UK

AI-proof score

Will AI take your job?

Name your job and we read what it is actually made of, from the role record in our graph. Then ten questions about your own week. You get the two side by side, where they disagree, and the qualification your role really routes to.

Find my score Discover how you learn
About 2 minutes No signup to see your result Grounded in named research

Ten questions about your week. One straight answer about your work.

Start here

Start typing and pick your job from the list, so we read the real role record rather than guess at a string. Then ten questions about your week. About two minutes, and the result shows straight away with no signup.

How this is put together, and the research behind it

Continue with Zavmo

Want to go further with Zavmo?

You have just seen a free read. If you want to go on and learn with Zavmo, here is how it would teach you, and who you would learn with.

Read the Zavmo manifesto to understand how you will be taught Meet your tutors

The method

It reads tasks, not job titles, and here is why that is the right question

The serious work on automation does not ask whether a job will be replaced. It asks which of the tasks inside that job can be written down as rules, and which cannot. That has been the mainstream position for over twenty years, and it is the position this tool takes.

So the ten questions are about your week, not your title. Two people with the same job title can spend their weeks doing very different things, and the research says it is the week that moves.

The ten questions, and what each one is for

  1. Routine and repetition 2 questions, one worded in reverse

    How much of your week is the same tasks in a predictable order, and how often the work hands you something you have never dealt with before.

    The routine-task hypothesis (Autor, Levy and Murnane, 2003). Rule-describable work is the work that substitutes first.

  2. Clear-cut, measurable work 1 question

    How much of your work turns clear inputs into clear outputs, with a right answer and quick feedback on whether you got it right.

    Task learnability (Brynjolfsson, Mitchell and Rock, 2018). A tight feedback loop is what makes a task learnable by a machine.

  3. Working with people 2 questions, one worded in reverse

    How much of the job runs on persuading, negotiating, coaching, caring and holding a room together.

    Social intelligence, the first of Frey and Osborne’s three engineering bottlenecks.

  4. Original and creative work 1 question

    How often the work needs something genuinely new rather than a variation on an existing template.

    Creative intelligence, the second bottleneck.

  5. Hands-on in the physical world 1 question

    How much of the work happens in unpredictable physical space: dexterity, movement, being there in person.

    Perception and manipulation, the third bottleneck, and the one Frey and Osborne single out as the constraint on the middle of the distribution.

  6. Judgement in the unknown 1 question

    When something ambiguous lands with information missing and no playbook, how much rides on your call.

    Non-routine problem solving in the Autor, Levy and Murnane sense: the half of the task space computers complement rather than replace.

  7. Adapting to change 1 question

    Your honest reflex when the tools change under you.

    Not an exposure measure. Exposure describes the job, and this describes what you would do about it, so it is reported on its own and never folded into the reading.

  8. One last check 1 question, not scored

    How well those answers actually captured your real working week.

    A calibration item. It sets the confidence level on your half of the result. It cannot move the reading up or down.

Where the three bottlenecks come from

Frey and Osborne estimated a probability of computerisation for 702 detailed occupations and put about 47 per cent of total US employment in their high-risk category. That figure is the one everybody quotes. The part worth borrowing is the method underneath it.

To decide what resists automation they identified three engineering bottlenecks: perception and manipulation, creative intelligence, and social intelligence. Those three are the three dimensions this page reads. We use them as a frame for what to look for. We do not use their data, their probabilities or their scoring, and nothing on this page is checked against their results.

Their own caveats matter here and we would rather repeat them than bury them. The estimate runs over what the authors call an unspecified number of years, perhaps a decade or two. And it describes an occupation being fully automated, which is exactly why it says nothing about how the tasks inside one particular week are shifting. That gap is the space this tool works in.

What the newer work adds

Felten, Raj and Seamans built a measure called AI Occupational Exposure by linking progress on specific AI capabilities to the abilities each occupation actually uses. Task-level again, not title-level.

Eloundou, Manning, Mishkin and Rock did the same thing for large language models. They report that around 80 per cent of the US workforce could have at least 10 per cent of their work tasks affected, and roughly 19 per cent of workers could see at least 50 per cent of their tasks affected. Read those two numbers together and the shape is clear: exposure is broad and shallow far more often than it is narrow and total. Most people are not facing a replaced job. Most people are facing a changed week.

Acemoglu and Restrepo went the other way and measured what actually happened to US local labour markets as industrial robots arrived. It is the useful corrective: exposure is a forecast, displacement is an outcome, and the two are not the same measurement.

Two readings, kept apart

Your result has two halves and they are never averaged. One half is what the record for your role in our graph is made of: its skills, responsibilities, decision-making authority and emerging skills. Your answers do not touch it. The other half is what you just told us about your own week.

An earlier version of this page blended everything into one number out of 100. It did not read your job at all, and it gave a nurse and a data-entry clerk the same 66. A single number implies a precision this reading does not have, so it is gone and it is not coming back.

Where the two halves disagree by more than one band, we say so and we show you what the role record says, so you can argue with it. That disagreement is the most useful thing on the page.

What this is not

  • This is not a validated instrument. There is no published reliability figure for it, because there has been no published study of it. It is nine scored self-report items and one calibration check, built on constructs that are well evidenced.
  • It does not predict whether you will lose your job. No model does that for one person, and the papers cited here do not claim to either.
  • Your half is self-reported, which means it reads how you see your week. The calibration question exists because that is worth stating rather than hiding.
  • The three dimensions use Frey and Osborne’s bottlenecks as a conceptual frame only. No value on this page is taken from, scored by, or checked against their data, or anyone else’s.
  • The role half is Zavmo’s own analysis of a role record. It is not a survey finding and not a labour-market forecast.

References

Every reference below was checked against the publisher record for its identifier before it was written here, and every one links out so you can read it yourself. Where a paper is used as a frame rather than as data, the note under it says so.

  1. Autor, D. H., Levy, F., and Murnane, R. J. (2003). The skill content of recent technological change: an empirical exploration . The Quarterly Journal of Economics, 118(4), 1279 to 1333.

    The routine-task argument this whole tool is built on: computers substitute for cognitive and manual work that can be written down as explicit rules, and complement people on non-routine problem solving and complex communication. Frame and method, not data.

  2. Frey, C. B., and Osborne, M. A. (2017). The future of employment: how susceptible are jobs to computerisation? . Technological Forecasting and Social Change, 114, 254 to 280.

    The three engineering bottlenecks (perception and manipulation, creative intelligence, social intelligence) that this page reads as its three dimensions. The paper estimates probabilities for 702 detailed occupations and puts about 47 per cent of total US employment in its high-risk category, over what it calls an unspecified number of years. Frame only: none of its data is used here.

  3. Felten, E., Raj, M., and Seamans, R. (2021). Occupational, industry, and geographic exposure to artificial intelligence: a novel dataset and its potential uses . Strategic Management Journal, 42(12), 2195 to 2217.

    The AI Occupational Exposure approach: score exposure from the abilities an occupation uses, rather than from the job title. Supports the task-level framing. No data from it is used here.

  4. Eloundou, T., Manning, S., Mishkin, P., and Rock, D. (2023). GPTs are GPTs: an early look at the labor market impact potential of large language models . arXiv:2303.10130. Published version: Science, 384(6702), 1306 to 1308 (2024).

    The two exposure figures quoted above, both read from this paper’s abstract: around 80 per cent of the US workforce could have at least 10 per cent of work tasks affected, and approximately 19 per cent of workers may see at least 50 per cent of tasks impacted.

  5. Acemoglu, D., and Restrepo, P. (2020). Robots and jobs: evidence from US labor markets . Journal of Political Economy, 128(6), 2188 to 2244.

    The measured-outcome counterweight to exposure forecasts. Cited to make the point that exposure and displacement are different measurements.

  6. Brynjolfsson, E., Mitchell, T., and Rock, D. (2018). What can machines learn and what does it mean for occupations and the economy? . AEA Papers and Proceedings, 108, 43 to 47.

    The task-learnability criteria behind the "clear inputs, clear outputs, quick feedback" question. Frame only.

  7. Ofqual (2026). Find a regulated qualification (the register of regulated qualifications) . Office of Qualifications and Examinations Regulation. Live register, accessed August 2026.

    Every qualification named anywhere on this site is a real, regulated one read from this register. Zavmo is not an awarding body and does not invent qualifications.

  8. UK Standards (2026). National Occupational Standards . The UK National Occupational Standards repository, accessed August 2026.

    The skills spine under every role in the catalogue. National Occupational Standards are the published statements of what competent performance in an occupation looks like.

Start the ten questions

Free, and no account needed to see your result.