AI Job Exposure Calculator
How exposed is your job to AI? Score your actual task mix.
Formula
exposure = routine% × 0.6 + cognitive% × 0.5 − physical% × 0.45 − human% × 0.4
Example
40% routine, 30% writing, 10% physical, 20% interpersonal → 46/100, moderate exposure.
Embed this calculator on your site
Add this free calculator to your own website with one line of code. The embedded version is responsive, ad-free, and includes a small attribution link back to CalcNest AI.
<iframe src="https://calcnestai.com/embed/ai-job-exposure-calculator.html" width="100%" height="700" frameborder="0" style="border: 1px solid #e5e5e5; border-radius: 12px; max-width: 720px;" loading="lazy" title="AI Job Exposure Calculator — Free Tool by CalcNest AI"></iframe>
What AI Exposure Research Actually Says
Tasks, not jobs, get automated
Every serious study — from Frey-Osborne's famous 2013 '47% of jobs' paper through OECD's task-based corrections and the post-ChatGPT exposure work by OpenAI and academic labs — converges on the same structural point: occupations are bundles of tasks, and automation eats specific tasks, not titles. The bank-teller precedent is canonical: ATMs automated cash handling, teller employment rose for decades as branches multiplied and the job shifted to relationships and sales. Exposure scores measure overlap with AI capability, which historically predicts transformation far more often than elimination.
The LLM-era inversion
Pre-2022 models ranked jobs by routineness: assembly lines exposed, creative and analytical work safe. Large language models scrambled that — the 2023 OpenAI/UPenn exposure study found the highest-exposure occupations were writers, interpreters, tax preparers, and programmers, while the least exposed were cooks, mechanics, and roofers. The economic logic is Moravec's paradox with a price tag: cognition ships as software at near-zero marginal cost, while dexterous physical work still requires a $100K robot that can't unclog your sink.
Reading your own score usefully
A high score is information, not a verdict. The practical responses the labor-economics literature supports: shift your mix toward the tasks AI complements rather than substitutes (judgment over production, client trust over drafting, reviewing over writing), become the person who wields the tools rather than competes with them, and weight accountability — society keeps humans in the loop wherever liability lives, which is why radiologists still sign every AI-read scan. The wage risk is real even without job loss: task automation can compress pay in exposed roles while raising it in complementary ones.
Exposure by task type: a reference grid
AI exposure tracks task mix more than job title. This grid shows how the four factors push a role's exposure up or down, with example occupations at each level.
| Exposure level | Dominant task mix | Example roles |
|---|---|---|
| High (70+) | Routine cognitive, writing, analysis | Data entry, basic bookkeeping, routine writing, entry-level legal review |
| Moderate (45–69) | Mixed judgment and production | Marketing, mid-level analysis, general management, customer support |
| Lower (25–44) | Physical or trust-based | Skilled trades, in-person sales, teaching, healthcare support |
| Minimal (under 25) | Embodied, interpersonal, unpredictable | Nursing, emergency services, childcare, plumbing, electrical work |
The post-2023 inversion is visible here: routine physical work — the classic "automatable" category — turned out harder to replace than junior knowledge work, because robots are expensive and language is cheap. Plumbers now score safer than copywriters, backwards from every 2015 prediction.
Common misreads of AI exposure
- Reading exposure as elimination. Jobs are task bundles; automation eats tasks, not titles. High exposure historically predicts transformation far more often than job loss — recall bank tellers rising after ATMs.
- Assuming physical work is most at risk. LLMs inverted the old hierarchy. Embodied, dexterous work resists automation because robots cost far more than software.
- Ignoring adoption lag. Capability runs ahead of deployment. Integration, regulation, and liability add years between what AI can do and what workplaces actually change.
- Treating a high score as a verdict. It's information for shifting your task mix toward what AI complements — judgment, trust, accountability — not a sentence.
What the exposure research actually says
Every serious study — from Frey and Osborne's 2013 paper through the OECD's task-based corrections and the post-ChatGPT exposure work — converges on one structural point: occupations are bundles of tasks, and automation eats specific tasks, not titles. The bank-teller precedent is canonical: ATMs automated cash handling, yet teller employment rose for decades as branches multiplied and the job shifted toward relationships and sales. Large language models then scrambled the old hierarchy. Pre-2022 models ranked jobs by routineness, with creative and analytical work presumed safe; the 2023 OpenAI/Penn exposure study instead found the highest-exposure occupations were writers, interpreters, tax preparers, and programmers, while the least exposed were cooks, mechanics, and roofers. The logic is Moravec's paradox with a price tag — cognition ships as software at near-zero marginal cost, while dexterous physical work still needs an expensive robot that can't unclog a sink. A high score is information, not a verdict: the responses the labor-economics literature supports are shifting toward tasks AI complements rather than substitutes, becoming the person who wields the tools, and weighting accountability, since society keeps humans in the loop wherever liability lives.
Frequently asked questions
Which jobs score highest and lowest on exposure?
Highest in current research: data entry, basic bookkeeping, telemarketing, routine writing and translation, entry-level legal review. Lowest: skilled trades, nursing and hands-on care, emergency services, childcare, and roles built on physical presence plus improvisation. Management sits mid-scale — meetings resist automation stubbornly.
My score is high. How worried should I be, honestly?
Directionally attentive, not panicked. Adoption runs slower than capability — integration, regulation, liability, and workflow inertia add years. The observable near-term effect in exposed fields is fewer entry-level seats and higher productivity expectations per person, which argues for accelerating past 'entry-level' tasks in your own skill mix now.
Is this based on a real methodology?
It's an honest heuristic distilled from the task-exposure literature (Frey-Osborne 2013, OECD task-based studies, 2023-era LLM exposure papers), weighting the four factors those studies repeatedly find decisive. It scores your self-reported task mix — a real assessment would decompose your occupation's O*NET task list.
My score is high. How worried should I actually be?
Directionally attentive, not panicked. Adoption runs slower than capability — integration, regulation, and workflow inertia add years. The observable near-term effect in exposed fields is fewer entry-level seats and higher productivity expectations per person, which argues for accelerating past entry-level tasks in your own skill mix now rather than fearing sudden replacement.
Which jobs are most and least exposed to AI?
Highest in current research: data entry, basic bookkeeping, telemarketing, routine writing and translation, entry-level legal review. Lowest: skilled trades, nursing and hands-on care, emergency services, and childcare — roles built on physical presence plus improvisation. Management sits mid-scale, since meetings and accountability resist automation stubbornly.