Can a Tax Preparer Map AI Pressure by Task?

Direct answer: use the checker as a prompt for three rows

Yes. A U.S. tax preparer can use the AI Proof Work checker as a prompt to map research, drafting, and client review separately, then consider its task-level change-pressure signals alongside the work. The product description supports reflection; it does not establish that the interface accepts these tax-specific subtasks or returns three scores. Keep the distinctions in your own notes unless the current inputs clearly support them. The work loops have different inputs and responsibilities. O*NET’s Tax Preparers profile lists consulting tax handbooks about atypical returns, preparing returns, interviewing clients, reviewing records, and checking entries as distinct tasks. This supports separate rows; it does not show how much time one preparer spends on each task or how exposed that person is to AI. A useful map records the recurring task, its output, what must be checked, and who resolves exceptions. Use the map to choose one bounded workflow to examine, subject to workplace approval and data-handling obligations. A checker signal is a prompt for inquiry, not evidence of employer adoption, role redesign, or a prediction of job loss. Start by mapping the work rather than treating one occupation-level result as a verdict.

Sources: 13-2082.00 - Tax Preparers

Why does one occupation label hide three different work loops?

A single job title can hide work with different inputs and different tests for “done.” O*NET’s 2026 Tax Preparers profile lists consulting tax law handbooks for atypical returns, preparing returns, interviewing clients, reviewing financial records, explaining tax law, and checking data or totals. Those duties support separating the work for reflection; they do not show how much time one preparer spends on each duty or how exposed that person is to AI. The three work loops below are editorial groupings, not official O*NET categories. The first row is research: a tax question is the input, and an applicable rule or source trail is the output. A preparer may need to locate authority, determine which tax year and jurisdiction it addresses, and decide whether the facts fit its scope. Finding a passage is only one step. The endpoint is an interpretation tied to facts and a source another person can check. This row depends on source quality, currency, and fit, not merely on whether a summary reads clearly. The second row is drafting: records, established rules, and a defined question become a first version of a return entry, explanation, or request for missing information. A completed-looking draft does not settle whether an input was accurate or a treatment applies. This row depends on record quality and keeping the reasoning attached to the draft. Note which output recurs in your own work rather than assuming every preparer writes the same materials. The third row is client-specific review: client answers and source documents are reconciled with the proposed treatment, calculations, and current rules. The output is a return or explanation ready for the next step, or an identified exception that must be resolved. Unlike proofreading alone, this loop depends on noticing missing or conflicting facts and deciding who must follow up. O*NET separately lists interviewing clients, reviewing records, and checking work; it does not prescribe one review procedure for every practice. Imagine an atypical deduction question. Research would identify and check the relevant authority; drafting might organize the proposed treatment or prepare a client question; review would compare the answer and supporting records with the draft and flag a gap. This illustrates how to divide a work loop, not a particular return, AI system, or firm practice. One issue may pass through all three rows, but each handoff asks a different question: Is the authority applicable? Does the draft reflect the supported treatment? Do the client facts and records support proceeding? An occupation-level list can guide a personal map but cannot substitute for one. O*NET describes tasks across an occupation; it does not calculate your task shares or show where your workflow begins and ends. Make three rows using recurring outputs you handle, and use existing time records if available. Do not infer percentages from the profile. For each row, record its input, output, what must be checked, and what happens when facts or sources do not settle the question. Those distinctions let a task-level checker prompt reflection without pretending one broad job label measures three separate workflows.

Sources: 13-2082.00 - Tax Preparers; Data Updates: Tax Preparers (13-2082.00)

What could assistance change in research, drafting, and review?

The three activities may all happen on a screen, but they close different questions. Research asks which rule or authority may apply. Drafting turns a checked conclusion into a return entry, explanation, or request for information. Review asks whether the conclusion and its supporting facts hold together for this client. O*NET’s “13-2082.00 - Tax Preparers” lists consulting tax handbooks for atypical returns, preparing returns, explaining rules, reviewing records, and checking entries as distinct duties. The list supports separating the work, not claims about an individual’s task shares or AI use. In research, a preparer might use an approved tool to locate candidate guidance or organize a question before consulting the underlying authority. The preparer still needs to establish that the source is authoritative for the question, covers the relevant tax year and jurisdiction, and applies to the facts at hand. The practical endpoint is therefore not “an answer appeared,” but a traceable source and a reasoned determination that the source fits. The IRS Office of Professional Responsibility’s “Introductory Guidelines for Responsible AI Use in Federal Tax Practice” says practitioners should verify AI-generated facts, citations, calculations, and written advice. It states a verification expectation, not a finding about any tool’s reliability. Drafting starts later in the chain. Where the treatment has already been determined, assistance might help arrange the explanation, prepare a draft request for missing records, or turn verified notes into clearer language. These are possible uses, not claims about adoption or guaranteed capability. The draft must preserve the preparer’s actual conclusion and identify what remains unknown. Generating confident wording does not decide whether a deduction, credit, or other treatment is supported. A useful draft should make its inputs and unresolved questions visible so a preparer can correct it before it travels to a client or becomes part of return work. If the task is to decide the treatment itself, drafting assistance cannot stand in for the research and professional judgment that decision requires. Review has a different finish line: reconcile the proposed work with source documents and client statements, notice contradictions or gaps, route exceptions, and leave an accountable person responsible for resolution. The IRS’s “Due diligence requirements for tax preparers” describes further inquiry for specified matters when information appears incorrect, inconsistent, or incomplete. That guidance applies to covered due-diligence obligations, not an identical checklist for every return, but it illustrates why a fluent draft does not close a factual gap. Review may also check calculations and whether rules support the explanation. Language edits alone cannot establish those things. For each row in a task map, record the input, intended output, evidence needed to accept it, likely exception, and person who owns that exception. Then include the time spent checking and correcting. A draft that takes less time to produce may still take more total effort if its sources need rebuilding or its errors are costly to resolve. This is a decision rule inferred from the distinct duties and IRS verification guidance, not a measured productivity result. Map what closes the task and who signs off; do not use screen-based or repeatable as a substitute for that analysis.

Sources: 13-2082.00 - Tax Preparers; Due diligence requirements for tax preparers; Introductory Guidelines for Responsible AI Use in Federal Tax Practice

Why is client review a separate accountability task?

Client review is a separate accountability task because it asks whether the return is supportable from the client’s facts and rules, not merely whether the draft reads clearly. A polished explanation can rest on a missing document, a contradiction between records, or an unchecked assumption. Editing improves expression; review must also decide what evidence is sufficient, which questions remain open, and whether an exception has been resolved. The IRS page “Due diligence requirements for tax preparers” makes this distinction concrete for specified matters. It says paid preparers must make additional inquiries when a well-informed preparer knowledgeable in the law would find information incorrect, inconsistent, or incomplete. It describes asking reasonable questions, evaluating whether information makes sense, following up on gaps, and using professional judgment about the credibility of the client and answers. If the preparer is not comfortable with the answers or credibility, the guidance says not to prepare the return until satisfactory information resolves the concern. These requirements concern covered matters, including named credits and filing status in applicable tax years; they are not an identical checklist for every return. The task-mapping lesson is narrower: some review is exception handling that depends on facts and follow-up, not proofreading alone. Consider an illustrative file where a client’s explanation of income does not appear to fit the records supplied. A drafting aid might help turn a preparer’s question into a clear request for clarification. That wording could make communication easier, but it cannot establish that records are complete, decide whether an answer is credible, or close the gap. The preparer still needs to determine what information is relevant and decide what to do if the inconsistency remains. This example is not a claim about a particular tool’s performance; it shows why the output and accountable decision belong in different parts of a workflow map. The IRS Office of Professional Responsibility’s “Introductory Guidelines for Responsible AI Use in Federal Tax Practice,” dated June 24, 2026, reinforces the boundary for AI-created material. It says practitioners must thoroughly review AI-created documents and language before incorporating them into writing delivered to clients or submitted to the IRS, and verify AI-produced facts, citations, and calculations. For written federal tax advice, it describes AI output as a starting point subject to thorough review and calls for independent authentication of factual and legal information. Review therefore extends beyond tone or grammar: the reviewer must trace claims to support and assess whether facts and authority fit the matter. The guidance is introductory federal-practice material, not a vendor test or a complete account of every professional or firm rule. For a three-row map, keep the handoff visible even if one screen combines drafting and review: what claim or fact is checked, what source or client evidence closes it, what exception routes back for follow-up, and who remains accountable for the decision. This makes review work observable instead of hiding it inside a general “prepare return” label. A plausible draft may still leave review work untouched.

Sources: Due diligence requirements for tax preparers; Introductory Guidelines for Responsible AI Use in Federal Tax Practice

What changes when assistance moves from private notes to client-facing work?

The boundary changes in two ways: material may now carry taxpayer-specific information, and the words may become advice for which a practitioner remains responsible. A private research note can still be wrong or contain sensitive details, so it is not risk-free. But once language is used in a client explanation, written advice, or a filing, clear prose is not enough. Ask whether facts and authority fit this client's circumstances, whether confidential information was handled in an approved way, and who must resolve uncertainty.

The IRS Office of Professional Responsibility’s June 2026 “Introductory Guidelines for Responsible AI Use in Federal Tax Practice” says practitioners must thoroughly review AI-created documents and language before delivery to a client or submission to the IRS, including verifying facts, citations, and calculations. For written federal tax advice, it describes AI output as a starting point and calls for independent authentication of factual and legal information. That is a practical distinction between editing a sentence and substantiating its claims: a polished explanation does not itself show that the cited authority is current, that the calculation uses the right inputs, or that the underlying facts are complete. The guidance discusses federal tax practice and Circular 230 duties; it is not a certification of any AI product or a complete account of every state rule, professional obligation, or firm procedure.

The data question runs alongside accuracy. The same guidance warns that uploading taxpayer information to unsecured or public systems can create unauthorized disclosure risks and says practitioners should use secure, enterprise-approved AI with safeguards. This does not establish that a particular vendor or account meets a preparer’s obligations. Before testing a workflow, check employer policy and applicable requirements: which tool is approved, what information may be entered, whether prompts or outputs are retained, and what review and escalation process applies. Do not put identifiable taxpayer information into a test unless its use has been explicitly cleared. Fluency, a checker result, or the label “private” cannot establish permission.

The IRS “Policy for Artificial Intelligence (AI) Governance” in the Internal Revenue Manual applies to IRS employees, contractors, and vendors working with IRS information or systems; it is not a rulebook for private practices. Its scope underscores why preparers must confirm the controls that actually apply to their own work. In a task map, record confidentiality controls separately from quality checks. A workflow is a realistic trial candidate only when the tool is approved, permitted data are clear, a qualified reviewer owns the output, and unresolved issues have a defined route. If those are unknown, ask the appropriate supervisor or compliance contact before testing with client data.

Sources: Introductory Guidelines for Responsible AI Use in Federal Tax Practice; Policy for Artificial Intelligence (AI) Governance

What can the checker tell you—and what remains a workplace question?

The AI Proof Work checker can give a tax preparer task-level change-pressure signals to use as prompts for reflection, according to the assigned product description. It organizes questions; it is not a validated probability that a preparer will lose a job. The current interface has not been verified for tax-specific inputs or for separate scores for research, drafting, and client review. Keep those three rows in your own task map unless the checker itself clearly supports that level of detail. Do not infer the feature from the fact that the work can be described in separate rows.

The distinctions matter because each signal answers a different question. Technical capability asks what a system can do under particular conditions. Exposure asks whether a task could be affected by that capability. Observed use asks whether workers actually use it; employer adoption asks whether an organization has approved and put it into a workflow. Job redesign concerns changes to duties, review, or handoffs. Labor demand concerns openings and employment patterns. Displacement concerns jobs lost or workers moved, and cannot be read directly from exposure. A checker result cannot establish that a tax firm has adopted a tool, changed procedures, reduced review time, or altered staffing. Those are workplace facts to investigate, not implications to fill in from a score.

The International Labour Organization’s research brief, Generative AI and Jobs: A 2025 Update, describes an index built from task-level data, expert input, and AI predictions, covering nearly 30,000 tasks and grouping occupations into four exposure gradients. It estimates that one in four workers globally is in an occupation with some degree of potential exposure. The brief says most jobs are more likely to be transformed than made redundant, in part because human input remains necessary. This is modeled occupational exposure, not a survey of tool use or a record of employers implementing systems. It gives no tax-preparer result, U.S. adoption rate, or individual job-loss chance. It is useful as context for why task change deserves attention, not as a number to transfer onto one worker.

For a preparer, the next questions are observable and local: Has the firm approved a pilot? Have written procedures changed? Which output is being assisted, and who checks it? Has the amount of correction or review work changed across repeated cases? Are exceptions routed differently? Has staffing or the mix of assigned duties actually changed? Documented answers can show whether a workflow is being tested or redesigned, not prove displacement. Distinguish what has happened from what someone expects might happen.

Read a high signal as a reason to identify the exposed task and ask what evidence would show useful, acceptable assistance. Read a low signal as a reason to check whether the task description missed important parts of the work. In either case, use the result to sharpen the task map, then compare it with approved procedures and actual workflow evidence. The checker can help frame that reflection; it cannot report your employer’s decisions or settle your career choice.

Sources: AI Proof Work task checker; Generative AI and Jobs: A 2025 Update; One in four jobs at risk of being transformed by GenAI, new ILO–NASK global index shows

What does current labor demand add—and not add—to the decision?

Current U.S. occupational projections add a useful but limited piece of context: O*NET’s Tax Preparers profile reports employment growth of 5% to 6% over 2024–2034 and 10,400 projected openings. The figures come from Bureau of Labor Statistics projections surfaced in O*NET, not a study of AI in tax firms. They describe projected national demand; they do not show that AI caused growth, that a particular city has openings, or that an individual preparer’s role is secure. The profile says projected openings include both employment growth and replacement. So the total is not a count of newly created positions alone or a promise that a particular worker can access one. The 5% to 6% figure describes estimated change in total employment across the projection period. It is a forecast, not a record of what has already happened. Dates, geography, and occupation definition matter when interpreting it. This demand measure answers a different question from a task map or change-pressure signal. A preparer may have research, drafting, and client-review duties with different assistance possibilities and verification needs. An occupational projection groups workers by classification; it does not show how duties divide inside one office, whether an employer is adopting a tool, or whether work is being redesigned. A task’s potential exposure also does not tell us whether occupational demand is rising or falling. These signals can coexist because they describe different things. The projection is neither a reason to dismiss workflow change nor to pivot immediately. It adds national employment context but leaves open what is happening in a particular workplace and local market. O*NET links the figures to BLS 2025 wage data and 2024–2034 employment projections. The national outlook cannot establish local pay, seasonal hours, access requirements, or the fit of an adjacent role. If considering a move toward research, planning, or advisory work, use the projection as background and check local openings directly. Review actual duties, seasonality, credential requirements, and location, then compare them with your experience, income floor, available learning time, and other constraints. An opening count cannot tell you whether a path fits your household or interests. Demand data belongs in a career comparison alongside your task map and workplace evidence. O*NET’s outlook is a national demand signal, not an AI forecast or personal guarantee. Investigate local conditions before treating an adjacent move as available.

Sources: 13-2082.00 - Tax Preparers; Data Updates: Tax Preparers (13-2082.00)

How can a preparer test one bounded workflow without overstating the result?

A useful test starts with one recurring output, not a broad instruction to “try AI on tax work.” Choose, for example, a research note on a non-client-specific question or a draft explanation of a rule whose authority is already known. Define success first: the output organizes the question and points to relevant primary authority, while a preparer independently confirms the current rule and each citation. The goal is not to judge whether the answer sounds polished, but whether assistance changes the whole work loop after checking and correction. Confirm that the employer permits both the tool and task. Set the data boundary before entering anything. Use only material policy explicitly allows; synthetic or de-identified examples are options only if the employer clears them for this tool and workflow. Removing a name does not itself make taxpayer information safe to upload. The IRS Office of Professional Responsibility’s “Introductory Guidelines for Responsible AI Use in Federal Tax Practice” calls for secure, enterprise-approved systems and warns about unauthorized disclosure through unsecured or public services. The test therefore begins with permission and allowed inputs. Preserve a source trail a reviewer can check: record the question, permitted input, output, primary authority consulted, and any correction. Independently verify facts, citations, and calculations before relying on generated material. The IRS guidance says AI-created documents and language incorporated into client-facing writing require thorough review; generated advice is a starting point, and factual and legal information must be independently authenticated. A fluent draft does not show that its source is current or applies to the question. Measure the full loop in ordinary work terms. Note time spent preparing the input, checking sources, correcting errors, resolving exceptions, and obtaining review, as well as whether the result met normal standards. Record what required human judgment and who owned unresolved questions. This is an editorially derived comparison, not an IRS-prescribed test or validated productivity measure. A shorter first draft may still add work if its authorities are missing, assumptions unclear, or a reviewer must rebuild it. Then stop at the scale the evidence supports. One permitted trial can show whether that output merits refinement, offers no practical benefit, or should not be used in that workflow. It cannot establish a firm-wide productivity gain, market adoption, job redesign, or employment effect. Broader process changes need repeatability across suitable tasks, documented policy, and review by responsible people. If permission or a safe data route is unclear, document the constraint and ask the appropriate manager or compliance contact what may be tested. This observation supplies local evidence an occupation-level checker cannot, while leaving larger career decisions to evidence about actual work and personal constraints.

Sources: Introductory Guidelines for Responsible AI Use in Federal Tax Practice; Due diligence requirements for tax preparers

Which next move fits the evidence and the preparer's constraints?

For a preparer who wants to stay in tax work, start by mapping research, drafting, and client review separately. If the workplace permits it, test one bounded output. A task map shows where a change-pressure signal may matter; it does not establish a need to retrain. O*NET’s Tax Preparers profile describes varied duties, but cannot tell an individual how an employer organizes work or which tasks dominate an individual’s week. Use recurring outputs and actual records rather than invented time shares. An upgrade within the current role is a practical experiment: choose one research or drafting task, confirm the tool and data are approved, and compare the assisted result with the usual process, including verification and corrections. If work is changing, adjacent research, planning, or advisory responsibilities may be worth exploring. Fit depends on experience, credentials, local openings, income needs, and time to learn; none is automatically available or better paid. A larger career change calls for a fuller comparison. Before committing to a course, certificate, degree, or new field, check prerequisites, total cost and training time, location, health and family constraints, and the salary floor you need. A checker result cannot settle those tradeoffs. If your employer has automated substantial tasks or is actively restructuring, consider timing and income risk. Write the three task rows and choose one permitted output to observe. The free AI Proof Work checker adds task-level change-pressure signals for reflection. The paid career roadmap can compare constrained paths and a 30/60/90-day plan. Neither predicts displacement or guarantees employment or income. The recommendation could change with workplace redesign and a path that fits your constraints.

Sources: 13-2082.00 - Tax Preparers; Introductory Guidelines for Responsible AI Use in Federal Tax Practice