AI was named in 120,136 job cuts this year. Nobody had to show the work.

On October 1, Challenger, Gray & Christmas published its September job cuts report. US employers have cited artificial intelligence in 120,136 announced cuts so far this year. That’s about 21% of all cuts, and it makes AI the leading reason of 2026. A day later the September jobs report came in at 29,000 new jobs against a forecast of 84,000, with unemployment up to 4.2%.

Those two numbers will get stapled together all month. Before they are, look at what the first one actually counts.

A stated reason is a sentence in a press release

Challenger tallies the reasons employers give when they announce cuts. It doesn’t audit them, and it doesn’t claim to. “AI” in that column means a company said AI. It doesn’t mean anyone checked which tasks a model now does, how well, or since when.

That matters, because the people running companies tell a different story when researchers ask them directly. Researchers at the Atlanta Fed, the Bank of England, the Bundesbank and Macquarie University surveyed nearly 6,000 CEOs, CFOs and senior finance managers between November 2025 and January 2026. More than 90% said AI had no effect on employment at their firm over the past three years. Eighty-nine percent saw no effect on productivity.

They can’t both be the whole story. Some displacement is real, and customer support is where it shows up most clearly. Salesforce has said its support headcount went from about 9,000 to about 5,000 after AI agents started handling customer conversations, according to CFO Dive. But MIT’s Paul Osterman put the other half plainly in Fortune this spring: “AI is a perfect excuse to justify big layoffs.” A reason that sounds like progress plays better with investors than “we hired too many people in 2022.”

So we have the most cited layoff reason in the country, and no standard for what it takes to claim it.

What real automation leaves behind

Here’s the useful part. If AI genuinely took over a body of work, that leaves evidence. Software that answers customers does specific things on specific days, and all of it can be written down.

Take a support team, since that’s the work we know. Honest automation of support produces a ledger with entries like these:

• Which topics the AI handles on its own. Order status, yes. Refund disputes, no. “Support” as a whole isn’t a topic.
• When each one went live, and who signed off on it.
• What it was tested on first. Real past tickets, with answers checked by the people who knew the job.
• What it still hands to a person, and how often.
• What it got wrong, and what changed afterward.

A company that has this can say something like “AI now resolves our shipping and account questions, which were 38% of volume, so we need fewer people on those queues.” That’s a claim you can argue with. A company that doesn’t have it is saying “AI” the way companies used to say “restructuring.”

If you run a business, keep this ledger even if you never plan to cut a single role. It’s how you’ll know whether the AI is working, and it’s the first thing you’ll want if anyone ever asks why a decision was made. If you’re a worker hearing “AI” in a meeting, it’s a fair question to ask for. Which work, exactly? Since when? Show me.

The second ledger is the money

The work ledger answers “did AI really do this?” There’s a second question that matters more to the people behind Challenger’s count. Where did the savings go?

Right now the default answer is the same place the old payroll went, minus the people. Nothing in a layoff announcement requires saying more than that.

Where we stand

We build an AI customer service agent, so we’re part of this story whether we like it or not. We’ve tried to build it so the first ledger exists by default. Celeste learns from a business’s own support history, then practices on real tickets privately, where nothing it writes reaches a customer. It goes live one topic at a time, only after the owner has read its answers and decided to trust that topic. Every graduation is a decision with a date and a name on it. Here’s how that works.

For the second ledger, we made a public commitment called The Dividend Standard: every year, we direct the greater of 5% of our revenue or 70% of our profit to the workers our AI displaces, granted as real ownership. It’s our answer to “where did the money go,” written down before anyone had to ask.

Neither of those fixes the national numbers. But when a company says AI did the work, it should be able to show you the work. And it should be able to tell you where the money went.