# How the exposure score works Tideproof's AI exposure score, out of 100, is a weighted sum of the ratings that hai, our scoring model, gives to the work you describe: routine information work (38%) and digital output (18%) push it up, while physical presence (16%), trust & accountability (16%) and novel judgement (12%) pull it down. ## What hai rates hai, our scoring model, reads the work you describe and rates it on 5 independent dimensions. Each rating is a value from 0 to 1, shown as a percentage next to its bar in your result. It also checks that the text describes real work at all (if it does not, you get no score), notes which field the work belongs to, and says how fully the result should be shown to you. None of those three moves the score. - **Routine information work:** Share of tasks current AI already does well. hai rates how much of what you describe is routine information processing, such as first drafts, summaries, standard analysis, code written to a clear specification, data entry and answers to common questions. - **Digital output:** How much of your output is text, code or data. hai rates how likely it is that the main thing your work produces is digital, such as documents, code, designs, spreadsheets or analysis. - **Physical presence:** Hands-on, in-person, real-world work. hai rates how much the work needs you physically there: hands-on skill, being on site, or coping with unpredictable real-world settings. - **Trust & accountability:** Value tied to you, your licence or relationships. hai rates how much of what the work is worth depends on you personally: the trust people place in you, the accountability you carry, your relationships, or a licence or legal responsibility you hold. - **Novel judgement:** Problems without a known procedure. hai rates how often the work brings new, ambiguous problems that no established procedure covers. ## How the score adds up The score is a weighted sum of the 5 ratings, multiplied by 100 and rounded to a whole number. Routine information work and digital output count towards exposure. Physical presence, trust & accountability and novel judgement count against it: for those, the sum uses one minus the rating, so more of them means a lower score. The weights add up to 100%. Each weight is also the most its dimension can move the score, in points: routine information work alone can move it by up to 38. ## Bands and range The score is reported on a 0 to 100 scale but kept between 3 and 97: a description that computes lower is reported as 3, and one that computes higher as 97, so 0 and 100 never occur. Your result also names the band the score falls in. - **Low:** 3 to 29 - **Moderate:** 30 to 49 - **High:** 50 to 69 - **Very high:** 70 to 97 ## The 600-character limit Before hai sees your text, control characters are replaced with spaces, every run of spaces and line breaks becomes a single space, spaces at either end are dropped, and only the first 600 characters are kept. Anything after that is not scored, so put the work that fills most of your week first. A description shorter than 12 characters after that is sent back with a request for more. ## How your 3 blocks are picked Your plan is drawn from all 6 building blocks: Skills AI amplifies, Ownership, Income streams, Trust & network, Place & cost base and Health & energy. Each block starts from a base need of its own, your ratings and your score add to it, and the 3 with the highest need become your plan, highest first. The reason shown under each block is fixed text chosen by your numbers. hai never writes prose, so it cannot invent advice. - **Skills AI amplifies:** Rises with your routine information work rating. - **Ownership:** Rises with your digital output rating, and less strongly with routine information work. Higher if you are trying to get into the work. - **Income streams:** Rises with your overall score. - **Trust & network:** Rises as your trust & accountability rating falls. Higher if you are trying to get into the work, whatever the rating. - **Place & cost base:** Rises as your physical presence rating falls, because remote-capable work can move somewhere cheaper. - **Health & energy:** Rises with your physical presence rating. ## If you are trying to get into the work Next to the description you say which applies: "I do this work now" or "I'm trying to get into it". It never changes your score, your band or your ratings: both answers come from the same single rating of your description, and switching after a result only swaps the plan. It changes two things. If you are trying to get in, Trust & network and Ownership weigh more in your plan: proof of your work and people who vouch for you, and, where the work does not hang on a licence or on personal accountability, earning in the field on your own. And from a score of 30 up, your result adds a note that hiring can close before the work does. The reason is a theory paper, not a measurement. Pascal Stiefenhofer’s "When Machines Apply for Your Job: Human Employment and Displacement Under Productive AI and AGI" (Economies, October 2026, https://www.mdpi.com/2227-7099/14/10/469) models firms that stop posting vacancies for people while a person already in the job still produces more than the job costs, because recruiting costs money and may not find anyone. In the paper that order holds under its example’s assumptions, not always, and it says nothing about how far apart the two points are for any real job or date. So we show no separate number for it, only the note and the change of plan. ## Limits of the estimate The score is a model estimate made from the words you type. It is not a forecast about your specific job, your employer or any timeline, and it knows nothing about you beyond those words. It rates the description, not the job title. Different wording, or a different emphasis on the same work, can produce a different score, so describe concrete tasks from a normal week. Exposure means how much of the work, as described, is the kind current AI already does well. On its own it says nothing about whether a job disappears, changes or grows. The confidence figure shown with your result is how sure hai was of its ratings. It does not change the score. If hai is unavailable, simpler keyword rules score the text instead, and your result says so. The scan is education, not advice: nothing here is financial, investment, legal or tax advice, and a score should not be the only basis for a career or employment decision. ## How much of this you see The same answer that rates your work also says how fully to show you the result: plain wording with one thing to start on, the standard view, or everything opened at once. hai judges that from how you described your work and nothing else. It never changes your score, your band or your blocks: the same description gets the same numbers whichever way they are laid out. It is a guess, so it is gated. If hai is less than 70% sure of the level it picked, the result is shown at the standard level instead of at a level nobody is confident in. And it is never final. Every result carries "Simpler" and "More detail" buttons that move it a step in either direction, as often as you like. The choice lives for that page only: no cookie, nothing stored, and nothing of it is attached to you. ## What is and is not stored Your text is sent over an encrypted connection to hai, our scoring model, and scored on five independent dimensions. The same answer also says how fully to show you the result, in plain wording or in full, which you can change yourself at any time with the "Simpler" and "More detail" buttons. We do not store it, and it is never attached to a waitlist signup: only the resulting score, band, sector and suggested blocks are. Running hai means sending your text to our AI model provider, which processes it for us under its data processing terms, may be located outside the UAE, and does not use it to train models. If you import a LinkedIn PDF, it is read in your browser and never uploaded; only the text you choose to scan is sent. We keep anonymous counts of a handful of moments: that a scan finished (with its exposure band, its sector, and the wording our model chose for it, which is plain, standard or detailed; if you then change the wording yourself, that press is counted separately and the two are never joined up), that a link to one of the tools we recommend was followed (which tool), and that a next step was followed, that "Next" was pressed, that a share button was pressed, that the full picture was asked for, or that you asked for the result to be simpler or more detailed. Each one is a count and nothing else: it carries no identifier, no IP address and no free text, it sets no cookie of its own, and it holds nothing from what you typed. Each count does carry the time it happened, so at low traffic a handful of counts a second apart could be guessed to belong to one visit; nothing in them identifies whose. They are stored in Cloudflare Analytics Engine, which holds them for three months. They are counted for every visitor, wherever you are, because a count that skipped a continent would tell us nothing. Our basis for keeping them is our legitimate interest in knowing, in aggregate, which parts of this page help and which do not. Your IP address is used to rate-limit the exposure scan and the page's layout calls per IP, and to prevent abuse. Our hosting provider, Cloudflare, keeps request logs for a short period as part of running and protecting the site, and sets one strictly necessary security cookie, named cf_clearance, which lasts up to a year and lets it tell a person from automated abuse. In our own account we keep one short technical line for each scan and each layout call, for three days: which engine answered, which model version, how long it took, how many tokens it used, and any error it returned. Each line also records which address on this site was called, and the reference numbers Cloudflare puts on a request so that one can be traced through its network; on their own those numbers identify no one. Those lines hold no IP address, no details of your browser, no location and nothing you typed, and the per-request record that would have held your IP address and browser is switched off. Page traffic is measured with Cloudflare Web Analytics, for every visitor, including in the EU: it is cookieless, sets no identifier, stores nothing on your device and does not follow you to other sites. ## Questions ### Why does my result look simpler than somebody else’s? Because hai also decides how fully to show it, from how you described your work. The numbers are the same either way. If it guessed wrong, the "Simpler" and "More detail" buttons on your result move it a step in either direction. ### Is the score a prediction? No. It is a model estimate of how much of the work you describe is the kind current AI already does well. It is not a forecast about your specific job, your employer or any timeline, and it is education, not advice. ### Why did my score change when I reworded my job? Because hai rates the text, not a job title. Different wording, or more weight on different tasks, can move the ratings and so the score. Only the first 600 characters are scored, so lead with the work that fills most of your week. ### What do the bands mean? They are fixed score ranges: Low is 3 to 29, Moderate is 30 to 49, High is 50 to 69 and Very high is 70 to 97. Scores never go below 3 or above 97, so 0 and 100 never occur. ### What raises or lowers my score? Routine information work (38%) and digital output (18%) raise it; physical presence (16%), trust & accountability (16%) and novel judgement (12%) lower it. Each percentage is that dimension's weight in the score. ### How are the blocks in my plan chosen? Each of the 6 building blocks gets a need from your ratings and your score, and the 3 with the highest need become your plan. For example, routine-heavy work lifts the Skills AI amplifies block, a digital output lifts Ownership, and a higher overall score lifts Income streams. ### Do you keep the text I type? Your text is sent over an encrypted connection to hai, our scoring model, and scored on five independent dimensions. The same answer also says how fully to show you the result, in plain wording or in full, which you can change yourself at any time with the "Simpler" and "More detail" buttons. We do not store it, and it is never attached to a waitlist signup: only the resulting score, band, sector and suggested blocks are. Running hai means sending your text to our AI model provider, which processes it for us under its data processing terms, may be located outside the UAE, and does not use it to train models. If you import a LinkedIn PDF, it is read in your browser and never uploaded; only the text you choose to scan is sent. ## How researchers frame exposure The two ideas this scan rests on, that the tasks in a job matter more than its title and that exposure is not a forecast, also appear in published research on AI and work. Quoted word for word; none of these sources reviewed or endorses Tideproof. > “Likewise, software may replace accountants who mainly prepare tax filings, but accounting jobs are varied, and those focused on other tasks may remain completely unaffected.” [Handel, Growth trends for selected occupations considered at risk from automation (Monthly Labor Review, U.S. Bureau of Labor Statistics, July 2022)](https://www.bls.gov/opub/mlr/2022/article/growth-trends-for-selected-occupations-considered-at-risk-from-automation.htm) > “Although new technologies may change the composition or weighting of tasks performed by workers in an occupation, sometimes dramatically, they may still have no employment impacts.” [Machovec, Rieley and Rolen, Incorporating AI impacts in BLS employment projections: occupational case studies (Monthly Labor Review, U.S. Bureau of Labor Statistics, February 2025)](https://www.bls.gov/opub/mlr/2025/article/incorporating-ai-impacts-in-bls-employment-projections.htm) > “We do not make predictions about the development or adoption timeline of such LLMs.” [Eloundou, Manning, Mishkin and Rock, GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models (arXiv, 2023)](https://doi.org/10.48550/arXiv.2303.10130) See where your own work lands: [scan your job free](https://tideproof.org/#scanner)