An AI jobs study found a shift toward senior staff

A press release from King's College London on 6 October 2026 carried the headline "AI adoption linked to increase in employment globally". On 9 October 2026, a post on X by unusual_whales went further: "AI is creating jobs, just not for young workers." The post cites the same study. Both rest on one paper by Bharat Chandar of Stanford University and Bouke Klein Teeselink of King's College London. I read the paper in full and checked each number against it. Its main finding is real and worth knowing. It is just not the finding in the headline.

What was claimed, and by whom

The press release says firm-level AI adoption "leads to around a 3.3% increase in the number of employees recruited overall". It also says "senior employment increases by 6.7%" and "junior employment falls by 2.5%". It quotes Chandar more carefully. In his words, the findings "suggest that generative AI is changing the mix of jobs within organisations rather than reducing overall employment." Klein Teeselink, the King's author, says it is hard to predict which jobs new technologies will replace, and that the study "doesn't fully settle that question". He then describes firms "employing fewer junior workers and more senior employees after adopting AI". In the paper, that fall in junior headcount is not statistically significant.

I read the full text of the X post through a public mirror of it, archived on 27 October 2026. The X post reads "Companies adopting AI are hiring more people overall", then "Senior employment: +6.7% over 5 years" and "Junior employment: -3%". It ends with the share figure: "The share of junior workers at those companies fell 1.9 percentage points."

What the paper actually measured

• Status. The paper is posted on the Stanford Digital Economy Lab's website. The PDF names no journal. I have not confirmed its peer-review status.
• Data. Revelio Labs records: 1.25 billion job postings and 154 million employment records across 41 countries, in a monthly panel from January 2021 to March 2026.
• Who counts as an adopter. A company that posted at least one job advertisement involving generative AI between November 2022 and March 2026.
• Design. 26,102 foreign affiliates of adopting multinationals, compared with 27,697 matched non-adopting affiliates in the same country and sector. To separate the effect of adopting from the kind of firm that adopts, the authors use a proxy: the share of other multinationals near the parent company's headquarters that adopted. In the paper's terms, adoption is instrumented with the adoption rate of other multinationals headquartered in the same metropolitan area as the parent.
• Who is junior. Revelio seniority levels 1 and 2, entry and junior. This is seniority, not age.
• Limits, in the authors' words. LinkedIn profiles "over-represent managers and professionals", and the estimates "should be understood as relevant to digitally attached multinational firms".

What it found, with its own significance language

Effects are measured from October 2022 to March 2026, against matched controls (Table 3).

Percentage points measure a change in a share. Percent measures a change in a count.

Table 3 reports the employment effects as changes in log employment, with standard errors. Converting those to percentages, I calculate an approximate 95% confidence interval from -0.6% to +7.5% for total employment. For junior employment, the same calculation gives about -6.8% to +2.0%. Both include zero.

In people, the paper says the average adopting affiliate had about 255 positions in October 2022, 145 of them junior. By March 2026 it employs "about 8 more seniors" and "about 4 fewer juniors" than it would have without adoption.

The paper also estimates 31 countries one at a time. The junior share fell in 23 of them, and seven of those falls are statistically significant. Total employment shows significant gains in two countries, Sweden and Spain, and no significant fall in any.

The paper's two checks without the instrument reproduce the share shift. The synthetic design, on the foreign-affiliate sample, finds total employment 5.4% higher, senior 7.2% higher and junior 1.5% higher. The junior share result is stable. The junior headcount result is not.

Where the claims go beyond the paper

1. Firm-level is not global. The paper says its estimate means adopting affiliates "grow relative to control affiliates, not that employment rises in the aggregate". Its model shows the two "can have opposite signs". It adds that growth at an adopting affiliate "may come at the expense of other parts of the same group". "Globally" in the headline describes the 41-country sample, not world employment.
2. Employment is not recruitment. The paper measures how many positions an affiliate holds, not how many people it hires.
3. The junior figure lost its caveat. The release reports "falls by 2.5%" with no mention that it is not significant. The X post rounds it to -3% and reads junior as young. The X post does report the 1.9 point fall in the junior share.
4. The time span grew. The X post says "over 5 years". The data run from January 2021 to March 2026, but the effects are measured from October 2022 to March 2026, which is 41 months.

Two details in the press release do not match the paper

The release says the study covered records "between 2021 and 2016". The paper's panel runs from January 2021 through March 2026, so that date is wrong. The release also says "158 million employee positions". The paper says 154 million, in the abstract, the data section and the conclusion. The release does not explain the difference. Neither detail changes the result, but both sit in the document most people will read.

Why the grade is overstated

I grade this overstated, not false. The paper does offer, in its abstract, "suggestive evidence of modest overall employment growth" at adopting firms. But the release headline can be read as a rise in total world employment, which the authors say their design cannot show. The release also calls headcount "employees recruited" and gives no measure of uncertainty. The X post goes further, with a rounded and uncaveated junior number and an inflated time span. A fair one-line summary: firms that adopted AI shifted toward senior staff, and may have grown a little compared with similar firms. Whether there are more jobs overall, this study does not say. I am confident in this reading, because it rests on the paper's own text.

Sources

• release AI adoption linked to increase in employment globally, King's College London, 6 October 2026 (archived copy)
• paper How Does AI Change Labor Demand? Evidence from 41 Countries, Chandar and Klein Teeselink, 20 September 2026 (archived copy)
• post unusual_whales on X, 9 October 2026 (archived copy)
• data Capture of the X post's text, retrieved 27 October 2026