Evidence library
BLS Occupational Outlook Handbook: Middle-school teachers
Instruction includes lesson preparation, assessment, classroom teaching, parent communication and supervision.
Published: Date unavailable · Source access checked: 2026-10-29T22:26:34.495908+00:00
- US occupational description; not measured AI task performance.
- Employment projections are forecasts, not observed AI displacement.
BLS Occupational Outlook Handbook: Accountants and auditors
Accountants examine records, evaluate financial risks, prepare taxes and communicate findings. BLS expects routine automation to change the task mix.
Published: Date unavailable · Source access checked: 2026-10-29T22:26:34.494407+00:00
- US occupational description; not measured AI task performance.
- Employment projections are forecasts, not observed AI displacement.
BLS Occupational Outlook Handbook: Heavy and tractor-trailer truck drivers
Drivers transport goods, inspect equipment and cargo, maintain records and report incidents; duties extend beyond driving.
Published: Date unavailable · Source access checked: 2026-10-29T22:32:12.413704+00:00
- US occupational description; not measured AI task performance.
- Employment projections are forecasts, not observed AI displacement.
ChatGPT in lesson preparation: Teacher Choices Trial
In the 2024 trial, 259 teachers in 68 English schools participated. For scoped science planning, reported weekly preparation was 56.2 minutes with ChatGPT plus guidance versus 81.5 without GenAI.
Published: 2024-12-12 · Source access checked: 2026-10-29T22:32:48.242601+00:00
- England science teachers; not representative of US teachers.
- Scoped preparation activity, not the whole teaching workload.
ChatGPT in Lesson Preparation: Evaluation Report
Randomization was at school level. Weekly diaries measured the outcome after a five-week learning period; 18.5% were excluded from the primary analysis for missing diaries. Sampled blinded resource review found no detectable quality difference.
Published: 2024-12 · Source access checked: 2026-10-29T22:27:48.809687+00:00
- Self-reported workload; missingness and England-to-US transfer limit inference.
- No detected quality difference does not prove equivalence.
Human + AI in Accounting: Early Evidence from the Field
The study combines 277 accountant surveys, proprietary records for 79 SMEs and a framed account-classification experiment. Field adoption associations coexist with experimental evidence of improved average classification accuracy and errors from following non-consensus advice.
Published: 2026-04-16 · Source access checked: 2026-10-29T22:31:01.095771+00:00
- Field associations do not identify a causal occupation-wide effect.
- Single partner platform and bounded account classification; transfer to tax/audit responsibility is unestablished.
- Raw proprietary data not available in this package.
Aurora driverless launch questions
Aurora reports second-generation trucks operating with an empty driver seat, sometimes carrying a roadside-assistance person, and a commercial network of ten driverless routes.
Published: 2026-07-22 · Source access checked: 2026-10-29T22:32:48.243439+00:00
- Company statement, not independent safety or operational verification.
- Bounded operator network; no occupation-wide displacement rate or nationwide capability inferred.
- Driving evidence does not establish cargo inspection, coupling or other duties.
BLS Occupational Outlook Handbook: graphic designers
Graphic designers create visual concepts and layouts, confer with clients, and check final work.
Published: Date unavailable · Source access checked: 2026-10-29T22:49:42.181102+00:00
- US occupational description; no AI performance finding.
BLS Occupational Outlook Handbook: operations managers
General and operations managers oversee daily work, assign duties and direct use of materials and human resources.
Published: Date unavailable · Source access checked: 2026-10-29T22:49:42.181102+00:00
- US occupational description; no AI performance finding.
The Double-Edged Roles of Generative AI in the Creative Process: Experiments on Design Work
In a randomized field experiment with 120 employed graphic/visual designers, AI-assisted ideation improved expert-rated creativity while taking longer. Implementation-only AI use added time without significant creativity gains.
Published: 2025-10-03 · Source access checked: 2026-10-29T22:49:42.181102+00:00
- One advertisement competition; task-specific creativity and time outcomes, not occupation-wide productivity.
- Study 2 used Midjourney V6.1. Student expertise groups in Study 1 are separate from the employed professionals in Study 2.
- No significant creativity improvement is not proof of equal quality. No employment-displacement inference.
The Cybernetic Teammate: A Field Experiment on Generative AI Reshaping Teamwork and Expertise
The April 2025 NBER working paper analyzes 776 randomized P&G professionals in a one-day product-development workshop. Individual AI users produced higher-rated proposals and spent 16.4% less task time than individuals without AI.
Published: 2025-04 · Source access checked: 2026-10-29T22:49:42.181102+00:00
- Commercial and R&D participants, not a representative general-operations-manager sample.
- One-day workshop and one-hour AI training; limited evidence on extended coordination, repeated revisions or actual operational execution.
- Use the April 2025 paper's 776 randomized participants; do not pool with HBS reporting of 791.
- Task working time excludes any inference of net organizational savings or staffing effects.
BLS: Registered nurses
Nursing includes documentation, monitoring, treatment, care coordination and education.
Published: Date unavailable · Source access checked: 2026-10-29T22:47:48.094260+00:00
- Official US description; no AI performance measurement.
BLS: Information clerks, hotel desk subtype
Hotel desk clerks register guests, assign rooms, process payments and maintain reservation records.
Published: Date unavailable · Source access checked: 2026-10-29T22:47:48.094260+00:00
- Use the named hotel desk subtype; the wider information-clerk category is not an occupation-specific AI estimate.
Generative AI-Based Nursing Diagnosis and Documentation Recommendation Using Virtual Patient Electronic Nursing Record Data
A fixed-order pilot of 40 South Korean nurses compared documentation in a virtual-patient training system, then with SmartENR AI. The authors report shorter documentation time, but internal numerical inconsistencies prevent a reliable normalized savings estimate.
Published: 2025-04 · Source access checked: 2026-10-29T22:46:44.636744+00:00
- Fixed order and repeated familiar scenarios create learning/order confounding; no randomized clinical workflow comparison.
- Virtual patients and a training ENR, not a production hospital EHR or patient-safety evaluation.
- Abstract and Table 1/Results report different time means; approximately 40% and 38.6% statements are inconsistent. No percentage saving is used.
- Five authors were current or former DKMediInfo employees, the SmartENR provider.
- CC BY-NC source; private raw XML is not redistributed. Only citation, brief excerpt and original summary are exported.
Outpatient reception via collaboration between nurses and a large language model: a randomized controlled trial
The published trial abstract reports higher patient satisfaction for nurse–LLM collaboration than nurse-only reception in a single-center study of 2,164 participants.
Published: 2024-07-15 · Source access checked: 2026-10-29T22:47:03.422779+00:00
- This is the author abstract retained through PubMed, not a full-text methods audit.
- Patient satisfaction in outpatient reception is distinct from nursing documentation accuracy, bedside care, staffing or job displacement.
- No whole-job performance or independent clinical safety conclusion follows.
The influence of human-robot collaboration on perceived job duality and workplace well-being: Hotel employees’ perspectives
A scenario experiment with 384 current hotel receptionists studied robot functionality, appearance and entertainment. It measured perceived social presence, job insecurity and wellbeing, rather than completed hotel transactions.
Published: 2025-12-20 · Source access checked: 2026-10-29T22:46:03.554345+00:00
- Online Prolific screening selected current receptionists; purposive platform recruitment is not a representative workforce sample.
- Participants evaluated hypothetical robot scenarios, not observed deployed robot or GenAI performance.
- Perceived job insecurity is a psychological response, not observed job loss or replacement probability.
- Full text was accessed for methods, but source text/images are not included in public downloads.
Learning AI by Doing: A Pilot Study of Human–AI Collaboration and Employee Stress in Guest Services
A front-desk GenAI simulation poster describes an initial four-student pilot, followed by seven students refining simulation logistics. It provides task-design observations without a demonstrated workforce productivity effect.
Published: Date unavailable · Source access checked: 2026-10-29T22:46:24.757286+00:00
- Conference poster preparation; no completed representative employee effectiveness trial.
- Student convenience pilots and scenario realism checks do not establish hotel employee productivity or stress reduction.
- Upload-path year is not treated as a verified publication date.