrier is asymmetric. Every other accessibility barrier on a website affects a user’s ability to find information. A broken checkout affects a user’s ability to complete a purchase. This is why eCommerce defendants face disproportionate exposure — not because eCommerce sites have more barriers per se, but because every barrier on a checkout page is, in effect, a denial of service that the law treats more seriously than a barrier on, say, an “About Us” page.

One specific failure we saw in our data multiple times: the “Add to Cart” success state. Many sites add an item to the cart without a page reload, displaying a small toast notification or modal that says “Added to cart!” If that toast is not in an aria-live region, the screen reader user gets no feedback that anything happened. They click Add to Cart again. And again. They may end up with three of the same item in the cart — or they may give up and abandon the purchase entirely. Both outcomes appear in complaints.

---

Keyboard navigation: The test that catches 37% of cases

Keyboard navigation issues appear in 2,821 cases (37%). The complaints are concrete:

The single most actionable insight from this category: the keyboard test catches the majority of accessibility violations on most websites in under five minutes. Press Tab. Watch where the focus indicator goes. If it disappears (no visible focus state) — that’s SC 2.4.7 (Focus Visible). If you reach an element and can’t tab past it — that’s SC 2.1.2 (No Keyboard Trap). If you reach the modal and tab order goes into the background page — that’s SC 2.4.3 (Focus Order). If the cart icon doesn’t receive focus at all — that’s SC 2.1.1 (Keyboard).

You don’t need a screen reader, an audit tool, or any specialized knowledge to find most of the issues that get sites sued. You need a keyboard. The fact that 2,821 cases describe keyboard failures means 2,821 sites failed a 5-minute test that any of their developers could have run.

---

The boilerplate problem: How “custom” complaints aren’t

One of the most striking findings from our pattern analysis: a substantial share of complaint language is identical across hundreds of separate cases. We ran exact-string matching on the first 80 characters of each issue description. Several phrases recur with extraordinary frequency:

Boilerplate phrase	Cases using verbatim
“These barriers to access have denied Plaintiff full and equal access to, and enj…”	991
“Plaintiff has been denied the full enjoyment of the facilities, goods and servic…”	843
“Administrative Code § 8-107(4)(a) in refusing to update or remove access barrier…” (NYC HRL)	504
“In fact, the access barriers make it impossible for blind and visually-impaired…”	312
“Plaintiff did not understand the purpose of the interactive element on the page…”	135
“Plaintiff could not determine in which part of the sub-menu the keyboard focus…”	78
“As a result, Plaintiff had difficulty in navigating the menu and could not deter…”	64
“Plaintiff received inaccurate information about the purpose of the element in fo…”	61
“The website had device-specific functionality, like mouse-dependency, making it…”	55
“Plaintiff was not provided with the mechanism to bypass repeated blocks of conte…”	50

This is what a litigation factory looks like. The boilerplate is efficient — once a complaint template has been litigation-tested (i.e., survived motions to dismiss in earlier cases), it can be reused across hundreds of subsequent complaints with only the defendant’s name and a few site-specific details changed. The issue-specific paragraphs are then dropped in as standard modules: “menu drop-down options not labeled,” “focus indicator missing,” “screen reader not announcing the cart counter,” and so on.

For defendants, this has two implications. First, the complaint they receive looks customized but isn’t — most of it is shared with hundreds of other defendants in the same firm’s litigation pipeline. Second, the defense response can also be templated, and many large defense firms now have standardized accessibility-litigation responses ready to deploy. This is part of why settlements close so quickly: both sides have done this before.

---

Which industries get sued the most

Our defendant data is messier than the issue data — the JSON files contain tens of thousands of unique defendant names, and many contain only company-name strings without industry classification. Using keyword analysis on defendant names, we get a rough industry distribution:

The remaining ~46% of defendants did not match any of our industry-keyword filters and are unclassified — many are LLCs, holding companies, or small businesses without industry-identifying names. This is a rough heuristic, not a precise classification.

The dominant pattern is eCommerce and retail at 36%, by a very wide margin. This makes sense given the structure: eCommerce sites have the most interactive functionality (search, browse, cart, checkout), the most pages, the most images, and the highest exposure per visitor. They also process transactions, which means a barrier denies a measurable economic benefit. Restaurants are the second-largest cluster — driven heavily by online ordering and reservation flows that fail accessibility.

What is striking is the diversity of defendants within each industry. Among eCommerce defendants, we see Williams-Sonoma, Hanesbrands, Fossil Group, Crocs, Burberry, Calzedonia, and thousands of much smaller retailers. The lawsuits are not concentrated on a small number of “bad actor” companies — they are spread broadly across the entire eCommerce sector.

---

Risk exposure calculator

Below is a rough estimator of likely exposure based on the patterns in our dataset. The numbers are not legal advice and not specific to any one defendant — they are aggregated approximations derived from the case-duration and settlement-pattern data above. Actual exposure depends on jurisdiction, industry, prior accessibility history, presence of a documented remediation program, and many factors not reflected in PACER metadata.

The dominant variable in this estimate is accessibility maturity. Sites with a mature program — meaning automated accessibility tests in CI, manual audits at least annually, accessibility built into design and code review — face exposure roughly 20× lower than sites with no program at all. The compound effect: investing in mature accessibility saves about 95% of the annual lawsuit risk for the same site profile, in our rough estimate.

---

What actually reduces risk: Patterns from the cases that don’t get sued twice

Our dataset has a useful negative signal: the 600 plaintiffs who appear in only one case. These are individual disabled users who encountered a barrier and pursued one claim — not high-volume serial litigation. The defendants in these one-off cases tend to fall into two groups: those who never get sued again (because they fixed the underlying issue), and those who get sued by a different plaintiff six months later (because they didn’t).

From the closing-date and case-pattern data, the defenses that show up as effective in actual litigation outcomes are not novel. They are the standard accessibility engineering practices. But our data does let us rank them by how often the underlying issue appears in complaints — i.e., which fixes prevent the most recurrence:

---

The bottom line

For organizations that are not currently being sued, the question is not “how do we avoid lawsuits?” but “how do we build a digital service that disabled users can actually use?” The two questions converge on the same answer, but they imply very different organizational priorities. The first leads to defensive add-ons that often introduce new barriers. The second leads to durable engineering practice that produces a better product for everyone — and incidentally produces the strongest legal defense available, which is a site that does not fail accessibility tests in the first place.

The 8,788 cases in our dataset and the 113,120 complaints inside them are not random. They are a near-comprehensive map of the small set of recurring failures that produce nearly all of the legal exposure. The map has been drawn. The work now is to use it.