AI job exposure checklist: 6 questions to review your tasks

This page is general education, not career, legal or financial advice. The self-test below is an original teaching tool. It has not been validated against outcomes and does not predict what will happen to any person’s job.

Headlines about artificial intelligence and jobs often mix three different ideas: what AI could do, what it is actually used for, and what has happened to employment. Separating them is the most useful step before deciding whether your own work is at risk.

Exposure is not job loss

Researchers usually break each occupation into tasks and ask how many of those tasks generative AI could perform or speed up. A job with many such tasks is called exposed. The ILO states that exposure “does not imply the immediate automation of an entire occupation”, but rather the potential for a large share of its current tasks to be performed using the technology. ILO Working Paper 140.

When AI meets a task, two different things can happen:

• Augmentation: a person keeps responsibility and uses AI to do the task faster or better.
• Automation: the task is largely handed over to the AI.

Displacement is a third, separate idea: jobs actually disappearing or no longer being filled. Exposure measures potential; displacement measures outcomes.

How many jobs are exposed?

The two numbers differ because the IMF uses a broader definition of AI and a different exposure measure, not because one is wrong. The IMF also estimates that in advanced economies, about half of exposed jobs may be negatively affected, while the rest could benefit from higher productivity. IMF Staff Discussion Note SDN/2024/001.

The ILO finds clerical work most exposed, including data entry clerks, typists, bookkeeping clerks and general office clerks, while adding that in practice many of these tasks still need substantial human effort. It also finds higher exposure for women, mainly because of their concentration in clerical occupations.

What AI is actually used for

Usage data shows writing is central at work. A 2025 NBER study of how people use one popular chatbot found that writing accounted for about 40% of work-related messages in June 2025, and about two-thirds of writing requests modified existing text rather than creating new text. NBER Working Paper 34255.

Usage data from one AI developer has shown roughly half of conversations as augmentation and half as automation, with the balance shifting between reports. The January 2026 edition, for example, recorded 52% augmentation and 45% automation for its consumer product. Anthropic Economic Index, January 2026. This is one provider’s data and is not representative of all AI use.

Take the test: six questions about your tasks

Answer for your actual working day, not your job title. Each answer scores 0 (A) to 3 (D). Add up your points.

1. How much of your working day is reading, writing, summarising or answering text on a screen? A) Almost none · B) Some · C) About half · D) Most of the day

Why it matters: text-heavy tasks are where current generative AI is strongest, and clerical occupations rank highest in the ILO index.

2. Where does your work happen? A) On site, with my hands, in a different setting every day · B) Physical work, mostly the same routine · C) A mix of physical and screen work · D) Entirely at a desk

Why it matters: language models operate on screens, so unpredictable physical work is less exposed to them. Routine physical work faces other automation: employers in the WEF survey expect manual dexterity, endurance and precision to decline in importance by 2030. WEF Future of Jobs Report 2025.

3. If your work goes wrong, who is accountable? A) Me personally, with a licence or signature on the line · B) Me, directly to a client or patient · C) My manager checks my work · D) Nobody checks it individually

Why it matters: in regulated fields such as law, finance and medicine, privacy, safety and liability rules have slowed actual AI adoption even where the technology could help according to US research by the Yale Budget Lab and Brookings (Brookings, October 2025). Someone still has to sign and answer for the result.

4. How often do you face a problem nobody has solved before, with no template? A) Almost every day · B) Every week · C) About once a month · D) Rarely

Why it matters: employers in the WEF survey rank analytical thinking as the most important core skill and expect creative thinking to grow in importance. Novel, messy problems rely more on judgement. This reasoning is a teaching assumption, not a measured statistic.

5. How quickly can someone tell whether your work is good? A) It takes months or years · B) It takes weeks · C) A few hours · D) Almost instantly

Why it matters: if a result can be checked quickly, an AI-produced version can be checked quickly too, which makes the task easier to delegate. When quality shows only after months, as with advice, strategy or care, delegation carries more risk. This is reasoning, not a measured statistic.

6. How long have you been in your current line of work? A) More than ten years · B) Five to ten years · C) Two to five years · D) Less than two years, or just starting

Why it matters: the clearest early signal in US data concerns young workers. In the Stanford “Canaries in the Coal Mine” study (November 2025 version), workers aged 22 to 25 in AI-exposed occupations saw a 16% relative employment decline, while employment for experienced workers remained stable. An August 2026 update on the project page puts the gap at 19% relative to less-exposed peers. That study groups people by age, not by years in a line of work; the tenure prompt here is not a validated substitute. The decline came mainly through fewer hires rather than layoffs, and the authors describe the results as early, descriptive indicators rather than causal estimates. Stanford Digital Economy Lab.

Use your score as a discussion prompt

The bands and equal question weights are teaching choices, not cut-offs from any study. They do not map to the ILO or IMF exposure measures. Two people with the same job title can score very differently because their tasks differ.

What the labour-market evidence shows so far

Across the whole US labour market, the Yale Budget Lab’s September 2026 update found no clear evidence of disruption associated with AI: occupational churn, AI exposure among the unemployed and usage measures remained flat, within historical ranges or on pre-AI trends. Yale Budget Lab.

This is consistent with the Stanford finding above. The overall market can look stable while a narrow group, early-career workers in exposed occupations, shows weaker hiring. Both sources cover the United States only.

Employers surveyed by the WEF in 2025 expected, across all drivers of change including technology, demographics and climate, 170 million jobs to be created and 92 million displaced by 2030, a net increase of 78 million. The fastest-declining roles named include postal service clerks, bank tellers and data entry clerks. These are employer expectations, not measurements.

What history suggests

In the United States, roughly 60% of employment in 2018 was in job titles that did not exist in 1940, according to Autor and co-authors. Autor et al., QJE 2024. Most of today’s work was created along the way.

Bank tellers are a useful case. As ATMs spread in the US, the number of tellers needed per urban branch fell from 20 to 13 between 1988 and 2004, but banks opened more branches and tellers moved towards customer relationships. Bessen, IMF Finance & Development. Today, employers list bank tellers among the fastest-shrinking roles. The lesson is that jobs shift over time; they are neither permanently safe nor instantly replaced.

Plan a practical task review

Employers in the WEF survey expect about two-fifths (39%) of workers’ existing skills to change by 2030. Of 100 workers, 59 would need training. AI and big data, networks and cybersecurity, and technology literacy top the fastest-growing skills, alongside creative thinking, resilience and lifelong learning.

Job postings that ask for AI skills advertise higher pay. Lightcast’s 2025 analysis of more than 1.3 billion postings found salaries 28% higher, and 51% of such postings were outside IT and computer science. Lightcast, July 2025. Posted salaries are offers, not guaranteed earnings, and other analyses measure the premium differently.

Practical steps that do not depend on any forecast:

1. Use AI tools on your own tasks to learn where they help and where they fail.
2. Move towards tasks that need judgement, accountability and work with people.
3. Learn an adjacent skill before your role formally requires it.
4. Keep a financial buffer so a change in work does not force rushed decisions; see how to start an emergency fund.

To see how a change in pay would affect your monthly and hourly income, use the salary calculator, and read whether your pay rise gave you more buying power.

A note on “pointless” jobs

The worry is not new. In 2015, 37% of British workers told YouGov their job was not making a meaningful contribution to the world; a 2024 repeat found 33%. YouGov, April 2024. That is a self-assessment of meaning, not of exposure to AI. A useful job can still be highly exposed, and the test above is about tasks, not value.