ChatGPT, Gemini and Claude do not even agree on which jobs are most exposed to automation, which is a pretty awkward look for a technology that loves to sound certain. Researchers tested ChatGPT-5, Gemini 2.5, and Claude 4.5 and found that their answers diverged enough to make ”AI job risk rankings” look more like educated guesswork than a reliable forecast.
The study matters because employers and workers are already treating AI tools like oracle machines. They are not. If the models disagree on something as basic as whether accountants are vulnerable, that is a reminder that the outputs depend heavily on training data, prompt framing, and whatever statistical soup sits underneath the polished interface.
Where ChatGPT, Gemini and Claude split
According to the researchers, the models were tested by economists Michelle Yin and Hao Wu of Northwestern University and Claudia Persico of American University. The results, as reported by The Wall Street Journal, showed sharp differences on occupations that blend mental and physical work, which is exactly the kind of job category people keep asking AI to judge with false confidence.
- Claude judged accountants to be highly exposed to AI disruption.
- Gemini disagreed and rated accountants as less at risk.
- Gemini also pushed back on risk assessments for ad managers and chief executives.
ChatGPT and Gemini were broadly aligned most of the time, but they still diverged in about 25% of cases. That is enough to make any neat-looking ranking suspect, especially when the ranking is being used to decide who should retrain, hire, or panic.
Why AI job risk forecasts keep wobbling
The researchers suggested the split may come from the different kinds of data each model was trained on. That is a polite way of saying the machines are not reading the same world, so they are not going to deliver the same prophecy.
There is also a bigger pattern here. A few months of hype have turned AI systems into career advisers, business strategists, and productivity gurus, but the models are still inconsistent even on questions they are specifically designed to answer. That makes the idea of using a single chatbot score to decide a career move look, at best, premature.
The jobs AI keeps arguing about
The source material also echoes a broader debate about which workers are safest in an AI-heavy economy. Nvidia chief executive Jensen Huang has argued that trades such as electricians, plumbers, welders, technicians, and builders will be among the most in demand, which is a neat reversal for a tech industry that spent years treating office work as the first domino to fall.
For now, the sensible takeaway is boring, which usually means it is right: do not quit your job because one chatbot drew a scary ranking. As economist Michelle Yin put it, people should not rely on a single metric to decide they need a new career or that their child needs a different major.
The next question is whether employers will stop outsourcing judgment to AI dashboards and actually compare forecasts against real labor market data. If they do not, the same models that cannot agree on accountants may end up shaping hiring decisions anyway.