Type your job title into a box, wait about six seconds, and a website will hand you a percentage. Customer service representative comes back at 63 percent, labeled very high risk. Hairdresser comes back at 3 percent.
The tool is called Will AI Take My Job?, it is free, it does not ask you to sign up, and it covers 756 occupations from accountants to zoologists. It was built by Kickresume, a resume building platform, and it is currently doing what these things always do, which is travel around group chats faster than anyone reads the methodology page.
So here is the methodology page. The number is real data, it is more carefully assembled than most tools of this kind, and it does not mean what almost everyone reading it assumes it means.
The short version
- What it is: a free calculator scoring 756 US occupations for AI exposure, no account needed
- What it runs on: Anthropic’s Economic Index, which measures how people actually use Claude at work, cross referenced with the Labor Department’s O*NET occupation database
- The adjustment: up to 25 percent movement based on how much a job depends on creativity, people skills and physical dexterity
- The gap nobody mentions: the underlying data measures usage, not replacement. Anthropic’s own numbers say 57 percent of that usage is augmentation rather than automation
- The other gap: roughly 30 percent of the occupations in the tool have no real usage data behind them at all
What the tool actually says
The scores fall roughly where intuition puts them, which is part of why the thing is persuasive. Desk work built on structured, repeatable digital tasks scores high. Work that happens with your hands, in a room, with another person present, scores low.
| Occupation | Score | Why it lands there |
|---|---|---|
| Customer service representative | 63% | Scripted, text based, high volume, already heavily piloted |
| Data entry keyer | 63% | The purest case of structured digital work there is |
| Office clerk | 41% | Mixed bag of digital admin and physical errands |
| Software developer | 25% | Enormous measured AI usage, but the job is more than writing code |
| Teacher | 23% | Heavy AI use for prep, almost none for the part that is the job |
| Actor | 9% | Lower than the deepfake panic would suggest |
| Construction laborer | 3% | Physical, variable, unstructured, outdoors |
| Hairdresser, cosmetologist | 3% | Nobody is handing scissors to a robot this decade |
Two of those deserve a second look. Actors at 9 percent runs directly against the loudest fear in that industry, and the reason is that the tool is measuring the job as O*NET defines it, which is largely showing up somewhere and performing, not being a face that can be synthesized. Voice work, which is far more exposed, sits in different occupation codes.
Software developer at 25 percent is the one that should make you suspicious of the whole exercise, and it is worth explaining why.
The number is built from chat logs
Here is the machinery. Anthropic’s Economic Index takes a large sample of real Claude conversations, works out which O*NET task each one corresponds to, and produces a picture of which occupations are actually leaning on AI and how much. Kickresume takes that measured usage as the base score, then nudges it by up to 25 percent depending on how much creativity, human contact and physical dexterity the role requires.
That is a genuinely better foundation than the older generation of these calculators, which mostly asked an economist to guess how automatable a task list looked. This one is grounded in what people are doing, not what someone thinks could be done.
But grounding it in usage creates a specific and important distortion. A high score partly means people in that job use AI heavily. That is not the same as the job going away. In some cases it is close to the opposite: a job where everyone has quietly adopted a powerful tool and become more productive is a job that is changing, not necessarily one that is disappearing.
Anthropic’s own index splits measured usage into automation, where the model does the task, and augmentation, where it works alongside somebody. The split runs roughly 57 percent augmentation to 43 percent automation. More than half of the signal feeding these risk scores is AI helping a person do their job faster.
The research the tool is built on is more cautious than the tool
On March 5 this year, two economists working inside Anthropic, Maxim Massenkoff and Peter McCrory, published a paper on the labor market impact of AI. It is the closest thing anyone has to hard evidence, and its findings are notably undramatic.
They found a gap of roughly three times between the share of tasks AI could feasibly perform and the share it is actually performing in real workflows. Computer and mathematical occupations came in at 94 percent theoretical exposure against 33 percent observed. The capability is there. The adoption is not, or at least not yet.
They also found no clear spike in unemployment among workers in the most exposed occupations. Not a small spike. No clear spike.
What they did find is narrower and more worrying. Hiring of workers aged 22 to 25 into high exposure roles has slowed by roughly 14 percent since ChatGPT launched. The damage, so far, is not showing up as people losing jobs. It is showing up as the bottom rung of the ladder getting harder to reach, which is the sort of thing that does not appear in an unemployment figure for years.
The other finding worth sitting with is who is exposed. The workers with the highest measured AI exposure are, on average, highly educated, experienced, and earning well above the median. This is not automation coming for the lowest paid work first, which is the shape every previous wave took.
Why the same question gets different answers
Coverage of that one Anthropic paper split almost perfectly in half. Some outlets read it as evidence that AI is not taking jobs. Others led with the worst case the authors sketched and framed it as a possible white collar recession. Forbes ran a piece arguing the study does not measure labor market impacts at all, which is a fair critique of a paper measuring usage and drawing inferences about employment.
The calculators disagree too. A separate tool released earlier this year by sports analytics firm Action Network, using its own O*NET derived scoring, ranks computer programmers as its single most at risk occupation at 45 percent implied odds. Kickresume puts software developers at 25 percent and well down the list. Same underlying database, same broad approach, opposite conclusion about the most talked about job in the economy.
That divergence is the most useful thing on either site. When two carefully built tools using the same source data land that far apart, the honest reading is that nobody has a reliable method for turning task exposure into a probability of your specific job existing.
Reading your score honestly
- A high score means AI touches your tasks a lot. It does not mean your employer has a plan to replace you
- Check whether your occupation had real data. Around 30 percent of the 756 use an estimate instead of measured usage
- O*NET codes are coarse. Your actual job is probably a blend of two or three of them, and the blend is what matters
- The measured usage comes from one model’s users. Claude skews technical, which inflates some occupations relative to others
- The number that would matter is not on the site: what share of your role is the part a model cannot do, and whether you spend your week on it
The pundits are not doing better
Against all that, the confident predictions look thin. Palantir CEO Alex Karp has argued that vocational trade work is the one area that will be spared, which assumes robotics stays where it is. Bill Gates has suggested professional sports is the only job that stays fully human, on the reasoning that nobody wants to watch a machine play baseball, which is probably true and covers a rounding error of the workforce.
Neither of those is a forecast. They are intuitions, and the tools at least have the decency to show their working.
It is worth remembering how fast the ground has actually moved in the places where it moved. Amazon spent twenty years running Mechanical Turk on the premise that some tasks needed a human, and this year shut the whole thing down because real AI finally did the work it was built to route around. Meanwhile the volume of machine written material on the open web has reached the point where roughly one in three pages created since ChatGPT launched was not written by a person, which is a labor market change that arrived without a single layoff announcement.
The honest read
Use the tool. It takes six seconds, it is built on better data than its competitors, and the ranking it produces is directionally sensible. Screen work is more exposed than physical work, and that is a real finding, not a guess.
Then treat the number as a description of your present rather than a prediction of your future. Sixty three percent does not mean a 63 percent chance of losing your job. It means the tasks in your occupation are the kind that people are already handing to a model, in volume, today.
What actually decides the outcome is not in any database. It is whether your employer chooses to run the same work with fewer people or the same people with more output, and that is a management decision, made building by building, that no calculator can see. The evidence so far says most of them have not decided yet. It also says they have quietly stopped hiring as many people in their twenties while they think about it.
