Clerks vs Care Workers: The Widest AI Exposure Gap in the Global Workforce

What the 9.0/10 score means for a general clerk

A general or keyboard clerk spends their working day on a specific and well-defined set of tasks: entering data into systems, routing documents to the right places, scheduling meetings and resources, drafting routine correspondence, and processing forms. Every item on that list is something that current AI systems can perform - at or above human speed and accuracy - using tools that exist and are being deployed by employers today.

Data entry: large language models read and transcribe structured and semi-structured data reliably. Document routing: AI classifies documents by content and destination. Scheduling: AI assistants manage calendars and resource allocation. Correspondence drafting: LLMs produce first drafts of routine business emails, memos, and notifications. Form processing: AI reads, extracts, and validates form content.

The 9.0 score reflects this nearly complete overlap between what a general clerk does and what AI can already do. It is not 10.0 because clerical roles retain some exception-handling function - the ambiguous document, the system error, the customer complaint that falls outside the defined workflow - and because some coordination tasks require human judgment in ambiguous situations. But 9.0 means the routine core of the job, the majority of hours worked, is directly replicable.

What the 2.0/10 score means for a care worker

A personal care worker's day is built around tasks that are almost entirely outside what current AI can do. Assisting a person with limited mobility requires physical presence and constant real-time adjustment to unpredictable conditions. Personal hygiene support involves intimate contact, dignity, and variable responses to an individual's changing state. Emotional support for dementia patients requires human presence, recognition, and attunement that cannot be scripted or systematized. Responding to a fall or a distress signal requires immediate physical judgment in an environment that is different every time.

Observing and reporting changes in a person's condition - a skill that experienced care workers develop over years - involves pattern recognition embedded in physical observation across time. AI systems can assist with scheduling care visits, reminding patients to take medication, and generating documentation. They cannot do the work itself.

The 2.0 score is not 0.0 because AI genuinely contributes to the administrative and monitoring side of care. Care scheduling software, digital care plans, medication management apps, and fall detection sensors all use AI components. But these are tools that support care workers, not replacements for them. The core of what a personal care worker does has no AI equivalent in 2026 and no credible roadmap for one within the next decade at scale.

Why the gap is 7 points and not more

The intuitive question is: if AI can do nearly everything a general clerk does, why not 10.0? And if AI cannot do almost anything a care worker does, why not 0.0?

Clerical workers are not at 10.0 for several reasons. Coordination tasks in many clerical roles require judgment in ambiguous situations: resolving a conflicting scheduling request, deciding what to do with a misfiled document that could belong in two different systems, handling a caller whose issue does not fit the standard workflow. Relationship management elements exist in many clerical roles, particularly those with regular contact with external parties. These are not large portions of the job, but they are genuine fractions that bring the score below 10.0.

Care workers are not at 0.0 because AI genuinely assists with care-adjacent tasks. Documentation that care workers are required to complete is increasingly AI-assisted. Scheduling of care visits and resources benefits from AI optimization. Medication reminders and some monitoring tasks (fall sensors, vital sign tracking) use AI components. The 2.0 score reflects that approximately 20% of what surrounds care work is AI-assisted, even though the core of the work is not.

The gap reflects reality, not a theoretical maximum. Both numbers are arrived at through the same structured scoring methodology applied consistently across all occupation groups in the WorldJobsData dataset.

The labour market consequences on each side

The two sides of the gap are moving in opposite directions in the labour market, and the data on this is consistent across multiple sources.

On the clerical side: the US Bureau of Labor Statistics (2024) projects -4.15% employment growth for clerical support workers over the next decade. This is against a background of +4% projected growth for overall US employment. Negative growth in an expanding job market is not a cyclical fluctuation - it is a structural contraction signal. The BLS projection does not depend on future AI development; it incorporates automation tools that already exist.

On the care side, the data points consistently toward shortage rather than surplus. The UK had 131,000 adult social care vacancies in 2024 (Skills for Care, State of the Adult Social Care Sector, 2024). Germany projects a deficit of 300,000 care workers by 2030 (Bertelsmann Stiftung, 2022). The World Health Organization projects a global health worker shortage of 10 million by 2030 (WHO, 2023). These projections are driven by demographic reality: aging populations in every high-income economy mean more people who need care, and no AI system makes that demand disappear.

The emotional reality on both sides

This analysis tries to be direct about what the data means, not just what it shows.

For clerical workers: the data is unsettling. When your role scores 9.0/10 and the Anthropic research (Massenkoff and McCrory, 2026) shows that the observed deployment gap for general clerks is 27 points (82% theoretical exposure, 55% currently observed AI exposure at work), the rational response is not panic - but it is not complacency either. The direction is set. The 27-point gap between theory and observed deployment is narrowing as enterprise AI adoption scales. Customer service clerks are already at an 8-point gap, meaning they are closer to the frontier. For general clerical workers, that gap represents runway - but runway with a measured length, not an unlimited horizon.

For care workers: the data is genuinely positive on AI risk, but the sector has a different structural problem that the data cannot solve. Personal care work is AI-proof and in structural shortage - which should, by basic economic logic, command higher wages. The fact that it does not is a policy and market-structure problem. Low wages in care are not explained by low demand or high replaceability. They are explained by how care work has historically been valued, who does it, and how it is funded. That is a problem the 2.0/10 AI score alone does not fix.

The 7-point gap as a career decision tool

The spread between 9.0 and 2.0 is not just a curiosity. It is usable information for career planning. The table below maps AI score ranges to what each range means in practical terms: