The checker can contribute task-level change-pressure signals for proposed duties, but the supplied product description does not establish that it automatically compares your current role with a promotion. Build that comparison from recurring tasks in both roles, marking what is removed, retained, added or changed. Use signals to identify questions about local tools, review and accountability, not as a verdict or a probability of job loss.
Can the checker audit a promotion by itself?
It can contribute a task-level change-pressure signal for the work you may take on, but the product description supplied for this article does not establish that the checker automatically compares your current role with a promotion. Build that comparison from the recurring duties in each role, then use the checker as one input about proposed tasks. If the promotion outline is incomplete, keep those duties marked unknown; do not treat a provisional list or a signal as a verdict on whether to accept. That distinction matters because a promotion changes a bundle of work, not just a title. A move may remove some duties, retain others, and add or alter the rest. A checker’s task-level signal can help identify which proposed activities merit closer attention. It cannot, by itself, establish how much of your week they occupy, what your employer uses, or who will be responsible for the outcome. The International Labour Organization’s Workers’ Exposure to AI: What Indicators Tell Us—and What They Don’t explains that exposure indicators differ in construction and generally describe technological susceptibility from task information. They do not establish an employer’s adoption, reliable performance in a particular workflow, job redesign, labor demand, or an individual’s chance of displacement. A capability signal may tell you that a task resembles work AI could affect; observed use and rollout are local facts to confirm; staffing consequences cannot be read from exposure alone. So, can the AI Proof Work checker audit a promotion with a changed task mix? It can support your audit with task-level signals, within the limits described; the worker-built comparison supplies the before-and-after view. List recurring duties, preserve uncertainty where the role is undefined, and use flagged tasks to form questions about workflow or review. The next step is to lay out what is removed, retained, added, or materially changed before drawing a conclusion.
Sources: AI Proof Work AI Job Risk Checker; Workers’ exposure to AI: What indicators tell us – and what they don’t
How should you construct a fair before-and-after comparison?
Build the comparison from the work you actually do now and the work the promotion is expected to involve. Start with recurring duties, not the two job titles. Then mark each duty as removed, retained, added, or changed. The distinction matters because a promotion can change how a familiar task is performed, even when the task name stays the same. For each item, note how often it occurs or roughly how much time it takes in an ordinary period, if you can estimate that responsibly. If workload varies by season or project, record that rather than forcing one representative number. A task that appears rarely may still matter if it carries substantial decision authority or serious consequences when done incorrectly. Conversely, a frequent, low-stakes task may occupy much of the week while leaving little discretion. A count of tasks alone misses both differences. Record who supplies the inputs, what output is expected, who relies on it, what decision the worker can make, and who is accountable for the result. Note whether another person checks the work, what that review covers, and what happens when an error passes through. “Prepare a report” and “approve the recommendation in a report” may sound related, but they differ in authority, consequence, and responsibility. The second can be a changed version of existing work rather than a wholly new duty. Keep it in the changed category if the underlying subject is familiar but the worker’s role, judgment, or sign-off has shifted. Separate the duty itself from the way it is carried out. If the current role drafts routine summaries and the promoted role approves summaries prepared by others, record the continuing subject matter and the changed work: checking sources, resolving exceptions, explaining a decision, or signing off. Review is not automatically more or less exposed. The inventory identifies what changed; it does not decide in advance what technology can do. Give each duty a confidence note and identify where the information came from. A written promotion outline, a manager’s explanation, and your own assumption are different kinds of evidence. Label them plainly. If the outline says “coordinate quality” but does not say what is checked, how much time is available, or who owns the final call, leave those points open. Do not turn an ambiguous phrase into a definite task description just to complete the table. The missing detail identifies what to clarify before treating the proposed task mix as settled. O*NET can help you notice duties that a short outline may omit. The O*NET 31.0 Task Statements data dictionary organizes task statements by O*NET-SOC occupation and labels them core or supplemental. The data dictionary says each task is linked to exactly one occupation; “core” tasks meet stated relevance and importance criteria, while supplemental tasks are less relevant and/or important by those criteria. This makes the catalog a prompt for a completeness check: look at likely related occupations, scan for work that resembles the proposed role, and ask whether any such duty belongs in your actual inventory. It is not a checklist that every worker in an occupation performs. This boundary matters when a promotion crosses teams or combines specialist work with coordination. O*NET is a U.S. occupational reference, not a description of your employer’s work allocation, task frequency, authority, or promotion terms. A match cannot establish local adoption or consequence. Use it to prompt questions; use confirmed role details to describe your work. Compare the lists side by side and inspect each changed item. Is a duty removed or transferred? Does an added task include preparation, review, and follow-up? Does the same label conceal a different level of discretion? Keep frequency and consequence separate. Do not collapse rough estimates into a home-made total. The result is an account of what differs and what remains uncertain. It can show which proposed duties merit a checker signal and where that signal cannot capture local authority or review. If a major duty is still assumed, ask the role owner to clarify it and revise the list. Compare recurring work on both sides, with its source and context, rather than relying on titles or a score.
Sources: O*NET 31.0 Task Statements; O*NET Database
What does an exposure signal leave out?
A task-level exposure signal answers a narrow question: does this kind of work resemble something current AI systems may be able to perform or assist with? It does not show whether your employer has adopted a system, whether that system works reliably in your particular workflow, whether the job will be redesigned, whether demand will change, or whether you personally will lose work. For a promotion review, treat the signal as an early clue about a task, then investigate the separate steps between technical possibility and a change in your job. The International Labour Organization’s 2026 research brief, “Workers’ exposure to AI: What indicators tell us – and what they don’t,” explains why exposure measures can disagree. They use different methods and definitions, and many start from static lists of existing job tasks. Some indicators judge tasks against assessed AI capabilities; those judgments can vary. The brief also says these measures leave out adoption constraints such as economic conditions and institutional barriers. Keep the sequence clear: capability, exposure, adoption, workflow redesign, labor demand, and displacement. Capability concerns what a system can do under specified conditions. Exposure asks whether a task overlaps with that capability. Adoption asks whether an employer chooses and is able to use it. Redesign concerns what people do after a tool enters the workflow, including who produces, checks, handles exceptions, or approves an outcome. Labor demand concerns how much work employers or customers need and how that work is organized. Displacement is a possible employment outcome, not an automatic consequence of any earlier link. That boundary matters when a promotion changes your mix of duties. A signal for a task you expect to perform more often does not tell you whether your employer has a tool for it, whether it is being piloted or used in production, or what quality checks apply. Nor does it establish whether that task will take more or less time, free capacity for other work, or shift responsibility toward review. Those are local work-design questions. Ask what system, if any, is actually in use; which parts of the workflow it handles; what a person must verify; who has authority to approve the result; and who is accountable when it is wrong. Mark unanswered items as unknown rather than treating them as evidence of either adoption or non-adoption. Labor-market evidence adds context but does not settle a single promotion decision. The OECD’s “Artificial intelligence and jobs: No signs of slowing labour demand (yet)” reviews studies, surveys, and case material available for its 2023 Employment Outlook. It describes employment effects as ambiguous because displacement can be offset or accompanied by productivity and new-task effects. In the evidence period it reviewed, the chapter found little significant negative employment evidence, while noting that adoption can take time and that the record largely predates generative AI. This is a useful counterweight to claims that exposure already means falling employment. It is not proof that a current employer will preserve a particular role, or that demand in your location and occupation will hold. A promotion could redistribute existing duties inside one team even if broad employment measures show no clear decline; broad demand could remain steady while an individual’s responsibilities become more demanding. Use the signal to make your next conversation more precise, not more alarming. For each flagged task, ask the role owner about the actual tool, current status, workflow boundary, review standard, decision rights, and expected time. If the answer is “not decided,” keep that uncertainty visible and ask when the scope will be confirmed. If use is confirmed but checking standards are unclear, ask who sets them and whether review time is included in the workload. A signal can point to a question worth resolving; only the local facts can tell you how that task fits into the promotion you are considering. It cannot be converted into personal job-loss odds. A high signal identifies an issue to investigate; a low signal cannot certify that the task or role will remain unchanged or secure.
Sources: Workers’ exposure to AI: What indicators tell us – and what they don’t; Artificial intelligence and jobs: No signs of slowing labour demand (yet)
What can change when a task moves from doing to reviewing?
A promotion can reduce hands-on production while adding selection, explanation, exception handling, checking, coordination, or sign-off. These duties may still involve information-processing AI can assist with. Judgment and accountability shape how the task is done, but do not make it automatically immune or automatable. The useful question is which work moves, what authority comes with it, and what must be verified. Imagine an analyst whose current week centers on drafting reports. In the proposed role, the analyst may spend less time producing each report and more time prioritizing incoming requests, explaining results to colleagues, validating summaries, and approving recommendations. This illustration shows why “doing” and “reviewing” are not opposites. A draft may be generated quickly, yet someone must decide whether its source data fit the question, whether an unusual result needs investigation, and whether a recommendation is sound enough to approve. The distinction between allocation and decision rights matters. A manager may assign a worker to check an output without giving that worker access to underlying records, time to investigate, authority to reject the result, or a clear escalation route. In that case, “human review” describes a step in the workflow, not necessarily meaningful control. Conversely, an analyst may retain responsibility for a decision while using AI only to organize material or draft a first version. To judge the role, record separately who prepares, who checks, who can change the result, who signs off, and who is answerable if an error reaches a client or internal decision. The field experiment summarized in “Navigating the Jagged Technological Frontier” helps explain why task labels do not settle capability. Researchers evaluated 758 consultants at one firm on tasks spanning creativity, analysis, writing, and persuasion. For tasks the study placed within the tested AI frontier, the HBS AI Institute summary reports that ChatGPT-4 increased speed by more than 25%, human-rated performance by more than 40%, and task completion by more than 12%. A separate task beyond that frontier showed a decline in performance, illustrating uneven capability across tasks. This was a bounded experiment with consultants and constructed business tasks, not evidence that reviewing an analyst’s real recommendations is safe or unsafe. It supports checking precise work and errors, not treating seniority as protection. A different mechanism appears in “Measuring the Productivity Impact of Generative AI.” The study followed 5,179 customer-support agents as a conversational assistant was introduced in one software company. It reports a 14% average increase in issues resolved per hour, with gains concentrated among novice and lower-skilled workers and minimal impact for experienced and highly skilled workers. The authors suggest that the system could spread practices used by more effective agents. Assistance may make some learned procedures easier to access while leaving other expertise relevant. It does not show displacement, protect review work, or establish effects in other occupations. Its setting and workflow limit the comparison. Together, these studies make a narrower case than either “AI replaces routine work” or “senior judgment is safe.” A tool can speed some components, perform unevenly across related tasks, and alter who can produce an acceptable first response. If production becomes faster, the remaining work may shift toward checking, exceptions, explanation, and accountability; whether that shift happens, and how much effort it takes, depends on the local workflow. This is a question to test locally, not a reported outcome of either experiment. Count review time and interruptions in the task inventory. Faster drafting may not reduce total effort if checks grow harder or approvals accumulate. Do not classify the new duties as “human” and stop there. Ask what judgment is genuinely required: for example, selecting among conflicting evidence, interpreting an unusual case, or weighing consequences that a standard output does not capture. Then ask what evidence the reviewer can inspect, what kinds of error are plausible, how they are caught, and whether the review standard is clear. Check whether review authority and time match sign-off responsibility. If they do not, the central promotion question may be workload and governance, not simply AI exposure. Senior work can include AI-assisted tasks and decisions that still require a named person; neither fact alone settles the role’s value or future.
Sources: Navigating the Jagged Technological Frontier; Measuring the Productivity Impact of Generative AI; Artificial intelligence, job quality and inclusiveness
How can you read checker signals against the task inventory?
Use the checker on the proposed recurring tasks to the extent its task-level process allows, then annotate what each signal makes you want to verify. The supplied product description says it offers task-level change-pressure signals and first actions; it does not establish a feature that compares your current job with a promotion automatically. Keep the result beside your task inventory, not in place of it. A flagged task is a prompt to ask how that work is done locally. An unflagged task is not proof that it is immune to change, and neither result is a probability that you will lose your job. For each proposed task, write a plain-language description and mark it removed, retained, added, or changed relative to your current work. Note approximate frequency or volume, the checker signal, whether a tool is confirmed in use, being trialed, planned, or unknown, what a person must check, what could happen if the output is wrong, who approves or owns the result, and the open question. Keep ‘unknown’ distinct from ‘no’: not knowing whether a tool is used is missing information, not evidence that it is absent. A changed task deserves its own description. Preparing a draft and approving one may share a subject but differ in authority and consequence. Let the inventory reveal that difference rather than hide it under one familiar label. Treat frequency and consequence as separate annotations, not ingredients in a new score. Repeated low-consequence work may occupy much of the week, while an occasional approval may carry substantial responsibility. A signal cannot tell you how much of the role a task represents or how costly an error would be. Avoid combining a qualitative pressure signal, estimated time share, and consequence judgment into a single number: the assignment supplies no validated basis for treating them as commensurate. The result should be questions and possible skill needs, not a precise-looking personal risk score. Task wording matters. A broad label such as ‘review reports’ may conceal differences in source data, acceptable error, privacy requirements, or the decision supported. If the checker’s process does not let you express that context, preserve it in your notes; do not assume the signal accounts for it. The International Labour Organization’s ‘Workers’ exposure to AI: What indicators tell us – and what they don’t’ explains that exposure measures vary by method, rely on static task lists, and omit adoption constraints such as economic conditions and institutional barriers. It says these measures indicate technological susceptibility, not employment outcomes or displacement forecasts. Use a signal to direct verification, while treating local implementation and consequences as separate questions. An occupational catalog can help you notice an omitted duty, but it cannot settle the content of your promotion. O*NET’s 31.0 Task Statements dataset links task statements to occupations and distinguishes core from supplemental tasks; its documentation says each task is linked to exactly one occupation. This is not a description of your employer’s cross-functional role or its particular workflow and accountability. Compare it with the role outline and ask the role owner to confirm gaps. Record whether each description came from the outline, a manager’s confirmation, or your own assumption, so uncertain entries stay provisional. Read the crosswalk as a conversation aid. Bring added or changed tasks, high-consequence review duties, and unknown tool-use entries to the role owner. Ask which workflows are in use, what a human must verify, who has final authority, and what support or training accompanies the duties. If the role description changes, revise the inventory and revisit relevant signals. This method is not an audit verdict; it shows what a signal can help investigate and what still depends on confirmed local facts.
Sources: AI Proof Work AI Job Risk Checker; Workers’ exposure to AI: What indicators tell us – and what they don’t; O*NET 31.0 Task Statements
How should local adoption and labor demand be checked separately?
Ask what is actually in use today, what remains a proposal or pilot, who reviews outputs, who remains accountable, and what support accompanies the expanded role. These differ from whether a task appears exposed to AI. A checker signal can make a task worth investigating, but it cannot establish that this employer has selected a tool, put it into production, or changed staffing because of it. Ask what is available now, which tasks use it, what people must check, and who corrects errors. A worker experimenting with an approved tool is evidence of that worker’s use, not proof of organization-wide adoption. A pilot is evidence of a trial, not necessarily a settled workflow.
Broader evidence is a useful counterweight, provided its boundaries stay visible. In the 2023 OECD Employment Outlook chapter “Artificial intelligence and jobs: No signs of slowing labour demand (yet),” the OECD reviewed econometric studies using cross-country exposure differences and local labor-market variation,. It reported little evidence of statistically significant negative employment effects in the evidence then available. Net effects are theoretically ambiguous: displacement can be offset by productivity effects or new tasks, and it cautions that much evidence predates recent large language models and generative AI. It also identifies slow adoption and implementation time as possible reasons effects may take time to appear. Historical aggregate findings cannot settle the future of one role or show that current generative AI has no labor effect. A promotion can reorganize work inside a team even if broad employment measures show no detectable change. A local workflow can also change without proving that occupational demand is collapsing.
A newer short-horizon study strengthens that distinction. “Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI,” an NBER working paper linking Danish adoption surveys to administrative labor records, studies the period after ChatGPT’s launch using difference-in-differences estimates. Its abstract reports precise null effects on earnings and recorded hours at worker and workplace levels, ruling out effects larger than 2% two years after launch in the studied setting. The authors also report adoption initiatives, new AI-related tasks, and task reorganization. This shows measured earnings and hours can remain stable while work changes; it is not a forecast for another country, employer, or promotion. The study is a working paper, covers an early period, and measures specific labor outcomes rather than every dimension of job quality or future staffing. A null result cannot certify that a particular offer is sustainable, and task reorganization cannot be translated into a job-loss claim.
Check demand. Start with the internal role outline: recurring work, authority, success measures, and whether added review or coordination fits the stated hours. Then examine a modest sample of comparable postings where you could realistically work. Note recurring responsibilities and requested qualifications; advertisements show what employers seek, not completed hires, stable demand, or security. A national forecast or exposure label is not a local vacancy count. Ask for observable facts rather than speculate about unannounced plans: confirmed duties, tool status, review standards, decision rights, training time, staffing context if shareable, and how compensation and workload reflect added responsibility. If the role is still being designed, mark answers unknown and keep the comparison provisional. These facts can inform whether to accept, negotiate scope or support, or wait for clarity while respecting your salary floor, location, health, family, and training constraints.
Sources: Artificial intelligence and jobs: No signs of slowing labour demand (yet); Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI
What if the promotion outline is incomplete or the signals are mixed?
Keep the audit provisional when the promotion outline is incomplete or the signals conflict. A title can be settled before the recurring workflow, authority, or review standard is clear. Mark each task as confirmed, assumed, or unknown, and note whether the information came from a written outline, a manager’s explanation, or your estimate. This shows which unanswered question could change whether the role fits your goals and constraints.
The four transition groups can expose omissions. A task listed as retained may change when you move from preparing material to approving it; a new accountability may hide behind “own the process.” Ask the role owner to clarify recurring responsibilities, decision rights, expected review, and available time or training. O*NET’s task statements can prompt you to notice duties missing from an outline, but the O*NET Resource Center describes occupation-linked U.S. data, not this promotion’s duties or authority. Use it to prompt questions, not infer employer plans.
Sort uncertainty by what is known locally. If a task appears exposed but you have no evidence a tool is used, ask whether there is a current system, trial, or only a possibility. Keep the signal as a reason to monitor, not a forecast. If a tool is already used but checking is unclear, ask who verifies its output, what errors matter, and what standard governs approval. A bounded trial or review procedure may resolve that work-design question if the organization offers one. These are questions, not claims that a rollout or trial exists.
If responsibility exceeds the attached time or authority, the issue may be role design rather than a missing AI skill. Ask whether priorities, staffing, review time, and decision rights will change with the scope. Accountable review can be valuable when the reviewer has evidence, authority, and time; it can also become extra workload when responsibility expands without them. Consider that alongside salary, location, health, family responsibilities, and available training time.
There is a credible alternative to refusing a promotion because some tasks overlap with current AI capabilities: it may combine tool-assisted work with complementary duties or useful decision authority. The International Labour Organization’s exposure-indicator review says indicators describe technological susceptibility and omit adoption constraints; they cannot settle how one employer will redesign a role. Conversely, “reviewer” does not mean protected. The relevant evidence is the actual workflow, allocation, review burden, and accountability. If those facts are unavailable, the conclusion is incomplete information, not safety or danger.
A demonstrated skill gap calls for a different response from uncertain job design. If the clarified role requires a skill you lack, define a small learning experiment tied to that task, such as checking outputs against source material and asking what competent review requires. Do not commit to broad training or a career change until the role’s requirements and your constraints justify it. Update the comparison when duties are firm; the checker remains one task-level input, not the deciding vote.
Sources: Workers’ exposure to AI: What indicators tell us – and what they don’t; O*NET 31.0 Task Statements
What is the smallest useful next move?
Start with the proposed work, not a career verdict. Request the role outline and mark each recurring duty removed, retained, added, or changed. Note frequency, decision authority, consequences of error, review expectations, and uncertainties. Use the free [AI Proof Work checker](https://aiproofwork.com/ai-job-risk-checker) for task-level change-pressure signals on proposed duties, to the extent its process allows. Treat a signal as a question, not a personal job-loss probability or an automatic comparison of both roles. Label assumed duties and ask when they will be confirmed; missing information does not show whether a tool is in use or outputs meet local standards. Take the annotated list to the role owner. Ask what tools are used, who checks outputs, which decisions you will own, and what time or training support comes with the responsibility. The OECD chapter “Skill Needs and Policies in the Age of Artificial Intelligence” distinguishes basic AI literacy and digital and data skills from specialized preparation. Use that distinction to choose a learning step only if the audit reveals a workflow gap. Weigh it against salary, location, health, family responsibilities, and available time. If several paths need comparison, the [career roadmap](/career-roadmap) compares scenarios and constraints without guaranteeing employment or income. The checker can support the audit; clear duties, authority, review, and support determine how useful its signal is.
Sources: AI Proof Work AI Job Risk Checker; Skill Needs and Policies in the Age of Artificial Intelligence
Questions readers ask
Does an AI task checker tell me whether a promotion is safe?
No. It can provide task-level change-pressure signals, but those signals do not establish employer adoption, workflow performance, job redesign, labor demand or a person’s probability of displacement.
How do I compare my current job with a proposed promotion?
List recurring duties in both roles and mark each as removed, retained, added or changed. Note frequency, authority, consequences, review and accountability, and label assumptions or unknowns before using task signals.
Sources and notes
- AI Proof Work AI Job Risk Checker
The assignment brief describes task-level change-pressure signals and first actions, not a validated displacement probability or automatic promotion comparison.
- Workers’ exposure to AI: What indicators tell us – and what they don’t
The ILO brief explains that exposure indicators vary, rely on task information and do not establish adoption or displacement outcomes.
- O*NET 31.0 Task Statements
The O*NET data dictionary describes occupation-linked task statements, including core and supplemental classifications, rather than a particular employer’s promotion duties.
- O*NET Database
The O*NET database is an occupational reference that can prompt task questions but cannot establish a local role’s allocation or authority.
- Artificial intelligence and jobs: No signs of slowing labour demand (yet)
The OECD chapter reviews earlier aggregate employment evidence, describes net effects as ambiguous and notes limits including the period’s evidence base.
- Navigating the Jagged Technological Frontier
The HBS AI Institute summary describes a field experiment with 758 consultants and different performance patterns on tasks within and beyond the tested AI frontier.
- Measuring the Productivity Impact of Generative AI
The NBER Digest summarizes a customer-support field study reporting heterogeneous productivity effects by workers’ experience and skill.
- Artificial intelligence, job quality and inclusiveness
The accessible search-result summary reports task-range broadening and decision support in case studies; it does not establish outcomes for a particular promotion.
- Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI
The working-paper abstract reports precise null effects on earnings and recorded hours in its Danish study setting during the measured early period.
- Skill Needs and Policies in the Age of Artificial Intelligence
The OECD chapter distinguishes basic AI literacy and digital or data skills from specialized AI preparation.
Apply this to your own work
See the whole job market at once.
Explore which occupations AI may reshape, then turn the signal into a practical response.
Explore the job map