How the lens works.

What the labels mean

AI can take it on

People widely hand this task to AI from start to finish. In the usage data it sits in the busiest quarter of work activities, and most of that use is people delegating the whole task rather than working through it.

Changing how it’s done

AI is regularly used on this task, but mostly as a collaborator: drafting, checking, iterating with a person in charge. Tasks in occupations the ILO rates as highly exposed also count here when usage is still light.

Stays human

Physical or hands-on care work, people-centred work such as supervising, negotiating and building relationships, or tasks where there's little observed AI use. People-centred work is never labelled “AI can take it on”.

Evidence still emerging

There's no AI-usage data for this task yet, not even for the broader activities it belongs to, and it isn't clearly physical or people-centred work. Rather than guess, we don't label it either way until evidence appears.

How a skill gets its label

1. Each skill in a role is one specific task from O*NET, the most detailed public list of what each job involves.
2. Each task is linked to the broader work activities it belongs to. We look up how much people use AI for those activities, and whether they hand the work over or work through it with AI, using April 2026 and May 2026 data.
3. We check the role's occupation against the ILO's global index of exposure to generative AI. Where the ILO rates an occupation as not exposed, none of its tasks can be labelled “AI can take it on”.
4. Each skill gets a confidence level. Directional means the usage data and the ILO index agree. Early signal means only one source covers it, or the evidence is indirect.
5. Your result counts your skills by label: how many AI can take on, how many are changing how they're done, and how many stay human.
6. We never give one exposure number. Instead we show a range for your week, weighted by how much of your week each skill takes if you tell us. The low end counts only skills AI can take on or that are changing, where the sources agree (Directional). The high end also counts early signals and skills whose evidence is still emerging. A wide range means the evidence is still thin.

Limits: the AI-usage data comes from one AI platform (Claude), so work people do with other AI tools, and automation by specialist software (route optimisers, robots, accounting systems), is under-counted. Exposure means the work changes, not that a job disappears.

Where the pay figures come from

Pay benchmarks are official statistics. For Australia, Singapore and the United States, official sources publish pay for specific occupations, so we show the pay of the closest official occupation to your role, and name it. Elsewhere we say plainly that there's no role-specific pay yet, and show the figure for a broad group, such as all professionals in India, underneath.

India. No official source publishes pay for specific roles in India; official figures cover only broad groups, such as all professionals. So for 17 roles we use the Michael Page India Salary Guide 2026: base salary ranges, excluding bonuses, by role, company segment and years of experience. It is recruiter placement data, mostly for mid-to-senior roles at larger companies, not official statistics, and it is named under every figure it supplies, with the segment it describes (for example "at larger companies" or "at global capability centres"). Where the guide has no figure for your role and experience, we don't show a large number: the broad official figure is shown only for comparison, behind the ⓘ.

Sometimes the closest official occupation is a poor fit. We say "No close official match" instead of showing its pay when it is a rare specialist occupation (under 0.0065% of the country's workers, about 10,000 people in the United States, where the source counts them), when its median pay is more than twice that of its occupation family, or when it is a different kind of work from the role's tasks. A role is matched to the same kind of occupation in every market.

Under each figure, one line names the source and year, who it covers and what the pay includes. Where the publisher's own notes don't say what the pay includes, it says "Pay basis not stated". Non-official sources say what the figure is based on instead, and "Not official statistics".

United States

US Bureau of Labor Statistics, Occupational Employment and Wage Statistics, May 2025 : Covers non-farm employees. Excludes overtime and most bonuses. Employees of non-farm businesses and government, May 2025. Excludes the self-employed, household workers and most farm workers. Includes commissions, production bonuses and tips; excludes overtime, shift pay and other bonuses. Annual figures converted to monthly.

International Labour Organization, ILOSTAT : Covers employees aged 16 and over. Includes overtime and tips. Employees aged 16 and over, from the US Current Population Survey, as processed by the ILO. The 2025 figure averages 11 months, because October 2025 wasn't collected.

Singapore

Ministry of Manpower, Singapore, Occupational Wages 2025 : Covers full-time residents at private firms with 25+ staff. Excludes bonuses. Full-time resident employees in private-sector establishments with 25+ staff, June 2025. Excludes bonuses, so pay is understated for roles where bonuses are large.

Australia

Jobs and Skills Australia, occupation profiles, using ABS Employee Earnings and Hours, May 2025 : Covers full-time non-managers on adult rates. Pay before tax. Full-time non-managerial employees paid at the adult rate, May 2025: median pay before tax and salary sacrifice. Weekly figures converted to monthly.

Australian Bureau of Statistics, Employee Earnings and Hours, May 2025 : Covers full-time adult non-managers. Includes overtime, not annual bonuses. Full-time non-managerial employees paid at the adult rate. Total cash earnings: pay for ordinary hours (including allowances, regular bonuses and commissions) plus overtime; excludes annual bonuses, payments in kind and other irregular payments. Weekly figures converted to monthly.All employees, full- and part-time, managerial and non-managerial. Total cash earnings: pay for ordinary hours (including allowances, regular bonuses and commissions) plus overtime; excludes annual bonuses, payments in kind and other irregular payments. Weekly figures converted to monthly.

Australian Bureau of Statistics, Employee Earnings and Hours, May 2025 (managers): Covers all employees, full- and part-time. Includes overtime, not annual bonuses.

India

Michael Page India Salary Guide 2026 : Recruiter placements, mostly mid-senior; base pay only. Not official statistics. Base annual salary, excluding bonuses, incentives and equity; from Michael Page's recruitment placements, mostly mid-to-senior roles at larger companies.

Malaysia

Department of Statistics Malaysia, Salaries & Wages Survey Report 2025 (Sumber: Jabatan Perangkaan Malaysia) : Covers citizens in full-time paid employment. Includes overtime, not bonuses. Malaysian citizens in full-time paid employment in the public or private sector (household survey). Basic wages, cost-of-living and other regular allowances, and overtime; excludes bonuses, gratuities, family allowances and social security payments.

Indonesia

International Labour Organization, ILOSTAT : Covers employees. Net pay after tax, in cash or kind. Employees, from Indonesia's National Labour Force Survey, as processed by the ILO.

Philippines

International Labour Organization, ILOSTAT : Covers employees. Basic pay only; excludes overtime and allowances. Employees, from the Philippines' Labour Force Survey, as processed by the ILO.

Europe

International Labour Organization, ILOSTAT : Covers employees aged 16 and over. Pay basis not stated. Employees aged 16 and over, from the EU Statistics on Income and Living Conditions survey, as processed by the ILO.

Currency conversion uses European Central Bank reference rates from 2026-10-09.

How your experience is used

You tell us your total years of work experience. We only show a figure for your experience where an official source publishes pay by age or experience. We don't treat a spread of pay as experience: the 25th percentile is not "a junior's pay".

• Singapore: estimated from pay by age. The Ministry of Manpower publishes median pay for each occupation at ages 25–29, 30–39, 40–49 and 50–59 (June 2025). We estimate your age as the age people in your role typically start work, plus the middle of your experience band (0–1 → 0.5, 2–3 → 2.5, 4–6 → 5, 7–9 → 8, 10–14 → 12, 15–19 → 17, 20+ → 22 years). The start age is 25 for roles usually entered with a degree and 21 for roles entered with a diploma, a trade qualification or from school; it's named under each figure. Each age group's median is placed at the group's middle age (27, 34.5, 44.5 and 54.5), and ages in between are read off a straight line. Under 27 we use the 25–29 figure, because MOM doesn't publish pay for under-25s, so the figure for someone under 25 may be a little high. We never go beyond the oldest group, and a figure never falls as experience rises. Where MOM hides the pay of the youngest groups, or pay at 25–29 is higher than at 30–39, we show the occupation's overall median instead, marked "not specific to your experience". Bonuses aren't included.
• The United States: a typical range, not an experience figure. The source gives the middle half of earners but not pay by experience, so we show the median and the range most people fall in.
• Australia: one typical figure. The source gives a single median per occupation, so we say we don't yet have pay by experience for this market.
• India: recruiter ranges by experience. The Michael Page guide gives base-pay ranges for bands such as 3–5, 5–10, 10–15 and 15+ years. We use the guide's band that contains the middle of yours (at a shared edge, the lower band), so 4–6 years uses 3–5 and 7–9 uses 5–10. Most of its tables start at 3 years, so there's usually nothing for 0–3 years.
• Everywhere else: the figure is for a broad group of occupations, such as all professionals, and isn't specific to your role or your experience.

If you enter your salary, it's compared with the estimate for your experience where there is one, and otherwise with the median.

Updates

The skill library is re-scored when new usage data is published, so labels can move. Last updated 11 October 2026, using data to May 2026, with scoring method version 1.1. The scoring code is public in the site's repository.

What we record

When you use the Skills Check, your browser sends anonymous events to our database, which is hosted in Singapore. Each event records a random ID created in your browser and one for this visit; the event (role chosen, skills confirmed, result viewed, "Was this useful?", PDF saved, a link clicked, a share button used, a waitlist joined, or a role you told us is missing); your role, job family, country, experience band and our scoring version; for the result, its counts and range; for the skills you confirm, their names from our library and how much of your week each takes; and for a missing role, the text you typed (up to 60 characters).

On the home page we count which of the two doors is chosen, and on Plans whether someone joins the waitlist. These are sent under a one-off random ID that isn't kept, with no country, role or visit, so they can't be linked to anyone. If you join a waitlist or the newsletter, your email goes to our newsletter service (beehiiv) only, never to these records.

We never record your name, email, IP address, custom skill text or the salary you type into the Skills Check. The only pay we keep is a range you choose to share, described below. Our database provider keeps standard server logs, which include IP addresses, for a short period for security; we don't copy them into our data. We use the events only to improve the Skills Check and its skill library, and delete them after 24 months.

Pay you choose to share (India only, optional). After your result, you can pick the range your monthly pay falls in. We keep only that range, with your role, country and experience band, the date, and a scrambled form of your browser's random ID, so each browser counts once. It's stored apart from the events above and isn't linked to them. We show it back only as "Reported by N users", beside the official figure and never instead of it, once at least 30 browsers have answered for that role and country over at least 14 days, with no single day supplying more than 25% of the answers. Suspected bots, and answers more than 3 ranges away from the typical one, are left out. Answers are deleted after 24 months.

You can turn this off at any time with the link under the Skills Check. If your browser sends a Global Privacy Control signal, nothing is recorded. If you choose Europe as where you work, we ask before recording anything, and nothing is sent unless you say yes.