A chart has been circulating for months. It shows a sharp, almost vertical decline in entry-level hiring starting around late 2022 — precisely when ChatGPT was released. The implication is clear: AI killed the first rung of the career ladder.
The chart has been shared thousands of times. It’s been cited in podcasts, newsletters, and Reddit threads. It’s been used to argue that AI is “coming for junior roles” and that the traditional career pipeline is broken.
But when you trace it back to its source, the story gets more complicated. The data is real. The interpretation is contested. And the chart itself — depending on which version you’re looking at — may be measuring something different than what it appears to show.
Here’s what the evidence actually says.
What the Chart Shows — and Where It Comes From
There isn’t one single “viral chart.” There are at least three distinct versions circulating, each with a different data source and a different claim.
Version 1: The “Rage Against the Machine” Chart
This is the version most commonly shared on X and Reddit. It plots junior headcount against time, showing a steep decline that begins around the GPT-3.5 release. The source is an SSRN working paper from 2025, and the chart has been widely shared as evidence that AI is eliminating entry-level roles.
The problem: the decline begins before AI’s mass adoption. As multiple analyses have noted, the downward slope in junior hiring predates ChatGPT by months, if not years. The chart shows a correlation, but the timing doesn’t support the causal claim.
Version 2: The Anthropic “Spider Chart”
In early 2026, Anthropic released a labor market impact report that included a spider chart showing AI exposure across different occupational categories. The chart was widely misinterpreted — and Anthropic itself later clarified that it measures tasks, not jobs.
The key distinction: AI exposure does not equal job elimination. A job can be highly “exposed” to AI — meaning many of its tasks could be automated — while still employing humans for the tasks AI can’t do. The Anthropic chart was never designed to show hiring collapse. It was designed to show task-level exposure. But once it went viral, that nuance was lost.
Version 3: The Stanford “Canaries” Chart
This is the version with the strongest empirical backing. In August 2025, a team led by Stanford economist Erik Brynjolfsson published a paper analyzing the impact of AI on early-career hiring, using large-scale, high-frequency administrative payroll data from ADP, which covers roughly one in six American workers.
The finding: a significant relative decline in employment for workers ages 22 to 25 in the most AI-exposed occupations since the widespread adoption of generative AI — even after controlling for other economic shocks. The effect has since been updated and expanded through the Stanford Digital Economy Lab’s Canaries Dashboard, which processes payroll data covering 4.6 million workers across 730 occupations.
The latest data shows that for workers ages 22 to 25, employment in highly AI-exposed occupations is shrinking at 3.8% per year, while employment for older workers in the same occupations remains stable or growing.
What the Original Source Actually Says
The Stanford paper is the most credible of the three. Here’s what it actually found — and what it didn’t.
What it found:
Early-career employment (ages 22–24) in the most AI-exposed quintile of industry-state cells declined by 12% over the 10 quarters following ChatGPT’s introduction.
The decline in hires is the primary cause of the employment decline, not increased separations.
The effect is observed across most sectors of the economy, not just tech.
The rate of hiring largely recovered by early 2025, but that’s attributable to a smaller employment base, not a return to pre-ChatGPT hiring levels.
What it didn’t find:
It did not find evidence that AI is causing mass job losses. The effect is concentrated at the entry level. Overall employment in AI-exposed occupations has actually grown since ChatGPT’s introduction.
It did not rule out alternative explanations. The paper notes that the decline in early-career hiring could also be related to the increase in remote work and increased educational attainment among workers in AI-exposed occupations.
It did not find that the decline began with ChatGPT. The paper’s triple-difference estimates provide “some evidence of earlier trend shifts” around the onset of the COVID pandemic.
The Counterargument: WFH, Not AI
The strongest challenge to the “AI caused the entry-level collapse” narrative comes from a 2026 paper analyzing the differential impact of work-from-home versus AI on entry-level hiring. Its conclusion: remote work hurt entry-level jobs more than AI has — so far.
The key finding: the decline in entry-level hiring pre-dates ChatGPT, and the slope of decline hasn’t changed much since. The Financial Times looked at the same data and reached the same conclusion: “Greedo (WFH) shot first, not AI”.
This doesn’t mean AI is innocent. It means the viral chart is oversimplifying a more complex story. Entry-level hiring has been declining for years — due to remote work, pandemic-era distortions, interest rate hikes, and shifts in how companies train junior employees. AI may be accelerating a trend that was already underway, not creating it from scratch.
The “Stupidest Chart in the World”
The most direct debunking of the viral chart came from TS Lombard economist Dario Perkins, who called a version of it — the one plotting S&P 500 against job openings — “the stupidest chart in the world”.
His argument: the chart implies that AI is destroying jobs, but the timing doesn’t work. Job openings peaked in March 2022 — months before ChatGPT was released — at more than 12 million jobs. That was the same month the Federal Reserve began raising interest rates to tamp down inflation. Job openings then started to stabilize in April 2024, just as ChatGPT traffic was really taking off.
Perkins’ conclusion: the chart reflects pandemic-era economic distortions and monetary policy, not AI-driven job destruction. The decline in job openings was the result of the labor market coming back into balance, not AI replacing workers.
This is a different chart from the Stanford one — it’s about total job openings, not entry-level hiring specifically. But it illustrates the same problem: viral charts often strip away context to tell a simpler, more dramatic story.
What the Data Actually Supports
If you strip away the viral charts and look at the underlying research, here’s what the evidence supports:
Claim | Evidence | Verdict |
|---|---|---|
AI has caused a collapse in entry-level hiring | Stanford Canaries data shows 12–16% relative decline for ages 22–25 in AI-exposed occupations | Partially supported — the decline is real, but concentrated and modest, not a “collapse” |
The decline began with ChatGPT | WFH paper shows the decline pre-dates ChatGPT; Stanford paper finds evidence of earlier trend shifts | Not supported — the decline began before AI’s mass adoption |
AI is replacing junior workers | AI is changing the task composition of entry-level roles, not eliminating them entirely | Nuanced — the job is being redefined, not eliminated |
Remote work is the real cause | WFH paper and FT analysis conclude WFH hurt entry-level hiring more than AI | Supported as a major factor |
The viral charts prove AI is destroying jobs | The charts show correlation without controlling for other factors; timing doesn’t align | Not supported — the charts oversimplify |
How to Trace a Viral Chart Yourself
If you see a chart making a strong causal claim, here’s how to evaluate it before sharing:
1. Find the original source. Not the tweet, not the newsletter, not the Reddit post. The underlying paper, dataset, or report. If you can’t find it, that’s a red flag.
2. Check the data source. Is it from a credible institution? Is the methodology transparent? Is the sample representative? The Stanford paper uses ADP payroll data — that’s credible. The “Rage Against the Machine” chart uses an SSRN working paper — also credible, but with different limitations.
3. Check the timing. If the chart shows a change starting at a specific moment, does the data actually support that? The viral charts show a decline starting around ChatGPT’s release — but the decline began earlier.
4. Check what’s being measured. Is it jobs, tasks, postings, or openings? The Anthropic chart measures tasks. The Stanford chart measures employment. The Perkins chart measures job openings. These are not the same thing.
5. Check for alternative explanations. Did something else happen at the same time? Interest rate hikes? Remote work shifts? Pandemic recovery? The strongest research controls for these. The viral charts don’t.
6. Check the effect size. A 12% relative decline sounds dramatic. But it’s concentrated in a specific age group and specific occupations. The aggregate effect is small. The viral chart makes it look like a cliff.
What I’m Still Trying to Figure Out
I don’t have this fully solved. Here’s what I’m wrestling with:
1. How do you measure “entry-level” consistently? Different datasets define it differently — by age, by experience, by job title, by salary band. The Stanford paper uses age 22–25. The SignalFire report uses “new grads.” The Indeed data uses job posting keywords. These aren’t interchangeable.
2. Is the Stanford effect causal or correlational? The paper controls for some factors but not all. The WFH paper suggests remote work is a bigger driver. How do you disentangle two trends that happened at the same time?
3. What happens to the “missing” entry-level workers? If companies aren’t hiring juniors, where are those workers going? Are they staying in school? Taking different jobs? Leaving the labor force? The data doesn’t fully answer this.
4. Does the effect persist? The Stanford dashboard is updated regularly. The most recent data shows the decline continuing for ages 22–25. But the paper notes that hiring “largely recovered by early 2025.” Which is it?
5. What am I missing? This is the big one. What source or perspective am I not aware of?
No feedback yet — submit the first.