ChatGPT, Gemini and Claude made different predictions on which professions will disappear

Three largest language models — ChatGPT-5, Gemini 2.5, and Claude 4.5 — gave drastically contradictory assessments of which professions and industries will suffer most from AI automation.

Models Did Not Agree With Each Other

Researchers asked each model to assess how different professions are vulnerable to artificial intelligence. The results diverged so drastically that they question the very idea of a universal vulnerability index. ChatGPT highlighted certain professions as high-risk, Gemini selected a completely different set, and Claude proposed a third variant. Particularly notable are the discrepancies in assessing creative specialties, technical positions, and management roles. If for one model a programmer was in the high-risk zone, for another their profession seemed relatively protected. The same applies to designers, analysts, and consultants — the three models see their future completely differently.

Why This Is Dangerous for Policy and Economics

These indices are relied upon by ministries, governments, and large corporations when planning workforce retraining, developing education, and shaping social policy. If three authoritative models give conflicting signals, there is a real risk of incorrect distribution of state resources. Imagine: the government invests billions in retraining thousands of workers in certain fields based on ChatGPT's forecast, but actual displacement occurs in completely different sectors, as Claude predicted. Workers will be retrained for an unnecessary field, and simultaneously a social crisis will occur in unprotected industries.

• Politicians don't know which index to rely on when choosing priority professions to support
• Companies may incorrectly plan automation investments based on contradictory forecasts
• Educational institutions are uncertain which skills to prioritize teaching

How This Happened

One of the main reasons is the different architectures and training data of these models. ChatGPT, Gemini, and Claude were trained on different sets of texts and have different sizes, so the assessments turn out to be subjectively colored. Each model interprets what "AI vulnerability" means in its own way. Additionally, vulnerability indices are crude tools in principle. They attempt to reduce the complex reality of the economy and technology development to a single number for each profession. In practice, AI's impact depends on many factors: region, salary level, company specifics, speed of technology implementation, and government political will.

What This Means

While AI is learning to assess its own impact on the economy, results will continue to diverge. This doesn't mean indices are useless — it's just time to view them as one of many signals, not as absolute truth. Politicians and employers need to combine data from different models, analyze local factors, and not rush into radical decisions based on a single index.