How we calculate your score

1. The research base

We start with occupation level risk estimates drawn from 20 published research sources and industry reports, including work from OpenAI and the University of Pennsylvania, the Anthropic Economic Index, the International Labour Organization, Microsoft, McKinsey, Stanford and Felten, Raj and Seamans. Each source scores occupations on dimensions like task routineness, cognitive complexity, and susceptibility to generative AI. The largest sources, and how many occupations each covers, are on the data page, and every occupation page lists the specific studies behind its own score.

These are aggregated into a baseline exposure score for each of the 1,133 scored occupations in our database. The baseline reflects what the research says about the typical version of that job, before we account for anything specific to you. Most of that research measures overlap with what language models can do, rather than how much of the job machines could take over outright. The two are different questions with different answers, and we have measured how different.

2. Personal modifiers

Your quiz answers shift the baseline up or down. Each question targets a dimension that research shows materially affects individual exposure:

• Years of experience: Deep domain expertise is harder to replicate. Seniority also correlates with tacit knowledge that resists codification.
• Sector: AI adoption rates and regulatory constraints vary significantly by industry. Finance and legal are moving fast; construction and public sector face structural barriers.
• Company size: Enterprises have the capital to automate aggressively and the scale to justify it. Small companies often lack the resources, but also move faster when they do adopt.
• Current AI tool use: People already using AI tools are more likely to augment their role than be replaced by it. Getting hands on now is one of the more reliable ways to stay ahead of the shift.
• Routine level: Predictable, repeating tasks are the primary target of automation. High routine = high exposure.
• Digital vs. physical work: AI acts on data and language. Work that requires physical presence, dexterity, or real world situational awareness is substantially harder to automate today.
• Relationship complexity: Negotiation, trust-building, and emotionally nuanced interaction remain genuinely hard for AI. Roles centred on human connection are more resilient.
• Creative originality: Generative AI is a powerful tool for creative work, but original strategic thinking, taste-making, and novel problem-solving still require a human in the loop.
• Specialisation depth: Rare, niche expertise has a smaller training corpus and higher verification cost. The narrower your domain, the harder it is for a model to fully replicate you.
• People management: Leadership, coaching, and accountability require human presence and authority. Management roles carry a structural moat against automation.
• Career goal: Not a risk modifier: this shapes what your full report emphasises, deepening resilience if you want to stay, or identifying pivot targets if you're open to change.

3. Risk categories

Your adjusted score (0 to 100) maps to one of three categories:

4. Percentiles

We show two percentile comparisons on your free report:

• Within your occupation: Where you sit relative to others in the same job, based on quiz-modifier variance. Two people with the same title can have meaningfully different exposure depending on their sector, seniority, and working style.
• Across all occupations: Where your role sits in the full distribution of 1,133 occupations scored in our database. This gives you a sense of absolute exposure, not just relative-to-peers.

While an occupation has only a small number of real assessments, its percentile is primarily estimated from the research-based score distribution for that occupation, and it calibrates toward real assessment data as volume grows. The cross-occupation percentile is suppressed entirely until we have enough data volume to make it meaningful.

5. What this is not

This score is a research-aggregated signal, not a prediction. A few important caveats:

• AI capabilities are advancing faster than any static dataset can track. A score calculated today may look different in 18 months.
• Individual organisations vary enormously in AI adoption pace. The same job title can be high-risk at one company and low-risk at another.
• Automation doesn't always mean elimination. Many displaced workers upskill into adjacent roles or AI-augmented versions of their current one.
• Your answers are self-reported. Honest answers produce more accurate results.
• This is not financial, legal, or career advice. It's a starting point for an informed conversation, not the final word.