AskTheRecruiter · AI Layoff Tracker

Data Sources

Every number in the tracker traces back to one of the sources below. That means an official government filing, a legally required layoff notice, an EU restructuring record, or a named news report. Nothing is estimated or modeled into existence. Each row links to the raw source so you can check it yourself.

← Back to the tracker · How we count · AI layoffs, in the employer's own words · Press kit and soundbites · Source code

Data last updated: Sep 5, 2026 · 5:12 AM EDT (the last time the database actually changed). Live collector status is on the health page. Note: this source list is a reference. It changes only when we add or remove a collector, which happens on a deploy, not every day. So it is honest for the list to stay the same between updates.

Every source at a glance

Verified means a filing or a named report is behind it. That is the headline number. Announced means a company plan reported at the announcement stage, before it takes effect. We keep it in a separate tier and never mix it into the verified total.

US state WARN registries we read (47 states + DC)

The federal WARN Act requires large employers to file advance notice of mass layoffs with their state’s dislocated-worker unit. We import those official notices daily from every state that publishes usable per-notice data. Each link is the state’s own official WARN page, the exact source our importer reads.

Two counts, and they answer different questions. The 47 states and DC above are the registries we read, every day. Elsewhere on the tracker you will see 46 US states and DC: that is the narrower count of jurisdictions whose notices are actually in the database so far. A registry we read that has not yet published a notice we could count appears in the first number and not the second.

Each state publishes on its own schedule: some daily, some weekly, some quarterly. We re-import every state daily at 11am ET, so a state’s newest posted notices appear here within a day. Where a state offers a page for each notice, the entry links straight to it. Otherwise the link goes to the state’s official WARN list or data file.

States not yet in the automated WARN feed, and why

All 50 states already appear in the tracker through SEC filings and news. The gap is only in state WARN notices. Each state below links to where it publishes (or an explanation of why it doesn’t), in plain terms:

“Publishes notices, no public headcounts” states post WARN notices without a per-employer employee count. Those notices cannot become countable rows, because we never invent a number. “No public register” and “Confidential by law” states keep WARN filings internal. In both cases those cuts still reach the tracker through SEC filings and named news, never invented and never estimated. The unemployment column is a separate official metric. It is the state’s monthly BLS rate, not a layoff count. We show it so every state carries an authoritative, sourced number even where individual notices are not public.

Why most countries appear through news, not a registry

Almost every country requires employers to notify a labour authority before a mass layoff. But most treat those filings as confidential. They publish totals only, never a public list of which companies are cutting. Only US states, Quebec and Poland’s Mazovia region publish a public, per-employer notice register we can read directly. The Mazovia register is WUP Warszawa, and a 2026 survey of all 16 Polish voivodeship labour offices found it is the only one that names employers. Everywhere else, the honest options are two. One is the EU’s Eurofound ERM, which compiles large restructuring events from national correspondents. The other is a reviewed allowlist of that country’s news outlets. That is why a German or Japanese layoff reaches this tracker through a named news report rather than a government file. The government has the file. It just does not make it public. We link each country’s official labour authority below, so you can check the filing requirement, and its confidentiality, yourself.

Worldwide news discovery: the countries and outlets we scan (180 configured country markets, 759 outlets)

Beyond official filings, we run worldwide discovery across a reviewed allowlist of reputable news outlets in the countries we cover, in 65+ languages. We check them once daily through GDELT and Google News, never the open web. Coverage varies by country, and each country’s measured coverage is shown below. The full list is built straight from the collector’s own configuration, so it updates automatically whenever a source is added. Each country also shows which official register, if any, we pull directly.

Every publisher we researched, connected or not

This is the full publisher research behind the tracker, country by country. Connected means a collector reads that publisher’s own feed. Researched, watched through its market sweep means the publisher is known and its stories reach us through its country’s own news edition, searched in that country’s own language, but we do not read a feed of its own. Researched, not connected means we checked it first hand and could not use it, and the reason is printed next to it. A country we hold nothing on should be a row that says why, not a gap you have to notice.

382 publishers across the research set: 19 connected, 336 watched through a market sweep, 27 researched and not connected. Feeds were probed first hand on 2026-08-14. Every refusal was measured at least twice. We never bypass a paywall, a bot wall or a certificate error, and we follow every robots.txt.

How verification works

Live collector status is on the tracker health page. The running corrections log is on the tracker itself.

How we classify an “AI-attributed” layoff (the rubric)

This is the number most likely to be quoted, so here is the exact, testable standard behind it. We report what the employer said, not our own judgment of cause. We never assert in our own voice that AI caused a layoff.

Two honest caveats. (1) The phrase “restructuring around AI” is sometimes PR cover for ordinary cost-cutting. The broad tier records the framing. It does not verify the cause. (2) Companies rarely say “replaced by AI” outright, so the strict measure is deliberately conservative. The broad measure exists precisely to show the wider, looser universe alongside it. We always report the two separately and never add them together.

Good reference sites (not sources we ingest)

Independent work we read and recommend. None of it feeds this tracker. No number on this site comes from any of them, and nothing below is counted, merged or reconciled with our figures. They measure different things from different data, which is exactly why they are worth your time alongside ours.

Why we name them at all: a reader checking our numbers should be able to find the other serious measurements quickly, including ones that may disagree with us. We link them because they are good, not because they agree.

Tip: click any column header to sort a table.