How ClaimWatch works

1. Where settlements come from

Settlements are discovered directly from the websites of court-appointed settlement administrators — the companies courts appoint to run a settlement's notice and claims process. ClaimWatch currently monitors Atticus, JND Legal Administration, Verita (formerly KCC), and ClaimDepot, and checks them daily for new and updated cases.

This means the source for any settlement is the administrator's own page for that case — not a press release, aggregator, or another settlement site. Cases in litigation are found separately, from newly filed federal complaints.

2. How each summary is written

The administrator's page is fetched and passed to a large language model (Anthropic's Claude Haiku 4.5), which fills in a fixed schema rather than writing free-form prose: what happened, who qualifies, how to claim, claim deadline, exclusion and objection deadlines, final hearing date, settlement fund, estimated payout, whether proof of purchase is required, the class period, and the case's court and docket number.

The model is instructed to leave a field empty rather than guess, so a blank on a settlement page means the source page didn't state it — not that it doesn't exist. A page that doesn't describe a real class action settlement, or that has no deadline and no claim link, is discarded rather than published.

3. What gets verified before publishing

• Links. Claim and official-site URLs are requested and must actually respond before they're published. Language models are prone to inventing plausible-looking claim URLs, and an invented link on a settlement page is worse than no link.
• Continued existence. A maintenance sweep re-checks published links, clears ones that have gone dead, and demotes settlements whose site has disappeared entirely.
• Claim status. A settlement's status is derived from its deadline and the administrator's own wording, so cases move to “closed to new claims” on their own rather than sitting open forever.

4. Court records

Where possible each settlement is matched to its real federal court docket via CourtListener/RECAP, the Free Law Project's public archive of U.S. court filings, using the court and docket number extracted from the settlement page. When a docket is matched, filings are read and key events are labelled — preliminary approval, final approval, fairness hearing, objections, appeals, and distribution to the class — and shown on the settlement's page.

This is the part of a settlement that's hardest to follow elsewhere, and it's the honest answer to “I filed months ago, where is my money?”

5. What this process can't catch

Stated plainly, because these limits matter more than the capabilities:

• Summaries can be wrong. They're machine-generated from a source page and are not reviewed by a lawyer. Confirm eligibility, deadlines, and payment terms on the official settlement site before you file. Where they disagree, the official site is right and we are wrong.
• Most state-court cases have no live docket. RECAP covers federal courts. Roughly half the settlements here — many privacy, biometric, and employment cases — are in state court, so they get a status but no filing-by-filing timeline.
• Payout figures are estimates. Almost all settlements divide a fixed fund among everyone who files, so the real per-person amount isn't known until claims close. An “up to” figure is a ceiling, not a promise.
• Coverage is not complete. Settlements run by administrators we don't yet monitor, or handled entirely by a law firm, won't appear. Nothing here should be read as “this is every settlement that exists.”