ADA lawsuit tracker / method
Which platforms get sued? We tried to measure it, and stopped.
Every accessibility vendor quotes some version of this statistic: the share of sued businesses running one platform or another, or running an accessibility widget when the case landed. We wanted a version we could stand behind, so we put a full month of federal filings through a documented pipeline and wrote the pass marks down before we started.
We missed them. Of the 348 businesses named as defendants in May 2026, we could confidently classify 146. The rest stayed unknown, and the rule we had set in advance said that at this much unknown, the answer does not get published. This page is the attempt in full: the method, the funnel, and the verdict against ourselves.
Run completed 2026-08-08. Covers May 2026, one month with substantially complete records. No parties are named here.
Any figure of the form “X% of sued businesses ran platform Y” or “X% had an accessibility widget installed”. We are not sitting on a number we privately believe. 58.0% of the month could not be classified, and the part that could is not a sample anyone should reason from, ourselves included.
Why this is harder than it looks
The tracker earns trust one way: every count links to the public search that reproduces it. A platform breakdown can never work like that. The court record names a defendant; it does not say what software that defendant’s website ran. Every classification is ours, which means the method has to carry the weight the query link normally carries.
Two further problems come with the territory. The suit code we count covers non-employment ADA cases of every kind, so website cases and physical-premises cases sit in the same pile with nothing to separate them. And the widget question cannot be answered by scanning sites today: a business sued in May may have installed a widget in June, often because of the case, so present-day scanning would systematically find widgets that were not there at the time.
So the bar we set before running anything was: a written method, a stated denominator, an explicit unknown bucket, evidence retained per case, and, for the widget question, evidence of what a site ran at the time it was sued.
The pass marks, set before the run
These three lines were written into the method document before a single case was fetched, precisely so that a disappointing result could not be argued into a publishable one afterwards.
| Measured | Line set in advance | Actual | Result |
|---|---|---|---|
| Websites identified with high confidence | at least ~60% of the population | 56.6% | Missed |
| High-confidence websites with a capture in the window | at least ~70% | 74.6% | Met |
| Everything left unknown | abandon the result above ~50% | 58.0% | Stop |
The middle line was the one we expected to fail, and it passed. The step that broke was the mundane one: matching a business named in a case caption to its actual website.
The pipeline
Take the month
Every filing in the month under the same nature-of-suit code the tracker counts, pulled from the public court-records archive. This is the one step anyone can reproduce with a link.
Keep only businesses
Company and organisation defendants only. People named as defendants are dropped outright and never processed further, which is also why no individual appears in any count on this page.
Find the website
Match each business to its main public website, recording a confidence for every match: high, medium, or none. Medium matches are deliberately not classified. This is the fuzziest step in the pipeline, and the run was designed to measure exactly how fuzzy.
Go back to the filing date
Look up the web archive for a capture of that site within 45 days of the date the case was filed. No capture in the window means the business is not classified. The capture timestamp, not today, is the classification date.
Read the archived page only
Detect the e-commerce platform and any accessibility widget from the archived HTML, using a published, version-stamped list of vendor markers. Never from the live site: what a site runs today is not evidence of what it ran when it was sued.
Tag what kind of business it is
Online store, brand site, restaurant or hotel, local services, government or education, or other, from the same archived page.
Publish totals only
Counts and buckets. No party names, no domains, no per-case rows, ever.
Where a month of filings actually goes
This is the finding. Every business that entered the run leaves it in exactly one bucket below, and all of the buckets are here, because the size of the unknown ones is the whole point.
| Stage | Count | Share |
|---|---|---|
| Filings in the monththe same public query the tracker publishes | 378 | |
| Unique named defendants | 373 | |
| Natural-person defendants, excluded by ruleindividuals are never processed or counted | 23 | |
| Case captions we could not resolve to a party | 2 | |
| Company and organisation defendants (the population) | 348 | |
| Website identified, high confidence | 197 | 56.6% |
| Website identified, medium confidencenot classified: the confidence rule excludes them | 43 | 12.4% |
| No identifiable websiteholding companies, landlords and small premises defendants | 108 | 31.0% |
| Archived capture within 45 days of filingof the high-confidence websites | 147 | 74.6% |
| No archived capture in the windowof the high-confidence websites | 50 | 25.4% |
| Archived page could not be readthe archive refused the fetch; recorded rather than retried into a result | 1 | |
| Classified under the rule | 146 | 42.0% |
| Left in an unknown bucketmedium confidence, no website, or no readable capture | 202 | 58.0% |
Shares are of the 348 business defendants, except the two archive rows, which are of the 197 businesses whose website we identified with high confidence. The filing count is the tracker’s own public query for the month, and rises as the archive receives more records.
The verdict
Unknowns came to 58.0% against a ceiling of about 50%, so the platform and widget shares are not published. Not published in a footnote, not published with a caveat, not published over a narrower denominator that happens to look better.
The reason that ceiling exists is visible in who ended up unknown. The businesses we could not match to a website are not a random slice: they are holding companies, landlords, and small premises defendants, the kind of case that has nothing to do with a website at all. A share calculated over what remains would not describe businesses sued under the ADA. It would describe businesses sued under the ADA that also have an easily findable web presence, and it would be quoted as if it meant the first thing.
One more reason to distrust it: the sites where no platform marker appeared were not obscure. Some of the largest stores in the month are built in ways that leave no vendor fingerprint in the page source at all, which means “no signature detected” has to stay its own honest bucket and can never be read as “no platform”.
Why publish a measurement that failed
Because the alternative is what this corner of the market already does. Accessibility statistics circulate for years with no stated denominator, no unknown bucket, and no way to check them, and they work, commercially, because nobody can test them.
We would rather be the source that shows the whole funnel, including the two thirds of it that defeated us. If we tell you a number later, this page is the reason to believe it: you have seen what we do with a number that did not survive its own test.
The funnel is a real finding in its own right. Anyone planning to measure the same thing now knows where the effort goes and which step will break first, and it is not the one you would guess.
What would change the answer
Three things would move this from a null result to a publishable one. None is scheduled, and if any of them happens, the new pass marks get written down before the re-run, not after it.
Separating website cases from premises cases
The suit code covers both, and it does not say which is which. Only the complaint text does, and getting it for a full month at a time is a document-access problem we have not solved. Until then, no honest version of this measurement can call itself a count of website lawsuits.
A second archive source
A quarter of the sites we identified had no capture near their filing date. Another archive, or a wider window, would recover some of them, but a wider window buys coverage by loosening the classification date, which is the guarantee that makes the widget question answerable at all.
Reporting medium-confidence matches separately
Twelve per cent of the businesses had a probable website match we refused to use. Shown as their own tier, clearly labelled, they would be informative. Merged into the main figures to lift the coverage rate, they would be exactly the corner-cutting this exercise exists to avoid.
Method notes
- Denominator. The 348 company and organisation defendants named in May 2026’s filings. This is deliberately not a count of website lawsuits: the suit code does not distinguish them from premises cases, and we do not guess.
- Classification date. The timestamp of the archived capture, taken within 45 days of the filing date, never the live site. Every classification carries the capture it came from.
- Signature list. v1 (2026-08-03). Platforms: Shopify, WooCommerce, BigCommerce, Wix, Squarespace, Salesforce Commerce Cloud, Magento. Accessibility widgets: accessiBe, UserWay, AudioEye, EqualWeb, Level Access, MaxAccess, TruAbilities. Detection is a match on vendor markers in the archived page source, with rules for known false positives. Anything outside the list is invisible to the method, which is one of the reasons the no-signature bucket is large.
- Reproducibility.The tracker’s counts are reproducible by query: click the link, get the number. This exercise is reproducible by procedure: the method is documented well enough for someone else to run it and compare, but there is no link that regenerates it. Those are different guarantees, and running them together is how accessibility statistics get their undeserved authority.
- People. Individuals named as defendants are excluded at the second step and never processed. All counts here are of companies and organisations.
- Evidence. Per case we retain the docket reference, the matched site and its confidence, the archive capture and its timestamp, and the markers found. That record names parties, so it is not published. If you have a legitimate reason to verify or correct something here, ask and we will share the relevant part of it.
Compiled from public US federal court records and public web archives. A lawsuit filing is an allegation only. We assert no wrongdoing by any party, nothing here suggests any platform or accessibility widget causes or prevents litigation, and none of this is legal advice. Corrections or questions: support@screenmy.site.
The numbers we do publish
Filing volumes, by month and by district, each one linked to the public query that reproduces it, on the ADA lawsuit tracker. And if the practical question behind all of this is what your own site would show, run the free scanner or read a sample report.