CoChat
Sign Up
Cited by 209Labor market dynamics and wage inequalityArtificial Intelligence ApplicationsEconomic Growth and Productivity

The DOI matches its Crossref record; 32 of the 32 references it cites were found in a registry or an index; none of the 39 sources its claims cite is checked for support yet.

The Rapid Adoption of Generative AI

Open access
Bick, Alexander, Blandin, Adam, Deming, David·2024·NBER Working Paper
DOI 10.3386/w32966
Open this paper in CoChat — free account

Abstract

Generative artificial intelligence (AI) is a potentially important new technology, but its impact on the economy depends on the speed and intensity of adoption.This paper reports results from a series of nationally representative U.S. surveys of generative AI use at work and at home.As of late 2024, nearly 40 percent of the U.S. population age 18-64 uses generative AI. 23 percent of employed respondents had used generative AI for work at least once in the previous week, and 9 percent used it every work day.Relative to each technology's first mass-market product launch, work adoption of generative AI has been as fast as the personal computer (PC), and overall adoption has been faster than either PCs or the internet.Generative AI and PCs have very similar early adoption patterns by education, occupation, and other characteristics.Between 1 and 5 percent of all work hours are currently assisted by generative AI, and respondents report time savings equivalent to 1.4 percent of total work hours.This suggests that substantial productivity gains from generative AI are possible.

Overview

AI snapshot

Written from the paper’s full text.

TL;DR

Using nationally representative U.S. surveys fielded in August and November 2024, the paper finds that nearly 40% of adults age 18–64 use generative AI, with 23% of employed respondents using it for work in the previous week. Adoption has been as fast as PCs at work and faster overall than PCs or the internet, with estimated time savings equivalent to 1.4% of total work hours and a potential aggregate productivity gain of about 1.1%.

Key findings

  • As of August and November 2024, 39.6% of respondents used generative AI either for work or outside work; 26.5% of employed respondents used it for work, including 9.0% every workday and 13.9% on some but not all workdays. Outside work, 35.5% used it, but only 5.2% every day.
  • Work adoption of generative AI after two years (27%) was similar to PC work adoption after three years (25%), while non-work adoption was much higher for generative AI (34%) than PCs (5%). Overall adoption was faster than both PCs and the internet.
  • Between 1 and 5 percent of all U.S. work hours involved direct generative AI use. Among users, mean time savings were 5.4% of work hours; including non-users, 1.4% of total hours. A standard aggregate production model implies a potential productivity gain of 1.1% using an AI-exposure-adjusted labor share of 0.57 (or 1.9% before that adjustment).
  • Adoption patterns by education, age, occupation, and industry closely matched early PC adoption; predicted task exposure from Eloundou et al. (2024) correlated with actual adoption at ρ = 0.67. The most useful work tasks were writing (39.5%), administrative tasks (25.6%), and interpreting/translating/summarizing (22.7%).

Methods

  • Main data are the Real-Time Population Survey (RPS), an online national labor market survey of U.S. adults age 18–64, fielded in August 2024 (N = 5,014) and November 2024 (N = 5,329), with a combined sample of more than 10,000 respondents. The paper also validates results using the December 2024 Survey of Working Arrangements and Attitudes (SWAA, N = 4,698).
  • The RPS mirrors the Current Population Survey (CPS) in question wording and structure; sample weights are constructed via raking to align with CPS demographic, employment, and occupational composition. The final weighted sample is 96.3% of initially collected responses.
  • Generative AI questions measure work and non-work use, frequency, daily time spent, products used, tasks helped, and self-reported time savings. Comparisons to PCs and the internet use CPS Computer and Internet Use supplement data from 1984–2003 and 2001–2009, plus ITU internet data. Aggregate productivity estimates use a Cobb-Douglas production model.

Limitations

  • The paper states that its data cannot distinguish whether higher wages among intensive generative AI users reflect task variation, worker selection, or direct productivity/wage effects. It also cautions that potential productivity gains may not appear in statistics if workers take time savings as on-the-job leisure, and that firm adoption lags far behind worker adoption.
  • The aggregate productivity calculation assumes all time savings are internalized as higher worker output and that labor supply is perfectly substitutable across workers; the paper notes imperfect substitution would reduce the estimated impact.
  • The RPS uses a commercial online panel that is not a random sample, though the authors weight it to be nationally representative and validate it against the CPS and SWAA. The text also notes that NBER working papers are circulated for discussion and have not been peer-reviewed.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3386/w32966

Cites

32 of 32 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

0 of 39 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    Between 1 and 5 percent of all work hours are currently assisted by generative AI.

    “Between 1 and 5 percent of all work hours are currently assisted by generative AI, and respondents report time savings equivalent to 1.4 percent of total work hours.”
    Quote in the paperSubstantial claim
  • Findingown data

    Respondents report time savings equivalent to 1.4 percent of total work hours.

    “Between 1 and 5 percent of all work hours are currently assisted by generative AI, and respondents report time savings equivalent to 1.4 percent of total work hours.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Respondents' self-reported time savings accurately measure the actual time savings from using genAI.
  • Findingown data

    As of late 2024, nearly 40 percent of the U.S. population age 18-64 uses generative AI.

    “As of late 2024, nearly 40 percent of the U.S. population age 18-64 uses generative AI.”
    Quote in the paperSubstantial claim
  • Findingown data

    In August and November of 2024, 26 percent of employed respondents reported using genAI for work.

    “In August and November of 2024, 26 percent of employed respondents reported using genAI for work: 9 percent used it every work day, 14 percent on some but not all work days in the previous week, and 3 percent did not use it in the last week.”
    Quote in the paperSubstantial claim
  • Findingown data

    9 percent of employed respondents used generative AI every work day.

    “23 percent of employed respondents had used generative AI for work at least once in the previous week, and 9 percent used it every work day.”
    Quote in the paperSubstantial claim
  • It may state two separate claims as one. 59
  • Probably not a substantial claim. 32
  • The quote only partly states this claim. 2

  • It may state two separate claims as one.

    Findingown data

    52 percent of daily genAI users also report using it an hour or more each day, compared to only 7 percent of those who used it in the last month but not the last week.

    “52 percent of daily genAI users also report using it an hour or more each day, compared to only 7 percent of those who used it in the last month but not the last week.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Findingnone given

    In 1984, 25 percent of workers reported using computers for their job, compared with 27 percent of workers who report using genAI for their job in 2024.

    “In 1984, 25 percent of workers reported using computers for their job, compared with 27 percent of workers who report using genAI for their job in 2024.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: In October 1984, the CPS launched a supplemental module to study computer adoption.

  • It may state two separate claims as one.

    Findingown data

    The unweighted distributions of usual weekly earnings in the RPS and CPS are already similar, and the weights improve the fit further.

    “The unweighted distributions are already similar, and the weights improve the fit further.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Findingown data

    ChatGPT is used most frequently, followed by Gemini.

    “ChatGPT is used most frequently, followed by Gemini, and our estimates of ChatGPT usage are similar to other recent surveys.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Findingown data

    In 2024, workers under the age of 50 were much more likely to use genAI than those over 50, and the same was true for PCs in 1984.

    “In 2024, workers under the age of 50 were much more likely to use genAI than those over 50, and the same was true for PCs in 1984.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Findingreasoning

    Evaluating equation (10) using the midpoint of the lower and upper bounds for l_i yields a wage-weighted share of total work hours spent using genAI of 5.7%, which implies a productivity parameter value of gamma = 0.33.

    “Evaluating (10) using the midpoint of these bounds, l i ≡ (l i +l i )/2 yields a wage-weighted share of total work hours spent using genAI of 5.7%, which implies a productivity parameter value of γ = 0.33.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: There is no change in TFP, capital, and hours worked.
    • The authors state this as an assumption: The economy's aggregate output is modeled using a Cobb-Douglas aggregate production function.
    • The authors state this as an assumption: In a competitive labor market, each worker is paid their marginal product of labor.

  • It may state two separate claims as one.

    Findingown data

    The most commonly used genAI products are ChatGPT (28 percent of all respondents), Gemini (17 percent) and embedded products such as Microsoft Copilot (14 percent).

    “The most commonly used products are ChatGPT (28 percent of all respondents), Gemini (17 percent) and embedded products such as Microsoft Copilot (14 percent).”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Findingown data

    We use survey data on actual adoption to estimate a wage-weighted usage share of 5.7%, which yields slightly higher estimates.

    “We use survey data on actual adoption to estimate a wage-weighted usage share of 5.7%, which yields slightly higher estimates.”
    Quote in the paperSubstantial claim

  • It may state two separate claims as one.

    Findingown data

    Workers in computer, mathematical, and management occupations use genAI between 9 and 12 percent of their work hours, and it saves them between 2.1 and 2.5 percent of their work hours.

    “Workers in computer, mathematical, and management occupations use genAI between 9 and 12 percent of their work hours, and it saves them between 2.1 and 2.5 percent of their work hours.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Findingown data

    Among workers who used genAI every day last week, 33.3 percent said it saved them four hours or more, compared to only 11.5 percent of those who used it only one day last week.

    “Among workers who used genAI every day last week, 33.3 percent said it saved them four hours or more, compared to only 11.5 percent of those who used it only one day last week.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Findingown data

    Workers report that genAI helps in a broad range of tasks, the most useful being writing, administrative support, and interpreting and summarizing information.

    “Workers report that genAI helps in a broad range of tasks, the most useful being writing, administrative support, and interpreting and summarizing information.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Findingcitation

    Almost all occupation groups have at least a modest amount of genAI adoption, which is consistent with the breadth of exposure found in Eloundou, Manning, Mishkin, and Rock (2024).

    “Almost all occupation groups have at least a modest amount of genAI adoption, which is consistent with the breadth of exposure found in Eloundou, Manning, Mishkin, and Rock (2024).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Eloundou, Manning, Mishkin, and Rock (2024)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • It may state two separate claims as one.

    Findingown data

    Among those who use genAI for work, 34 percent used it every workday in the previous week, 52 percent used it some but not all days in the previous week, and 14 percent did not use it at all the previous week.

    “Among those who use genAI for work, 34 percent used it every workday in the previous week, 52 percent used it some but not all days in the previous week, and 14 percent did not use it at all the previous week.”
    Quote in the paperSubstantial claim

  • It may state two separate claims as one.

    Findingnone given

    Some occupations have high adoption relative to predicted exposure (especially managers) and some have low adoption relative to predicted exposure (especially office and administrative support occupations).

    “At the same time, some occupations have high adoption relative to predicted exposure (especially managers) and some have low adoption relative to predicted exposure (especially office and administrative support occupations).”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Findingown data

    Managers have relatively high adoption rates relative to predicted exposure, while legal and social services and office and administration occupations have relatively low adoption rates.

    “At the same time, managers have relatively high adoption rates relative to predicted exposure, while legal and social services and office and administration occupations have relatively low adoption rates.”
    Quote in the paperSubstantial claim

  • It may state two separate claims as one.

    Findingown data

    Outside of work, usage is higher overall (35.5 percent) but less intensive, with only 5.2 percent of all respondents using it every day last week.

    “Outside of work, usage is higher overall (35.5 percent) but less intensive, with only 5.2 percent of all respondents using it every day last week.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Findingown data

    The RPS and SWAA waves yield very similar adoption rates, both overall and by frequency of use, as well as the type of products being used.

    “The RPS and SWAA waves yield very similar adoption rates, both overall and by frequency of use, as well as the type of products being used.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Findingown data

    Personal service and blue-collar occupations have relatively low predicted exposure and also relatively low adoption, while computer and mathematical occupations have both high predicted exposure and high adoption.

    “Personal service and blue-collar occupations have relatively low predicted exposure and also relatively low adoption, while computer and mathematical occupations have both high predicted exposure and high adoption.”
    Quote in the paperSubstantial claim

  • It may state two separate claims as one.

    Findingreasoning

    Overall genAI adoption (including work and non-work usage) has been faster than both the personal computer and the internet.

    “While we cannot distinguish between work and non-work internet use, overall genAI adoption (including work and non-work usage) has been faster than both the personal computer and the internet.”
    Quote in the paperSubstantial claim

  • It may state two separate claims as one.

    Findingown data

    Workers in personal service occupations use genAI in only 1.3 percent of their work hours, and it saves them only 0.4 percent of work time.

    “By contrast, workers in personal service occupations use genAI in only 1.3 percent of their work hours, and it saves them only 0.4 percent of work time.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Findingown data

    40.7 percent of SWAA respondents reported using genAI either for work or outside of work, compared with 39.6 percent of RPS respondents.

    “40.7 percent of SWAA respondents reported using genAI either for work or outside of work, compared with 39.6 percent of RPS respondents.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.

    Findingnone given

    Women were earlier users of PCs, in part because of rapid adoption of computers in white-collar clerical occupations.

    “However, women were earlier users of PCs, in part because of rapid adoption of computers in white-collar clerical occupations.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Findingown data

    We also find similar estimates of use for work (27.1 percent in the SWAA versus 26.5 percent in the RPS) and outside of work (35.5 percent in the SWAA versus 33.7 percent in the RPS).

    “We also find similar estimates of use for work (27.1 percent in the SWAA versus 26.5 percent in the RPS) and outside of work (35.5 percent in the SWAA versus 33.7 percent in the RPS).”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Findingown data

    Among workers who used genAI in the previous week, 33.2 percent said genAI saved them an hour or less of time last week, 26.4 percent reported saving two hours, 20.0 percent reported three hours, and 20.4 percent reported four hours or more.

    “Among workers who used genAI in the previous week, 33.2 percent said that genAI saved them an hour or less of time last week, 26.4 percent reported saving two hours, 20.0 percent reported three hours, and 20.4 percent reported four hours or more.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Findingown data

    32 percent of genAI users report using genAI for an hour or more per day at work, 47 percent used it for between 15 and 60 minutes per day, and 21 percent used it for less than 15 minutes per day.

    “Figure 10 shows that 32 percent of genAI users report using genAI for an hour or more per day at work, 47 percent used it for between 15 and 60 minutes per day, and 21 percent used it for less than 15 minutes per day.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Findingown data

    Information services has both the largest share of work hours spent using genAI (14.0 percent) and the highest time savings (2.6 percent).

    “Across industries, information services has both the largest share of work hours spent using genAI (14.0 percent) and the highest time savings (2.6 percent).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim. The paper’s conclusions may not depend on it.

    Findingreasoning

    This estimate is slightly higher than the average time savings due to genAI from the previous section (1.4 percent).

    “This estimate is slightly higher than the average time savings due to genAI from the previous section (1.4 percent).”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Findingown data

    The first bar shows that 39.6 percent of respondents report using genAI either at work or at home: 9.2 percent used it every day, 23.1 percent used it some but not all days, and 7.3 percent did not use it in the previous week.

    “The first bar shows that 39.6 percent of respondents report using genAI either at work or at home: 9.2 percent used it every day, 23.1 percent used it some but not all days, and 7.3 percent did not use it in the previous week.”
    Quote in the paperSubstantial claim

  • It may state two separate claims as one.

    Findingown data

    On days that workers used genAI, 32 percent used it for an hour or more per day, 47 percent used it between 15-59 minutes per day, and 21 percent used it for less than 15 minutes per day.

    “On days that workers used genAI, 32 percent used it for an hour or more per day, 47 percent used it between 15-59 minutes per day, and 21 percent used it for less than 15 minutes per day.”
    Quote in the paperSubstantial claim

  • It may state two separate claims as one.

    Findingnone given

    Internet adoption increased rapidly from 20 percent in year two to 60 percent in year seven, and then more gradually to 90 percent over the ensuing two decades.

    “Internet adoption increased rapidly from 20 percent in year two to 60 percent in year seven, and then more gradually to 90 percent over the ensuing two decades.”
    Quote in the paperSubstantial claim

  • It may state two separate claims as one.

    Findingown data

    Leisure, accommodation, and other services has both the lowest share of work hours spent using genAI (2.3 percent) and the lowest time savings (0.6 percent).

    “Leisure, accommodation, and other services has both the lowest share of work hours spent using genAI (2.3 percent) and the lowest time savings (0.6 percent).”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Findingown data

    Use of both PCs and genAI is associated with a weekly earnings premium of 21 log points, a wage premium of about 15 log points, and 5 log points weekly hours.

    “Use of both technologies is associated with a weekly earnings premium of 21 log points, a wage premium of about 15 log pints, and 5 log points weekly hours.”
    Quote in the paperSubstantial claim

  • It may state two separate claims as one.

    Findingown data

    The second bar shows that 26.5 percent of workers used genAI at work: 9.0 percent used it every work day and 13.9 percent on some but not all workdays, bringing total weekly usage to 22.9 percent.

    “The second bar shows that 26.5 percent of workers used genAI at work: 9.0 percent used it every work day and 13.9 percent on some but not all workdays, bringing total weekly usage to 22.9 percent.”
    Quote in the paperSubstantial claim

  • It may state two separate claims as one.

    Findingreasoning

    Early genAI usage is higher for men, while computer usage was initially higher for women.

    “The one exception is gender: early genAI usage is higher for men, while computer usage was initially higher for women, reflecting rapid computer adoption among secretaries and other administrative occupations, which had large female shares.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Findingnone given

    Adoption patterns by occupation and industry are similar for computers and genAI, although computer adoption was somewhat more concentrated in a few occupations whereas genAI adoption is more evenly dispersed.

    “We also find similar adoption patterns by occupation and industry, although computer adoption was somewhat more concentrated in a few occupations whereas genAI adoption is more evenly dispersed.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Findingown data

    The highest ranked tasks at work were writing (39.5 percent), administrative tasks (25.6 percent), interpreting/translating/summarizing text or data (22.7 percent), and searching for facts or information (18.0 percent).

    “The highest ranked tasks at work were writing (39.5 percent), administrative tasks (25.6 percent), interpreting/translating/summarizing text or data (22.7 percent), and searching for facts or information (18.0 percent).”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.

    Findingnone given

    Our estimates of ChatGPT usage are similar to other recent surveys.

    “ChatGPT is used most frequently, followed by Gemini, and our estimates of ChatGPT usage are similar to other recent surveys.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Findingown data

    GenAI adoption is 7.5 percentage points higher for men, while early PC adoption was 6 percentage points higher for women.

    “Figure 5c shows that genAI adoption is 7.5 percentage points higher for men, while early PC adoption was 6 percentage points higher for women.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Findingown data

    Both computer and genAI adoption have been faster for younger, more educated, and higher-wage workers.

    “Both computer and genAI adoption have been faster for younger, more educated, and higher-wage workers.”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Findingown data

    The speed of genAI adoption has been similar to the adoption of personal computers in the 1980s.

    “Still, it is notable that the speed of genAI adoption has been similar to the adoption of personal computers in the 1980s, since the latter led to an accelerate of productivity growth and rising inequality (e.g.Autor, Levy, and Murnane 2003).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Autor, Levy, and Murnane 2003

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds. It may state two separate claims as one.

    Findingnone given

    Time savings are highly correlated with the intensity of genAI use and thus vary widely across occupations and industries.

    “Time savings are highly correlated with the intensity of genAI use and thus vary widely across occupations and industries.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Respondents' self-reported time savings accurately measure the actual time savings from using genAI.

  • Probably not a substantial claim.

    Study design and conductreasoning

    The RPS closely follows the CIU question wording and structure, allowing comparison of adoption of different technologies relative to their first mass market product release.

    “We closely follow the CIU question wording and structure, which allows us to compare adoption of different technologies relative to their first mass market product release.”
    Quote in the paperSubstantial claim

  • It may state two separate claims as one.

    Study design and conductnone given

    The RPS asks the same core questions and follows the same timing and structure of the Current Population Survey (CPS).

    “The RPS asks the same core questions and follows the same timing and structure of the Current Population Survey (CPS).”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim. The paper’s conclusions may not depend on it.

    Premisecitation

    Several recent studies find that genAI improves worker productivity.

    “Several recent studies find that genAI improves worker productivity (Brynjolfsson, Li, and Raymond, 2023; Cui et al., 2024; Dell'Acqua et al., 2023; Noy and Zhang, 2023; Peng, Kalliamvakou, Cihon, and Demirer, 2023; Toner-Rodgers, 2024)”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Brynjolfsson, Li, and Raymond, 2023

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Cui et al., 2024

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Dell'Acqua et al., 2023

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • It may state two separate claims as one.

    Study design and conductreasoning

    The parallel structure of the RPS to the CPS allows benchmarking of survey estimates against the CPS and construction of weights that ensure a nationally representative sample.

    “This parallel structure allows us to benchmark our survey estimates against the CPS and to construct weights that ensure a nationally representative sample.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim.

    Study design and conductreasoning

    To minimize measurement error, the RPS and CPS samples are restricted to individuals with weekly earnings below the CPS topcode of $3,960.00 and an implied hourly wage of at least the federal minimum wage of $7.25.

    “To minimize measurement error, we restrict both the RPS and CPS samples to individuals with (i) weekly earnings below the CPS topcode of $3,960.00,(ii) an implied hourly wage of at least the federal minimum wage of $7.25.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

    Study design and conductown data

    Appendix D.3 presents the same figures for each survey wave and analogous plots for industry and college major, which show qualitatively and quantitatively similar patterns to the occupation shares.

    “Appendix D.3 presents the same figures for each survey wave and analogous plots for industry and college major, which show qualitatively and quantitatively similar patterns to the occupation shares.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Interpretationown data

    Our results offer partial validation of task-based exposure predictions, supporting the utility of these measures for researchers.

    “Our results offer partial validation of task-based exposure predictions, supporting the utility of these measures for researchers.”
    Quote in the paperSubstantial claim

  • It may state two separate claims as one.

    Study design and conductown data

    Results from the August and November waves separately, as well as a pilot survey from June 2024, all yield similar estimates.

    “Results from the August and November waves separately, as well as a pilot survey from June 2024, are included in Online Appendix B. All three waves yield similar estimates.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim. The paper’s conclusions may not depend on it.

    Study design and conductnone given

    Our weighting scheme incorporates a broader set of variables, including employment status.

    “Our weighting scheme incorporates a broader set of variables, including employment status.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Study design and conductcitation

    The RPS survey matches questions on demographics and labor market outcomes in the basic CPS and CPS Outgoing Rotation Group, using the same word-for-word phrasing when practical and replicating the intricate sequence of questions necessary to elicit labor market outcomes in a manner consistent with the CPS.

    “The RPS survey matches questions on demographics and labor market outcomes in the basic CPS and CPS Outgoing Rotation Group, using the same word-for-word phrasing when practical and replicating the intricate sequence of questions necessary to elicit labor market outcomes in a manner consistent with the CPS (US Census Bureau, 2015).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites US Census Bureau, 2015

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim. The paper’s conclusions may not depend on it.

    Study design and conductcitation

    Bick, Blandin, Caplan, and Caplan (2024a,b) show that estimates of work from home rates in the RPS align with estimates in the American Community Survey, the American Time Use Survey, the CPS, and the Survey of Income and Program Participation.

    “As an additional point of comparison,Bick, Blandin, Caplan, and Caplan (2024a,b) show that estimates of work from home rates in the RPS align with estimates in the American Community Survey, the American Time Use Survey, the CPS, and the Survey of Income and Program Participation.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Bick, Blandin, Caplan, and Caplan (2024a,b)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • It may state two separate claims as one.

    Study design and conductnone given

    The paper's data come from the Real-Time Population Survey (RPS), a survey constructed and weighted to be nationally representative and following the same survey design as the CPS.

    “Our data come from the Real-Time Population Survey (RPS), a survey that is constructed and weighted to be nationally representative and follows the same survey design as the CPS, a widely used national data source.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim. The paper’s conclusions may not depend on it.

    Study design and conductnone given

    All surveys started fielding during the same weeks that the CPS conducted its corresponding surveys.

    “All surveys started fielding during the same weeks that the CPS conducted its corresponding surveys.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

    Premisecitation

    Acemoglu (2024) assumes that these productivity gains would evolve gradually over the next ten years, whereas our estimates are based on current usage.

    “A final difference is that Acemoglu (2024) assumes that these productivity gains would evolve gradually over the next ten years, whereas our estimates are based on current usage.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Acemoglu (2024)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim. The paper’s conclusions may not depend on it.

    Premisecitation

    Bick and Blandin (2023) show that the RPS produced very similar statistics for employment, hours worked, earnings, industry composition, and employee tenure during the pandemic.

    “Bick and Blandin (2023) show that the RPS produced very similar statistics for employment, hours worked, earnings, industry composition, and employee tenure during the pandemic.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Bick and Blandin (2023)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • It may state two separate claims as one.

    Study design and conductown data

    Employment rates are very similar across the CPS and RPS, although individuals classified as unemployed according to the CPS definition are somewhat overrepresented in the RPS.

    “Employment rates are very similar across the two surveys, although individuals classified as unemployed according to the CPS definition are somewhat overrepresented in the RPS.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Interpretationcitation

    Firm adoption of genAI lags far behind worker adoption, suggesting that worker adoption is still mostly informal.

    “While firm adoption may increase over time, it still lags far behind, suggesting that worker adoption is still mostly informal.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim. It may be too vague to check against evidence. The paper’s conclusions may not depend on it.

    Study design and conductcitation

    According to Qualtrics staff, the screener question provides the best results in terms of screening out respondents.

    “According to Qualtrics staff, this question provides the best results in terms of screening out respondents.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Qualtrics staff

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • It may state two separate claims as one.

    Study design and conductown data

    Individuals aged 18 to 24 and with no more than a high school degree are underrepresented in the RPS relative to the CPS, while individuals with household income of $50,000 or less are overrepresented.

    “The most notable discrepancies are that individuals aged 18 to 24 and with no more than a high school degree are underrepresented in the RPS relative to the CPS, while individuals with household income of $50,000 or less are overrepresented.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim. The paper’s conclusions may not depend on it.

    Premisecitation

    Pew Research Center (McClain, 2024) found that in February 2024, 27 percent of U.S. adults age 18-64 reported ever having used ChatGPT.

    “They find that in February 2024, 27 percent of U.S. adults age 18-64 reported ever having used ChatGPT, compared to 29 percent of individuals who report using ChatGPT in our survey in August and November 2024.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites McClain, 2024

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • It may state two separate claims as one.

    Interpretationown data

    Relative to the introduction of the first mass-market product, genAI has been adopted at a faster pace than both PCs and the internet.

    “We conclude that, relative to the introduction of the first mass-market product, genAI has been adopted at a faster pace than both PCs and the internet.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The first mass-market computer was the IBM PC, which was released in August 1981.
    • Rests on an assumption: The first mass-market genAI product was ChatGPT, which was released in November 2022.

  • Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.

    Interpretationnone given

    Imperfect substitution across worker types with differing usage patterns would reduce the impact of genAI on aggregate productivity.

    “However, imperfect substitution across worker types with differing usage patterns would reduce the impact of genAI on aggregate productivity.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Labor supply is perfectly substitutable across workers.

  • The quote only partly states this claim.

    Interpretationown data

    The pattern of genAI adoption by education is strikingly similar to that of PCs.

    “This pattern is strikingly similar for PC's.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim. The paper may give no reason why it holds. It may state two separate claims as one.

    Interpretationreasoning

    If workers can complete the same tasks in less time without their employer's knowledge, they may take their time savings as on-the-job leisure, which would increase welfare but not productivity.

    “If workers are now able to complete the same tasks in less time without their employer's knowledge, they may take their time savings as on-the-job leisure, which would increase welfare but not productivity.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Study design and conductreasoning

    Replicating key portions of an existing high-quality survey ensures that survey concepts are comparable, which allows researchers to validate RPS outcomes against a widely used benchmark with a larger sample size and, where necessary, to construct sample weights.

    “Replicating key portions of an existing high-quality survey ensures that survey concepts are comparable, which allows researchers to validate RPS outcomes against a widely used benchmark with a larger sample size and, where necessary, to construct sample weights.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim.

    Premisecitation

    Eloundou, Manning, Mishkin, and Rock (2024) assign each occupation a predicted exposure score based on what fraction of that occupation's tasks they believe will be substantially affected by genAI.

    “Eloundou, Manning, Mishkin, and Rock (2024) assign each occupation a predicted exposure score based on what fraction of that occupation's tasks they believe will be substantially affected by genAI.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Eloundou, Manning, Mishkin, and Rock (2024)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim. It may state two separate claims as one.

    Study design and conductcitation

    We construct sample weights using the raking algorithm of Deming and Stephan (1940), ensuring that the weighted sample proportions align with the demographic characteristics targeted in the sampling procedure.

    “To address remaining discrepancies, we construct sample weights using the raking algorithm of Deming and Stephan (1940), ensuring that the weighted sample proportions align with the demographic characteristics targeted in the sampling procedure.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Deming and Stephan (1940)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

    Study design and conductreasoning

    The weighting scheme necessitated dropping some observations due to missing occupation and other discrepancies, resulting in a final sample size of 96.3% of initially collected responses.

    “The weighting scheme necessitated dropping some observations due to missing occupation and other discrepancies, resulting in a final sample size of 96.3% of initially collected responses.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim. The paper’s conclusions may not depend on it.

    Interpretationreasoning

    For an individual working 40 hours per week, saving 5.4 percent of work hours implies a time savings of 2.2 hours per week.

    “For an individual working 40 hours per week, saving 5.4 percent of work hours implies a time savings of 2.2 hours per week.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim. The paper’s conclusions may not depend on it.

    Premisecitation

    Acemoglu (2024) estimates a potential productivity gain of 0.7 percent using estimates of task-based exposure to genAI rather than actual adoption data.

    “This is similar to Acemoglu (2024), who estimates a potential productivity gain of 0.7 percent using estimates of task-based exposure to genAI rather than actual adoption data.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Acemoglu (2024)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim. The paper’s conclusions may not depend on it.

    Premisecitation

    Fletcher and Nielsen (2024) found that 18 percent of U.S. adults used ChatGPT at least weekly.

    “They found that 18 percent of U.S. adults used ChatGPT at least weekly, compared to 19 percent who used ChatGPT in the previous week in our study.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Fletcher and Nielsen (2024)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim. It may state two separate claims as one.

    Study design and conductnone given

    Individuals who take the survey too fast, i.e. in less than 50% of the median time in a soft launch, or who do not state that they will provide their best answers, are automatically dropped from the sample.

    “Individuals who take the survey too fast, i.e. in less than 50% of the median time in a soft launch, or who do not state that they will provide their best answers, are automatically dropped from the sample.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim. The paper’s conclusions may not depend on it.

    Interpretationreasoning

    Lower adoption costs may also allow genAI to diffuse more rapidly.

    “On the other hand, lower adoption costs may also allow genAI to diffuse more rapidly.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

    Premisenone given

    The Qualtrics panel includes about 15 million members and is not a random sample of the U.S. population.

    “The Qualtrics panel includes about 15 million members and is not a random sample of the U.S. population.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim. It may state two separate claims as one.

    Premisenone given

    The SWAA has been running monthly since May 2020 and is conducted using a different survey provider (IncQuery), but targets a similar sample that is intended to be representative of the U.S. ages 20-64 by age, sex, income, race, and region.

    “The SWAA has been running monthly since May 2020 and is conducted using a different survey provider (IncQuery), but targets a similar sample that is intended to be representative of the U.S. ages 20-64 by age, sex, income, race, and region.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

    Premisecitation

    The adoption of personal computers in the 1980s led to accelerated productivity growth and rising inequality.

    “since the latter led to an accelerate of productivity growth and rising inequality (e.g.Autor, Levy, and Murnane 2003).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Autor, Levy, and Murnane 2003

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

    Premisecitation

    The widespread adoption of computers and related information technologies beginning in the 1980s contributed to rising aggregate productivity and growing income inequality.

    “The widespread adoption of computers and related information technologies beginning in the 1980s contributed to rising aggregate productivity and growing income inequality (Autor, Levy, and Murnane, 2003).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Autor, Levy, and Murnane, 2003

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim.

    Study design and conductown data

    In December 2024, we ran our genAI module through a different online survey, the Survey of Working Arrangements and Attitudes (SWAA).

    “To provide a second source of data, in December 2024 we ran our genAI module through a different online survey, the Survey of Working Arrangements and Attitudes (SWAA) (Barrero, Bloom, and Davis, 2021).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Barrero, Bloom, and Davis, 2021

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

    Interpretationcitation

    Our estimate is based on reported genAI use and time savings, while Acemoglu (2024) uses experimental evidence on productivity gains from genAI usage and task-based predictions of genAI exposure.

    “Our estimate is based on reported genAI use and time savings, while Acemoglu (2024) uses experimental evidence on productivity gains from genAI usage and task-based predictions of genAI exposure.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Acemoglu (2024)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim.

    Study design and conductreasoning

    We use existing technology adoption questions from the CPS as a template for our own questions regarding genAI.

    “To facilitate this comparison, we use existing technology adoption questions from the CPS as a template for our own questions regarding genAI.”
    Quote in the paperSubstantial claim
81 of 81

  • The paper relies on this without support.

    In October 1984, the CPS launched a supplemental module to study computer adoption.

    “In October 1984, the CPS launched a supplemental module to study the adoption of a new technology: the computer.”

  • The paper relies on this without support.

    Respondents' self-reported time savings accurately measure the actual time savings from using genAI.

    “We ask users how many additional hours they would have needed to complete the same amount of work in the previous week if they had not had access to genAI.”

  • The paper relies on this without support.

    Time saved by genAI translates one-to-one into aggregate productivity gains.

    “Using a standard model of aggregate production, we estimate a potential aggregate productivity gain of 1.1 percent at current levels of worker genAI usage.”

  • The paper relies on this without support.

    The choice of whether to adopt a given technology is a function of both its benefits and its cost.

    “The choice of whether to adopt a given technology is a function of both its benefits and its cost.”

  • The paper relies on this without support.

    PC adoption required the purchase of relatively expensive hardware that was initially not very mobile, implying a substantial cost of adoption.

    “PC adoption required the purchase of relatively expensive hardware that was initially not very mobile, implying a substantial cost of adoption.”

  • The paper relies on this without support.

    Internet adoption required the purchase of a modem and an Internet Service Provider subscription.

    “Internet adoption required the purchase of a modem and an Internet Service Provider subscription.”

  • The paper relies on this without support.

    Many genAI products are currently free or inexpensive and user-friendly.

    “By contrast, many genAI products are currently free or inexpensive and user-friendly.”

  • The paper relies on this without support.

    Online-based surveys may suffer from selection based on unobservable characteristics that are correlated with genAI use.

    “A potential concern with online-based surveys is that they may suffer from selection based on unobservable characteristics that are correlated with genAI use.”

  • The paper relies on this without support.

    The first mass-market computer was the IBM PC, which was released in August 1981.

    “The first mass-market computer was the IBM PC, which was released in August 1981.”

  • The paper relies on this without support.

    The first mass-market genAI product was ChatGPT, which was released in November 2022.

    “The first mass-market genAI product was ChatGPT, which was released in November 2022.”

  • The authors state this as an assumption.

    The AI-exposure-adjusted labor share is 0.57.

    “If we assume an AI-exposure-adjusted labor share of 0.57, following Acemoglu (2024), the implied potential productivity gain is 1.1%.”

  • The authors state this as an assumption.

    There is no change in TFP, capital, and hours worked.

    “Assuming no change in TFP, capital, and hours worked, the percent change in aggregate productivity from genAI equals the percent change in aggregate output.”

  • The authors state this as an assumption.

    Labor supply is perfectly substitutable across workers.

    “We also assume above that labor supply is perfectly substitutable across workers.”

  • The authors state this as an assumption.

    All gains from genAI are internalized as higher worker output rather than increased leisure.

    “However, we caution that this assumes all of the gains are internalized as higher worker output rather than increased leisure.”

  • The authors state this as an assumption.

    The economy's aggregate output is modeled using a Cobb-Douglas aggregate production function.

    “We model the economy's aggregate output using a Cobb-Douglas aggregate production function where labor supply is perfectly substitutable across workers.”

  • The authors state this as an assumption.

    In a competitive labor market, each worker is paid their marginal product of labor.

    “In a competitive labor market, each worker is paid their marginal product of labor:”
16 of 16

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • RPS outside-of-work adoption is reported as both 35.5% and 33.7%

    Minor

    The RPS results give two different outside-of-work adoption rates. The 33.7% figure also matches the components shown for that bar in Figure 2a (6.6% + 21.9% + 5.2%).

  • Women's early PC adoption advantage does not match Figure 5c

    Minor

    Figure 5c gives PC adoption rates of 29.6% for women and 21.6% for men, a difference of 8.0 percentage points rather than 6.

    Reported 6 · recomputed 8

  • Figure 11b is described as both an industry and occupation comparison

    Minor

    The text says both panels 11b and 11c plot results by industry, whereas the figure notes specify that panel 11b is by occupation and panel 11c is by industry.

  • Figure 3 describes overall adoption as work adoption

    Minor

    Figure 3's note calls its series usage rates at work, but the discussion identifies the genAI point as the overall adoption rate, and the paper says its internet data do not distinguish work from non-work use.

4 of 4

Citations

Cites

Flagged first, then by citations

Research works printed but not linked: 1

  1. Nelson, R. R. and E. S. Phelps (1966). "Investment in humans, technological diffusion, and economic growth". In: The American economic review 56.1/2, pp. 69-75.

Cited by

Sorted by citations

Related

Metadata

Authors
Bick, A., Blandin, A., & Deming, D.
Affiliations
  • Federal Reserve Bank of St. Louis (US)
  • Vanderbilt University (US)
  • National Bureau of Economic Research (US)
ORCID
Bick, Alexander
Venue
NBER Working Paper
Publisher
National Bureau of Economic Research
Published
2024-09
DOI
10.3386/w32966
OpenAlex
W4402736895
Cited by
200
References
2
Citations by year
2026: 130 · 2025: 67 · 2024: 3
Type
report
Language
en
Open access
Locations
National Bureau of Economic Research (publishedVersion)
Topics
  • Social Sciences › Economics, Econometrics and Finance › Economics and Econometrics › Labor market dynamics and wage inequality
  • Physical Sciences › Computer Science › Artificial Intelligence › Artificial Intelligence Applications
  • Social Sciences › Economics, Econometrics and Finance › Economics and Econometrics › Economic Growth and Productivity
Keywords
generative AI, technology adoption, productivity improvement, time savings

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-11-08, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-11-08 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review