Methodology and sources

What this checker can tell you

The report combines published occupation-level indicators with task-level guidance. It helps you choose a small AI workflow to evaluate. It does not estimate the probability that you will lose your job, the number of hours you can save or a date when a job will disappear.

AI applicability, automation of a task and employment change are different concepts. A task can be suitable for AI assistance while the person remains responsible for its outcome.

Published AI applicability

We display ai_applicability_score from Microsoft’s Working with AI release 1.1, on its original 0–1 scale, rounded to three decimal places. It combines observed activity coverage, estimated task completion and scope of impact. A larger value means greater applicability under that study’s method, not greater probability of replacement.

The study analyzes anonymized U.S. Bing Copilot conversations from January–September 2024. Its activity classifications and completion assessments use model-based classifiers. The sample is consumer-assistant usage and does not represent all workers, employers, specialist AI software or present-day model capabilities.

The research uses O*NET 29.0 and detailed 2018 SOC codes. We retain the published score; we do not recalculate it with O*NET 31.0 tasks. Repository snapshot: 550c00c9d1b0

Observed activity coverage

The report uses coverage_user from Microsoft’s soc_metrics.csv. This is the importance- and relevance-weighted fraction of an occupation’s intermediate work activities with a user-goal conversation share above 0.05% in the study. We display it as a percentage, rounded to one decimal place.

It is not the percentage of people using AI, the percentage of a workday automated or a measurement of usage for each task in our checklist. We do not assign the occupation’s coverage to individual tasks.

Anthropic and ILO indicators are not included in this release. Their samples and definitions differ from Microsoft’s and should not be averaged into a single score without a separate method.

Occupation matching and confidence

Each report explicitly records an O*NET 31.0 occupation and a research SOC code. Both codes must exist in their source data for the import to succeed. We match titles and occupational scope; we do not automatically infer a match from a similar title.

• Direct occupation match: the job title refers to the named standard occupation. Specialties and individual workplaces can still differ.
• Broader occupation group: tasks come from an O*NET specialty, while the published research and BLS data cover its wider SOC group. For example, Business Intelligence Analysts is within Data Scientists.
• Approximate occupation match: the standard occupation only covers part of the requested role. UX Designer uses Web and Digital Interface Designers; Data Engineer uses Database Architects; Product Manager uses Marketing Managers. Each report states the limitation.

These labels describe mapping quality, not statistical confidence. We do not publish a numeric confidence estimate or an invented confidence interval. A direct code match does not establish that the study predicts your own work.

Tasks, categories and skills

Each occupation has eight editorially selected tasks linked to O*NET 31.0 task IDs. We shorten task labels and may split one broad O*NET task into several practical activities. The original statement is available under “Task source.” The selection is not an exhaustive job description or a ranking of all duties.

O*NET task importance is shown on its 1–5 scale when a non-suppressed rating is available. The skill list uses the five highest available importance ratings across essential and transferable skills. Importance describes the occupation; it does not measure AI suitability.

• Partial automation: a bounded draft or routine step may be delegated after the input and acceptance criteria are defined. A person still reviews the result.
• AI assistance: AI may propose options or carry out intermediate steps; a person checks evidence and integrates the result.
• Human-led: the task centers on responsibility, negotiation, contextual decisions, authorization or direct validation. AI can help prepare material.

These categories, explanations, learning exercises and tool matches are AI IDE List editorial guidance. They are not classifications published or endorsed by O*NET, Microsoft or BLS, and are not empirically validated estimates of automation. No live language model generates your score.

How personalization works

“Not part of my job” has weight 0; “Occasionally,” “Regularly” and “Often” have weights 1, 2 and 3. Every task starts at 2. A category’s share is its selected task weights divided by the sum of all selected task weights. The displayed percentages are rounded independently.

This is a frequency-weighted task mix, not working hours or time saved. Selecting no tasks produces an empty plan. Your selections never change the published occupation-level indicators or BLS projection.

The practice list prioritizes tasks you do more often, keeps different skill areas and shows up to three. At equal frequencies, assistance and partial-automation tasks come before human-led tasks. Tools match the capabilities of selected assistance tasks; a match is a suggestion to evaluate, not evidence that a tool is best for the occupation.

Employment projections and geography

BLS Table 1.2 supplies the published percentage change in U.S. employment between 2025 and 2035. We join it by the explicit SOC mapping and name the BLS group in each report. These projections combine many economic and demographic factors and are not a measure of AI’s causal effect.

This release does not provide UK forecasts or an “AI changing UK STEM jobs” score. A UK report would require ONS SOC 2020 mappings and UK evidence. U.S. indicators should not be presented as local forecasts for another country.

Tool recommendations and privacy

Tool links come from published, active English entries in the existing AI IDE List directory. Model links come from the site’s model catalog. Unpublished or discontinued tools are excluded. Recommendations use task capabilities and frequency; they do not use paid placement or the research score as a ranking.

Your task choices are stored in this browser. You can reset them from the report. A copied report link contains an occupation and frequency codes, so the recipient can reproduce the selection. We do not request a name, CV, job description or account.

When analytics consent allows it, the site records actions such as opening a known occupation, adjusting tasks, copying a report and opening a tool. These events do not include your search text or individual task frequencies.

Attribution and licenses

• O*NET 31.0

  National Center for O*NET Development; U.S. Department of Labor, Employment and Training Administration. Task labels are adapted; task guidance is by AI IDE List.

  CC BY 4.0

• Working with AI, v1.1

  Kiran Tomlinson, Sonia Jaffe, Will Wang, Scott Counts and Siddharth Suri. Working with AI: Measuring the Applicability of Generative AI to Occupations. NeurIPS 2026 (in press). Published scores are retained without recalculation.

  CC BY 4.0

• BLS Employment Projections, 2025–2035

  U.S. Bureau of Labor Statistics, Table 1.2. Selected occupation growth percentages transcribed from the official table on 2026-10-25.

  U.S. government data

Use of these sources does not imply endorsement of AI IDE List. O*NET is a trademark of the U.S. Department of Labor, Employment and Training Administration.

Data version and updates

English, Simplified Chinese and Japanese versions use the same data. Translations do not change occupation codes, task IDs, research indicators or task-selection weights. O*NET task statements and official occupation names remain in their original English for reference.

October 25, 2026 — Version 1.0.0. Initial English release: 20 occupation reports, 160 task mappings, O*NET 31.0, Microsoft v1.1 and BLS 2025–2035 projections.

Source snapshots and explicit mappings are versioned with the application. Updating a source requires rechecking occupation codes, task IDs, suppression flags, the projection period and editorial guidance. New source versions are not silently substituted into existing reports.

Found an incorrect mapping or statement? Include the occupation, task and source. Send a correction