AI and your job: what the federal numbers actually say

For three years, "AI is coming for your job" has been a claim without a scoreboard. Everyone had an opinion. Nobody had a federal number.

On August 27, 2026, that changed. The Bureau of Labor Statistics — the agency that counts American jobs — published a rating of how exposed each of 831 occupations is to artificial intelligence. Low, Moderate, High, or Very high, for every job it tracks.

We crossed that against BLS's own ten-year job projections. Here is the first thing we found, and it is not what we expected.

1. The number is smaller than the noise

Add up every job that BLS rates highly exposed to AI and projects to shrink by 2035, and you get 1.3 million jobs.

That is 0.76% of American employment. About one job in 131.

Over the same decade, BLS projects total employment to grow by 5.9 million jobs — up 3.5%.

We want to be careful here, because this cuts both ways. It does not mean AI is harmless, and it does not mean 1.3 million people losing work is a small thing to them. It means that the U.S. government's own ten-year projection does not currently describe an employment collapse. If you have been reading otherwise, the federal data does not back it up yet.

BLS also revises these projections every single year. This is a snapshot of what the government expects today, not a prophecy.

2. "Exposed to AI" is not the same as "going away"

This is the part most coverage gets wrong, and BLS says so in its own documentation.

Its exposure ratings, in its words, "do not distinguish between AI impacts from automation versus augmentation." A job can be highly exposed because AI will replace it — or because AI will make the person doing it twice as fast and twice as valuable.

The clearest example sits right in the data. BLS rates software developers Very high exposure. It also projects software developer employment to grow 15.8% over the decade. It rates office clerks Very high too, and projects them to fall 6.0%.

Same rating. Opposite futures.

So if you rank the states by exposure alone, you are not measuring risk. You are mostly measuring how many office jobs a state has. We checked: a state's exposure score tracks how white-collar its workforce already is at a correlation of 0.89. Four fifths of that ranking would just be restating something anyone could already see.

That is why we did not use it.

3. What we measure instead

We keep only the occupations that are both highly exposed to AI and that BLS projects to lose jobs. That is 100 of the 831. Then we weight each one by how far it is projected to fall, so a job shrinking 6% counts far more than one shrinking 0.3%.

The result is a single number per state: the share of that state's jobs expected to disappear from AI-exposed shrinking work. We call it AI Displacement Risk, and it appears as a Problem Intensity score on every state page.

The jobs doing the work are not exotic:

Cashiers alone carry between 16% and 21% of every state's total. This is not a story about robots. It is a story about checkout lanes, phone queues, and filing.

4. Where the risk actually is

Here is the surprise.

New Hampshire has the highest AI Displacement Risk in the country — 0.94% of its jobs, a Problem Intensity of 84.9. Not California. Not Washington. Not Massachusetts.

Why? New Hampshire levies no general sales tax, so it pulls in shoppers from Massachusetts, Maine and Vermont. Retail is 3.5% of its jobs against 2.5% nationally, and cashiers make up 18% of its at-risk total. The thing that makes New Hampshire a bargain for shoppers is the same thing that concentrates its workforce in the jobs most likely to thin out.

South Dakota is third. Bookkeeping and accounting clerks are 2.25% of its jobs against 0.89% nationally — two and a half times the national share — because of the financial back-office sector around Sioux Falls.

And at the bottom of the table: Massachusetts, 50th of 50. Colorado 49th. Washington 47th. California 46th. The states everyone associates with artificial intelligence have the least of their workforce in the jobs AI is projected to take, because they have the fewest cashiers and phone-support workers per person.

Eight of the ten highest-risk states have a U.S. Senate seat on the ballot in 2026.

5. And the coverage does not follow the risk

We measure two things for every problem: how bad it is (Problem Intensity) and how severely the local press treats it (Media Focus). The distance between them is the Perception Gap — the thing this whole site exists to show you.

For AI Displacement Risk, that gap is the widest we have seen on any problem.

Across the 35 states with a 2026 Senate race, the median gap is 37.8. Twenty-seven of the 35 cross the threshold we label a Significant Narrative Divergence. Only two states are Aligned.

Read the middle column again. South Dakota scores 9.1 for Media Focus and Kansas 6.7 — on a 0–100 scale. The press in the states most exposed to this is close to silent about it.

And the relationship between the two is essentially nothing: across those 35 states, Problem Intensity and Media Focus correlate at +0.147. Coverage of automation is almost unrelated to how much of a state's workforce is actually in the jobs projected to go.

The two Aligned states are the tell. Massachusetts — lowest risk in the country — has a gap of 1.0: mild problem, mild coverage, in agreement. Alaska is at 3.1. Where the problem is small, the press has it about right. Where it is large, the press is not there.

We are not claiming the media is hiding something. Automation stories are national by nature — a piece about cashiers is written once and syndicated, rather than reported out of Sioux Falls or Topeka. That is a structural feature of how news works, not a conspiracy. But the effect on a voter in Kansas is the same either way: the local signal about a real local exposure is faint.

6. What we are not claiming

We would rather you trust this number less and understand it better, so here is everything working against it.

The gap between states is thin. Every state falls between 0.66% and 0.94%. The worst state is 1.4 times the best, not five times. Problem Intensity spreads that across the 0–100 scale because that is how every Polipad score works — it shows you where a state sits relative to the others, not the size of the underlying number. Roughly one job in 350 separates the most and least affected state.

BLS does not blame AI for these declines. It projects cashier jobs to fall. It does not say AI is the reason — self-checkout and online shopping were shrinking that work before ChatGPT existed. We are selecting declining jobs that also happen to be AI-exposed. That is narrower than proving cause, and we are not going to pretend otherwise.

Pairing the two datasets is our decision, not the government's. BLS publishes exposure ratings and it publishes job projections. Putting them together is our inference. BLS states plainly that its exposure categories are "not a forecast of employment growth or decline."

The underlying research is already dated. BLS built its ratings from five outside studies whose view of AI capability stops around mid-2023, and which mostly look at language models — not image generation, not video, not robotics.

7. Why we did it this way

There was an easier path, and we turned it down.

Several respected academic indices score AI exposure by occupation. We could have picked one. The problem is that they disagree with each other in a way that would have decided our answer for us: one leading index puts blue-collar manufacturing states most at risk, another puts professional white-collar states most at risk. Applied to the same employment data, they produce opposite maps of America.

If we had chosen, the ranking you just read would have been a product of our choice rather than of the data. So we waited for a federal statistical agency to make that call in public, with its methodology written down. That is what arrived on August 27.

Both halves of this metric — which jobs AI can do, and which states have those jobs — now come from the same agency that counts American employment.

That is the standard we hold every Problem Intensity score to. It is why AI sat unmeasured on Polipad for months rather than shipping a number we could not defend, and why it is measured now.

8. What to do with it

Open your state. Look at where AI Displacement Risk sits among your state's other problems — for most states it is in the middle of the pack, well below housing or healthcare.

Then use the Crown Matcher to say how much you personally weigh it, and see which candidates have an actual federal record on labor-market and technology policy rather than the ones who talk about AI most.

That last distinction is the whole point of this site. Coverage is not severity — and on this problem, in 27 of 35 states, the two are further apart than on anything else we measure. A candidate's willingness to say the word "AI" is not the same as a record of doing anything about the jobs behind it.

AI Displacement Risk is built from BLS Occupational Employment and Wage Statistics (May 2025) and the BLS AI exposure categories released August 27, 2026, alongside the 2025–35 employment projections. It covers all 50 states and refreshes annually, when BLS publishes new figures.