How the record is built

Methodology and corrections

Every row aggregates public statements. This page explains how those statements were sourced, tiered, and refreshed, and how to submit a correction. Hand this URL to anyone who challenges a number.

01How this was built

What counts, how it is graded, and how to correct it

What counts as an entry

A company enters the dataset when an executive, an SEC filing, an official press release, or a Tier 1 news outlet publicly attributes a layoff, hiring freeze, restructure, or headcount plan to AI, automation, or productivity gains from AI tools. Generic "restructuring" without AI attribution is excluded.

Tier 1 outlets

Reuters, Bloomberg, Wall Street Journal, CNBC, TechCrunch, The Information, Business Insider, Financial Times, plus SEC filings and company press releases. Challenger Gray & Christmas is treated as Tier 1 for aggregate US layoff totals. Trade press and analyst notes are treated as supporting sources, not primary.

Credibility tiers

Verified means a named AI tool, a measurable workflow or headcount change, and an executive on the record in a Tier 1 outlet or filing. Reported means credible Tier 1 press attribution, but one of those three elements is missing or paraphrased. Inferred means the AI attribution is partial or directional, supported by public disclosure but not fully confirmed. Two cases, HP Inc. and Cloudflare, are graded Verified without a named tool because their SEC filings causally join AI to the workforce reduction; those exceptions are set out in full under How we grade credibility and on the rows themselves.

Named tool vs. undisclosed

Every row records whether a specific AI vendor or product was named in the public statement. Several companies attribute headcount decisions to AI without naming a tool. That absence is preserved in the dataset and surfaced in the vendor block. It is not an accusation. It is a fact about disclosure.

Refresh cadence

Automated tasks run every Friday morning ET. An overnight sourcing sweep collects filings, earnings language, WARN notices, and news leads. Tier 1 collects new AI-attributed cases from Tier 1 outlets. Tier 3 tracks soft signals (analyst notes, rumors, labor pushback, 10-Q language) that are not yet Verified. A separate task refreshes the Challenger Gray & Christmas run-rate KPI. The changelog logs every change with a run date.

Corrections policy

If a company named here believes a row misrepresents their situation, or a number is out of date, email info@futurefluenthr.ai with the row, the correction, and a public source. Verified corrections are published in the next weekly refresh and logged in the changelog with the original value, the corrected value, and the source. The correction record is public. Requests to remove a row without a factual dispute are declined; the dataset aggregates statements that are already public.

How to read the briefing

Seventy-five separate cases across seventy-four mid-to-large companies have now publicly tied a workforce change to AI, from headcount cuts and restructurings to hiring rebuilt around AI. Deployments with strong metrics but no workforce decision (Chipotle, McDonald's) sit on the watch list, not in the record. The window opens in Jan 2021, two years before ChatGPT, so the timeline carries a pre-ChatGPT baseline; the first case lands in Apr 2023 and the record runs to Oct 2026: the AI-attribution era goes public in spring 2023, accelerates from May 2024, and steps again from January 2026. Each case is translated into what HR leaders should actually do about it. If you only want the record, jump to the company table.

What “AI-attributed” means

A workforce decision counts here only when a company or executive publicly named AI, automation, or AI-driven productivity as the reason. This is the same threshold Challenger Gray & Christmas uses in its monthly job-cut report. Generic “restructuring” does not qualify. Neither does “efficiency” without a named tool or a stated AI rationale.

Why the window opens in 2021

The observation window opens in January 2021, two years before ChatGPT’s launch on November 30, 2022, so the timeline carries a measured pre-ChatGPT baseline. The first case lands in April 2023, when layoffs stopped being framed as post-pandemic correction or general belt-tightening and companies began naming AI. IBM CEO Arvind Krishna’s May 2023 comments on pausing back-office hiring marked the start of the AI-attribution era in HR reporting. Earlier AI workforce programs with no announced workforce decision, Unilever's 2021 recruitment build the archetype, sit outside the record: qualification requires a decision, not just a deployment.

Who is (and isn’t) in the dataset

The dataset skews heavily toward companies over 1,000 employees, because that is where AI workforce decisions get publicly disclosed. Small-business signal is quieter and less press-driven, tracked in the weekly watchlist rather than the main record. If you lead HR at a sub-1,000-person org, read the mid-market cases as your closest analogs and the reversals twice.

See the mid-market cases in the explorer

How the company put it

The credibility tier says how solid the evidence is. This says how plainly the company itself tied AI to the cut, coded only from its own words or a named executive’s on the public record, never from the timing of events. Many companies are careful about naming AI as the reason for a reduction; this is how the record shows that care, case by case. Watch-list entries carry it too, except those with no workforce action. Each case card also quotes the company’s own words on the record about the cut, verbatim and linked to its source: where the company named AI, the sentence that does; where it did not, what it said instead. A memo seen only by reporters is never quoted as the company’s words, and a card without a quote means the company said nothing usable on the record.

Names AI as the reason: The company or a named executive says AI, or the efficiency AI brings, drives or enables this cut. Stored as stated-cause.
One of several reasons: AI is named as one reason among others, such as management layers, cost or structure, and not the lead one. Stored as contributing.
Cutting to fund AI: The cut frees money or capacity for AI strategy, products or teams. AI is where the resources go, not what does the eliminated work. Stored as reallocation.
Mentions AI, no link: AI appears in the same announcement or context, but the company does not tie it to the cut. A general remark about AI and jobs lands here too. Stored as adjacent.
Link only off the record: The AI link exists only in a leaked or internal document, anonymous sourcing, union claims or press inference; the company’s public account leaves AI out. Stored as off-record.
Denies AI caused it: The company says this reduction was not caused by AI. A general remark that AI will not cut many jobs is not a denial, and a denial of one wave does not carry to another. A denial removes a case from the record. Stored as denied.
Says nothing about AI: There is a workforce action and the company has said nothing about AI on the record. “Automation” without AI counts here unless the company equates the two. Stored as silent.

Which jobs a case touched

Each case records the kinds of jobs the cut touched, in plain words: customer support, engineering and product, managers, back office, sales and marketing, HR and payroll, content and media, operations and logistics, or consulting. The coding uses only what the company itself said about whose jobs went, never press reports alone, lists the main type first and stops at three. When the company never said, the case is marked not specified, and that is a finding in itself.

Each type points to its closest occupations in the federal Standard Occupational Classification (SOC 2018), linked to O*NET OnLine. Companies describe cuts by department while SOC classifies the work a person does, so these are the nearest occupation groups, not a count of the people cut. This page includes information from O*NET OnLine by the U.S. Department of Labor, Employment and Training Administration (USDOL/ETA), used under the CC BY 4.0 license. O*NET is a trademark of USDOL/ETA.

The download, column by column

Every case is in the CSV export, one row per case, in the order below. Numbers are numbers: the columns marked as such are safe to sum, sort and divide without parsing prose. Blank means the record does not carry the value, never zero.

Company: The employer named in the announcement. A company appears more than once when it announced separate waves carrying different evidence, so this is not a unique key.
Industry: The company's industry as the source describes it.
Country: Where the workforce action was announced, as free text. Use Countries (ISO) to group.
Announcement Date: The date as the source expresses it, which is sometimes a year or a range. Use Announced Month for anything computed.
Total Headcount Reduction: The reduction in the company's own words, including any qualifier it attached. Prose, not a number.
Reduction Percentage: The percentage in the company's own words.
HR Function Impacted: The HR work the company named, when it named one.
AI Tool / Platform: The AI system named in the statement, or a note that none was named.
Job Types: Which kinds of jobs the cut touched, in plain words: customer support, engineering and product, managers, HR and payroll, back office, sales and marketing, content and media, consulting, operations and logistics. Coded only from what the company said, the main type first and at most three, separated by semicolons. Not specified means the company never said whose jobs went.
Post-Automation Role Shifts: What the company said happened to the work and the people afterwards.
Credibility Tier: Verified, Reported or Inferred. The definitions are above on this page.
How The Company Put It: How the company itself described AI’s role in the cut, stored as one of seven values from stated-cause to silent. The values are defined above on this page.
HR Scope: A row-level note on how close the change sits to the HR function. Retired as a headline axis in September 2026; kept because Tier 1 still sets it.
HR Scope Note: The workflow the change touched, in a sentence.
Strategic Pivot: Whether the company also announced a strategic shift, in its own words.
What This Means For Your Role: Our brief for an HR reader. Editorial, and labeled as such.
Source URL: The primary source for the row.
Source Citation: The full citation, including any secondary sources named in prose.
Source Publisher: The publisher of the primary source.
Announced Month: The announcement month as YYYY-MM, or YYYY where the source names only a year. This is the field to bin by.
Announced Precision: month or year, saying which of the two shapes Announced Month holds.
Countries (ISO): ISO 3166 alpha-2 codes, space separated. Two rows name two countries.
Metric Status: disclosed where the row carries a headcount or a percentage, none where it carries neither. Seven cases are none.
Headcount Reduction (number): The announced reduction as an integer, where one was disclosed. Announced, not realised: it includes reductions delivered through attrition and multi-year plans.
Reduction Percentage (number): The disclosed percentage as a number.
Company Headcount: Approximate total employees at the company, for the denominator.
Share Of Workforce (percent): Headcount Reduction divided by Company Headcount, as a percentage. Computed on export, never stored, and blank where either side is missing.
Company Size Bucket: SMB, Mid-market or Enterprise.
Industry Parent: The industry rolled up to the group the charts use.
Source Type: What kind of document the primary source is: filing, company, wire, outlet, transcript, analyst and so on.
Source Tier: primary for a filing or company document, tier1 for the outlets the methodology names, then regional, trade or other.
Archived Copy: A Wayback Machine capture of the primary source, where one exists. 24 of the record's sources have none yet.
Last Reviewed: When the row was last checked by the weekly pass.

Accessibility

This site aims at WCAG 2.2 AA, and the parts that can be checked automatically are checked on every change rather than audited once a year.

Contrast. Every color pair in the stylesheet is computed against its real background in CI, and the weekly Playwright audit re-measures the text as rendered. A pair below 4.5:1 fails the build.
Keyboard. Every control is a real button or link: the case rows, the seat pills, the chart filters, the reversal timeline and the shelf toggles are all reachable and operable without a mouse, and each carries its state in aria-expanded or aria-pressed.
Structure. One h1 per page, headings in order, tables with real header cells and captions, and a skip link to the main content.
Motion. Everything that animates is inside prefers-reduced-motion, including the seat swap and the rail.
Without JavaScript. The whole record is pre-rendered into the explorer and the library, and every case has a static page of its own, so the dataset is readable with scripting off.

Known gaps: the charts are canvas, and while each carries a text description and the underlying numbers are in the tables and the CSV, they are not navigable point by point. The heatmap is dense on a small screen. If something here does not work for you, email info@futurefluenthr.ai and say what you were trying to do; that goes in the same queue as a data correction.

Independence. This dataset is maintained by Future Fluent HR. It is not sponsored by or affiliated with any company, vendor, or platform listed on this page. No listed company has editorial input.

Not legal, financial, or investment advice. The insights on this page are editorial commentary on public statements. They are intended for HR leaders, practitioners, and analysts as strategic context, not as a recommendation or a compliance opinion.

Contact. Corrections and press: info@futurefluenthr.ai. Live changelog: see this week's refresh.