Human input
No human ratings have been submitted for this paper yet, so there is no human aggregate or synthesis score. If you know the work, rate or discuss it on the dashboard card; the team reviews feedback before it affects prioritization.
Loading public input and its source labels…
AI assessment: why consider this paper?
Using a nationally representative survey, the paper builds occupation- and task-level indexes of generative-AI use and concludes that adoption is widespread but shallow. It informs forecasts of labour-market effects and how quickly AI diffuses inside jobs. An evaluation would focus on survey measurement of AI use, the task taxonomy, representativeness, and how adoption translates into productivity or employment effects.
AI-generated criterion ratings and reasoning
Scores come from AI prioritization; the accompanying explanations include AI-assisted source checks. These are provisional judgments, separate from human ratings and commissioned evaluations.
Decision relevance
6.0/10
The expected contribution is moderately positive: better adoption measurement could improve statistical collection, downstream labor-market research and eventually the targeting of worker support. The principal beneficiaries would initially be US workers and institutions; poorer workers' welfare interests deserve particular attention, but neither their realized gains nor LMIC transfer is established. Incremental value comes from supplementing existing exposure and platform measures, rather than capturing the total value of managing AI's economic transition. The specific measurement gap is tractable, while policy implementation and welfare effects are uncertain. There is no established direct animal-welfare or catastrophic-risk benefit. Documented communication and accessible indexes support realization, but changing technology and ambiguous links from adoption to productivity or displacement constrain the impact score.
- Paper claim to check Figure 6 reports that over 80% of occupations and 40% of tasks have adoption above 20%, while only 2.8% of tasks exceed 50%. These are weighted distributions from pooled August 2025–May 2026 surveys, excluding cells with fewer than 20 observations. [Working paper](https://s3.amazonaws.com/real.stlouisfed.org/wp/2026/2026-017.pdf) Source abstract
Value of added scrutiny
8.0/10
The authors' Federal Reserve Bank of St. Louis published a public summary, and a CEPR VoxEU column on measuring what work generative AI does appears alongside it. These show policy-audience reach. We found no independent critique, and the St. Louis summary is from the authors' own institution.
Timing
9.0/10
Feedback remains actionable. The [Federal Reserve working-paper version](https://s3.amazonaws.com/real.stlouisfed.org/wp/2026/2026-017.pdf) identifies an initial April 27, 2026 version and an August 25 revision; the NBER issue is dated August 2026. Thus the current working-paper version is recent even though development began earlier. The paper acknowledges conference and seminar feedback, which provides some expert input but does not establish a public claim-by-claim assessment. The supplied targeted search found no distinct substantial-review leads, and its citation-search component failed. That uncertainty does not justify declaring the paper unreviewed in every sense. Prompt evaluation is valuable before measurement choices become embedded in subsequent work, with the current version fixed explicitly at commissioning.
- Source record Publication-stage and date evidence should be checked in the linked paper record. NBER
Methodological potential
7.0/10
The central challenge is external validity: demographic weighting cannot by itself eliminate selection into an online panel or reporting error. Reviewers should examine whether restricting questions to occupation-linked tasks omits AI-assisted activities systematically, and distinguish workers, worker-task pairs, conversations, prompts and time spent. Cross-platform differences may reflect genuine user and product specialization as well as classification error; disagreement alone does not establish which source is closer to the relevant target. Within-worker dependence and sparse cells matter for uncertainty and rankings, while residual adoption variation does not identify a causal effect of experience or training. The public data page lists adoption indexes but does not establish availability of this paper's complete current microdata and code; a package for a related paper is insufficient. Commissioning should therefore begin by confirming access and fixing the version. A focused survey-and-measurement evaluation, with clearly stated limits on platform validation, is more feasible than attempting to validate employment or welfare effects the paper does not directly identify. The paper's funders and fuller author list, including Tyler R. Schumacher, are recorded in the [working paper](https://s3.amazonaws.com/real.stlouisfed.org/wp/2026/2026-017.pdf).
- Paper claim to check Figure 6 reports that over 80% of occupations and 40% of tasks have adoption above 20%, while only 2.8% of tasks exceed 50%. These are weighted distributions from pooled August 2025–May 2026 surveys, excluding cells with fewer than 20 observations. [Working paper](https://s3.amazonaws.com/real.stlouisfed.org/wp/2026/2026-017.pdf) Source abstract
Prominence
9.0/10
The scoring model estimated prominence from the paper's venue, authors, institutional setting, and visibility. The model did not supply a criterion-specific explanation. Current public-attention status: Policy-audience dissemination; no independent critique found.
- Author-institution summary The Federal Reserve Bank of St. Louis published a summary of the findings. Its research department issued the working paper, so this is not independent. St. Louis Fed
Likely influence
8.0/10
The scoring model estimated how far the findings could shape later research or decisions, without supplying a criterion-specific explanation. Current public-attention status: Policy-audience dissemination; no independent critique found. See public-attention evidence below.
- Author-institution summary The Federal Reserve Bank of St. Louis published a summary of the findings. Its research department issued the working paper, so this is not independent. St. Louis Fed
What the paper says
Source abstract · Source-supplied abstract wording; HTML entities and whitespace normalized. Not independently compared with the paper PDF.
We measure how workers use genAI for their jobs in a nationally representative survey linking genAI adoption to detailed occupations and tasks. Our data provide the first task-level genAI adoption indexes, which we show can inform analyses of genAI’s labor market impact. Exposure scores explain some, but far from all, of the variation in adoption across occupations and tasks. We also distinguish our indexes from measures based on genAI platform chat logs, which differ conceptually and tend to over-classify chats into generic activities spanning many occupations. Finally, we highlight that current adoption is widespread but shallow: genAI is used across many occupations and tasks, yet within most of them, fewer than half of workers adopt. This indicates substantial variation among workers doing very similar work, suggesting that understanding who adopts may matter as much as understanding which tasks genAI assists.
Claims to check
- Figure 6 reports that over 80% of occupations and 40% of tasks have adoption above 20%, while only 2.8% of tasks exceed 50%. These are weighted distributions from pooled August 2025–May 2026 surveys, excluding cells with fewer than 20 observations. [Working paper](https://s3.amazonaws.com/real.stlouisfed.org/wp/2026/2026-017.pdf)
- Table 2 reports worker-level adjusted R-squared of 0.176 for detailed-occupation fixed effects, versus 0.081 and 0.086 for two exposure measures: occupation and exposure explain limited individual adoption variation. [Working paper](https://s3.amazonaws.com/real.stlouisfed.org/wp/2026/2026-017.pdf)
- Across 332 intermediate work activities, survey-based AI-assisted task shares correlate 0.34 with Anthropic, 0.10 with Microsoft and 0.11 with OpenAI shares. The attribution of disagreement to generic-task classification needs assessment beyond verifying these correlations. [Authors' comparison](https://cepr.org/voxeu/columns/measuring-what-work-generative-ai-does-survey-evidence-versus-chat-logs)
Methodological or theoretical issues flagged for evaluation
The central challenge is external validity: demographic weighting cannot by itself eliminate selection into an online panel or reporting error. Reviewers should examine whether restricting questions to occupation-linked tasks omits AI-assisted activities systematically, and distinguish workers, worker-task pairs, conversations, prompts and time spent. Cross-platform differences may reflect genuine user and product specialization as well as classification error; disagreement alone does not establish which source is closer to the relevant target. Within-worker dependence and sparse cells matter for uncertainty and rankings, while residual adoption variation does not identify a causal effect of experience or training. The public data page lists adoption indexes but does not establish availability of this paper's complete current microdata and code; a package for a related paper is insufficient. Commissioning should therefore begin by confirming access and fixing the version. A focused survey-and-measurement evaluation, with clearly stated limits on platform validation, is more feasible than attempting to validate employment or welfare effects the paper does not directly identify. The paper's funders and fuller author list, including Tyler R. Schumacher, are recorded in the [working paper](https://s3.amazonaws.com/real.stlouisfed.org/wp/2026/2026-017.pdf).
Opus 5.5 re-evaluation (experimental)
Read with care. This is an experimental re-evaluation by a different model (Claude Opus 5.5, 2026-10-01). Across a calibration sample Opus scored about 15 points lower than GPT-5.5, so 15 points are added to compare it with GPT-5.5 scores. The calibration is a small sample (n=44) and is not human-validated, and the two models disagree about the order of papers within the top group. The AI score at the top of this page is the usual evaluation-priority score shifted by the difference between the adjusted Opus score and the earlier holistic score; the synthesis uses that shifted value, and the unshifted score is shown beside it. On the dashboard you can switch the Opus scores off. Read the methods note.
Opus rationale (AI-generated)
This is an NBER working paper by Bick (St. Louis Fed), Blandin (Vanderbilt) and Deming (Harvard), extending their influential nationally representative survey of generative-AI adoption to the occupation-task level. It is clearly within Unjournal's scope as quantitative social science on the economic impacts of AI. It speaks directly to a live measurement debate: whether theoretical exposure scores or AI-platform chat logs (such as the Anthropic Economic Index and OpenAI usage studies) are good proxies for actual work use. The finding that chat logs over-classify generic activities and that adoption is 'widespread but shallow' would matter to statistical agencies (BLS, Census BTOS team), Fed and CBO forecasters, and AI-labour researchers, and to the AI labs whose usage data are increasingly cited in policy. As a recent working paper with no visible independent review, an evaluation could usefully probe survey design, the validity of self-reported task use, task-cell sample sizes, and the fairness of the chat-log comparison while the authors can still revise. Concerns: it is descriptive measurement of US workers, with limited direct welfare or LMIC pathways. Unjournal assessors have historically rated tech/AI papers lower. The headline numbers will date quickly. I would place it in the monitor band, as a reasonable candidate if Unjournal wants an anchor evaluation in the AI-labour measurement cluster.
Dashboard details and provenance
Full AI dashboard scoring rationale
This is clearly in scope and worth keeping near the commissioning threshold: it could help the US Department of Labor, Bureau of Labor Statistics and Census Bureau choose how to measure workplace AI use, with downstream implications for worker-support and training decisions. Policy-facing uptake is already visible in a [Federal Reserve speech](https://www.federalreserve.gov/newsevents/speech/jefferson20261113a.htm), although I have not verified an operational decision based on the detailed indexes. As an NBER working paper, it presents a substantial scrutiny gap: influential measurement claims could propagate before their survey assumptions and cross-source comparisons receive a public independent assessment. A focused evaluation by a labor economist with survey-measurement expertise could assess representativeness, uncertainty and whether differences from chat logs support the proposed classification explanation; this would give users a consolidated account of which claims justify action. I score it just below prioritize-now because the connection to welfare-improving resource allocation remains indirect, and a bounded measurement review offers a clearer commissioning case than a broad evaluation of AI's employment effects.
AI decision-relevance rationale
The expected contribution is moderately positive: better adoption measurement could improve statistical collection, downstream labor-market research and eventually the targeting of worker support. The principal beneficiaries would initially be US workers and institutions; poorer workers' welfare interests deserve particular attention, but neither their realized gains nor LMIC transfer is established. Incremental value comes from supplementing existing exposure and platform measures, rather than capturing the total value of managing AI's economic transition. The specific measurement gap is tractable, while policy implementation and welfare effects are uncertain. There is no established direct animal-welfare or catastrophic-risk benefit. Documented communication and accessible indexes support realization, but changing technology and ambiguous links from adoption to productivity or displacement constrain the impact score.
AI timing assessment
Feedback remains actionable. The [Federal Reserve working-paper version](https://s3.amazonaws.com/real.stlouisfed.org/wp/2026/2026-017.pdf) identifies an initial April 27, 2026 version and an August 25 revision; the NBER issue is dated August 2026. Thus the current working-paper version is recent even though development began earlier. The paper acknowledges conference and seminar feedback, which provides some expert input but does not establish a public claim-by-claim assessment. The supplied targeted search found no distinct substantial-review leads, and its citation-search component failed. That uncertainty does not justify declaring the paper unreviewed in every sense. Prompt evaluation is valuable before measurement choices become embedded in subsequent work, with the current version fixed explicitly at commissioning.
Earlier public-scrutiny search
The supplied November 16 targeted search and public-annotation check identified no substantial independent review, replication or methodological critique of this specific paper. Their coverage is incomplete: the OpenAlex search failed, and annotations on other versions or hosts could be missed. Consequently, no supplied evidence establishes that particular claims have already been independently checked. The consequential remaining gap concerns representativeness and task coverage, uncertainty in adoption rankings, and whether survey–platform discrepancies identify classification problems rather than different populations, denominators or usage intensity. A feasible public evaluation could reproduce available aggregates, examine survey construction and sensitivity, and separate robust descriptive findings from explanations requiring stronger validation. Computational reproduction alone would not validate sampling assumptions or policy interpretations. The neglectedness score is high because errors could affect policy-facing measurement and downstream research; it remains below the maximum because operational reliance is unverified and the search cannot establish the absence of scrutiny.
No supporting source links are stored.
Intake, review, and crux connections
Public attention and use
AI-assisted review of the linked public sources.
Author-institution summary visible circulation
The Federal Reserve Bank of St. Louis published a summary of the findings. Its research department issued the working paper, so this is not independent.
Source: St. Louis Fed · Relationship: author institution
Policy-audience column public mention
A CEPR VoxEU column compares survey evidence with chat-log evidence on the same question. We did not read it to confirm whether it discusses this paper directly, nor its authors' relationship to it.
Source: CEPR VoxEU · Relationship: relationship not checked
Research dissemination listing / discoverability
The paper is listed by the Washington Center for Equitable Growth.
Source: Equitable Growth · Relationship: bibliographic index
What the search did not establish
- No exact-title EA Forum or LessWrong discussion surfaced in the targeted search.
- No comparison paper or replication using other survey sources surfaced.
- The most decision-relevant next signal is whether forecasters and agencies adopt its adoption indexes.
Targeted search checked 2026-10-01. Search scope: Targeted exact-title searches across the open web, institutional and author pages, news and public social-media results, EA Forum/LessWrong, and policy/white-paper contexts. Evidence records distinguish commissioning or report use from independent discussion, media attention, indexing, and post-publication policy use. A search miss is reported as uncertainty, not proof of absence. Entries with check_depth "quick" rest on roughly one to two searches and should be read as especially uncertain; entries checked on 2026-10-01 were added for the shortlist of papers without human ratings.