
Job Displacement and Broader Societal Impacts
Public opinion and political debate
The poll behind the twenty-year claim
Pew Research Center · U.S. adults · June 22–28, 2026; n=3,488
What the public expected
In Pew Research Center’s June 2026 survey, 71% of U.S. adults said AI would lead to fewer U.S. jobs over the next twenty years, compared with 64% in the August 12–18, 2024 observation. The answer concerns the direction of an aggregate change, not its magnitude. In 2026, 5% expected more jobs, 10% expected little difference and 14% were unsure; totals may not sum to 100 because of rounding or unanswered items. Pew fielded the 2026 survey with 3,488 participants from its American Trends Panel, designed to represent the noninstitutionalized U.S. adult population. The full-sample sampling margin was ±1.8 percentage points at the 95% confidence level; it is not the error bound for any age subgroup. [15] [17] [18]
What the poll cannot estimate
It is not an estimate that 71% of jobs will disappear, that 71% of workers will be displaced, or that the probability of national job loss is 71%. A retiree and a working adult are both in the population; neither has to report personal employment exposure to answer. People may also interpret fewer jobs relative to today or relative to the jobs otherwise available. The question does not supply an economic counterfactual. Present the result as a national expectation over the survey horizon. Repeating the same question strengthens descriptive comparison, but cannot show whether exposure to AI or another event changed beliefs. [15] [28]
Exhibit 02 · A policy monitoring matrix
A dashboard can relate early indicators to observed outcomes without pretending they share a scale. Review disaggregated cohorts and compare to credible baselines before concluding that AI caused a change.
48PEGASUS Research conceptual exhibit. Not empirical data.
Young adults changed most, not always highest
Ages 18–29 · 61% to 73% · 2024–2026
Compare the four age bands
Pew’s reported shares expecting fewer U.S. jobs were 61% to 73% for ages 18–29, 66% to 74% for ages 30–49, 69% to 72% for ages 50–64, and 58% to 63% for adults 65 and older, from 2024 to June 2026. The youngest band registered the largest observed increase, 12 percentage points. In the later reading, however, its 73% was close to 74% among ages 30–49 and 72% among ages 50–64. Pew characterizes the youngest group’s current views as similar to the middle age bands. [15]
Interpret the gradient carefully
“Particularly among younger adults” is most defensible as a description of the change in Pew’s trend or the comparison with adults ages 65 and older. It is not a claim that respondents under 30 were uniquely alarmed or clearly more pessimistic than all working-age adults. In June 2026, Pew reported age-band sampling margins of ±4.7, ±3.1, ±3.5 and ±3.9 percentage points, respectively, with unweighted bases of 569, 1,256, 902 and 756. Adjacent rounded numbers do not establish a statistically significant ranking. [18] The age bands in each wave are repeated population slices. They do not isolate a fixed birth cohort or prove that each respondent reversed position. Keep the time comparison and the cross-age comparison distinct in a headline or exhibit. [15] [28]
Exhibit 03 · Expectations of fewer jobs rose
Pew surveyed 3,488 U.S. adults June 22–28, 2026. The youngest group had the largest observed change since 2024 (+12 percentage points); Pew describes their 2026 result as similar to ages 30–64. These are beliefs about the future, not forecasts or counts of jobs lost.
Pew Research Center, June 22–28, 2026 field period: primary source
Daily-life skepticism is a separate measure
U.S. adults · Concern versus excitement · 2021–2026
An attitudinal change beyond jobs
Pew’s daily-life item asks respondents how they feel about growing AI use, not whether total jobs will rise or fall. In June 2026, 52% were more concerned than excited, 37% equally concerned and excited, and 9% more excited than concerned. The comparable more-concerned share was 37% in 2021. Among ages 18–29, more-concerned sentiment increased from 31% in 2021 to 55% in 2026, while more-excited sentiment declined from 25% to 11%. Adults 65 and older still registered a higher 2026 point estimate for general concern (59%). Pew’s 2021 wording included an explanatory preamble about AI computer programs, noted in its topline. [15] [17]
Familiarity and concern can coexist
A younger person can use a chatbot for routine work while doubting that companies will share the gains. An older person may be more uneasy about AI overall while being less likely to predict job destruction. The measures have different outcomes and cannot be collapsed into one index without an explicit method. Pew’s separately fielded February 2026 report measures experience, use and expected impacts; it reinforces the need to distinguish task utility from broad assessments, not a rule that heavy use must increase or reduce concern. Report timing and population before interpreting a difference. [15] [16]
A second survey points in the same direction
Quinnipiac University · Job opportunities · March 2026
A distinct outcome and question
Quinnipiac’s March 2026 poll found that 70% of adults thought advancements in AI were likely to decrease job opportunities, up from 56% in its April 2025 release. In the March survey 7% expected an increase and 18% expected little difference. Its adult Gen Z definition used birth years 1997–2008. Eighty-one percent of that group expected fewer opportunities, followed by Millennials (71%), Gen X (67%), Baby Boomers (66%) and the Silent Generation (57%). Quinnipiac’s question does not specify Pew’s twenty-year horizon. [20]
Agreement, not a merged trend
Both instruments are consistent with widespread pessimism. They are not sequential readings of the same variable: one asks about a direction in future U.S. job count, the other about opportunities without a named horizon. Generations defined by birth year are also different from fixed age brackets across waves. Putting all percentages on the same unlabeled line would imply precision the evidence does not supply. A small-multiple figure is more honest: show changes within each survey, print the two question summaries, and label the different field dates. Treat the direction of corroboration as an observation rather than a pooled estimate. [20] [15] [28]
Changing the horizon changes the claim
Bentley–[*REDACTED*] · U.S. adults · Ten-year expectation
Broad pessimism at a shorter horizon
The July 2026 Bentley–[*REDACTED*] publication reports that 79% of Americans expected AI to reduce the total number of U.S. jobs in the next decade, compared with 73% in 2025 and 75% in 2024. Among adults ages 18–29, 75% expected a reduction in 2026, up from 62% in 2025; adults ages 45–59 recorded 84% in 2026, up from 75%. In the youngest group, the share expecting an increase fell from 14% in 2025 to 5% in 2026. [21]
Do not infer a time preference
The 79% ten-year response cannot be subtracted from Pew’s 71% twenty-year response to estimate how beliefs change as the horizon doubles. They come from separate instruments and samples, with potentially different wording, response options, question placement and field dates. A direct inference about time horizons would require a designed comparison, such as randomly assigning respondents to otherwise identical questions with different horizons. The existing figures support a narrower conclusion: pessimism appears under several formulations. Preserve the source authors’ terminology and avoid an average that erases the question actually asked. [21] [15] [28]
National pessimism and personal concern differ
Quinnipiac University · Adults versus employed adults · Separate denominators
A national expectation is not personal risk
Quinnipiac’s March 2026 survey found 70% of adults expected a decrease in opportunities. Separately, 30% of employed adults were very or somewhat concerned that AI might make their own job obsolete, compared with 21% of employed adults in April 2025. In March, 69% of employed respondents were not very concerned or not concerned at all in the poll’s reported breakdown. In the separate national-outlook item, 71% of white-collar workers and 73% of blue-collar workers expected fewer opportunities overall. [20]
Avoid the fallacy of subtraction
The later national and personal shares should not be subtracted to infer how many particular people are worried about others but exempt themselves. The respondent universes differ and the question objectives differ. Some workers may anticipate slower hiring in other industries, entry-level roles, or distant years while judging their present position secure. Some may infer personal consequences without expecting national contraction. Neither interpretation can be determined without linked response data. For an employer pulse check, ask separately about industry outlook, own role, hours or earnings, and trust in planned redeployment. These measures imply different interventions. [20] [28]
Technology obsolescence as a worker concern
[*REDACTED*] Work and Education · Employed respondents · August 3–24, 2026
A long-running measure
[*REDACTED*] reported that 27% of U.S. workers worried that technology could soon make their own jobs obsolete, up from 20% in 2025 and 13% when measured in 2017. In the August 2026 wave, 34% of workers ages 18–44 expressed this worry versus 19% of workers ages 45 and older. Worry was reported by 25% of college graduates and 29% of workers without a college degree. The item measures technology broadly rather than isolating generative AI. [22]
Put the question alongside ordinary insecurity
[*REDACTED*] reported 33% of workers worried about benefit reductions, 24% about layoffs, and 21% each about reduced hours and lower wages. These are different outcomes, not alternatives that sum to 100. A company that reassures workers only that it has not scheduled layoffs may leave the underlying issue of hours, pay or work intensity unanswered. Conversely, an employed-worker poll does not cover people excluded from hiring, recent graduates looking for their first job, or many people who already left paid employment. Combine workforce perceptions with hiring, separations and scheduling records before drawing a conclusion about realized risk. [22] [28]
Exhibit 04 · Workers worried about technology making jobs obsolete
Separate [*REDACTED*] Work and Education survey, August 3–24, 2026. The measure concerns workers’ own jobs and technology generally, not the Pew question about the national impact of AI over twenty years. Age bands also differ.
[*REDACTED*], 2026 Work and Education survey: primary source
Young adults see a threat to financial security
Harvard Youth Poll · Young Americans ages 18–29 · Sept. 30–Oct. 7; n=2,024
Distinguish seriousness from likelihood
In Harvard’s fall 2026 youth poll, 64% described AI replacing jobs as a serious threat to their long-term financial security. The reported shares were 76% among young Democrats and 61% among both Republicans and independents. In the same poll, 68% regarded the power of large corporations as a serious threat and 62% said so about the wealth and power of billionaires. The study is a survey of young adults. It cannot be used by itself to compare respondents with older Americans or to estimate the causal effect of AI on economic insecurity. [24]
Why this differs from job-count polling
A respondent can believe job totals will grow while treating the risk of losing a foothold in a specific career as serious. Threat seriousness can reflect either a substantial perceived chance or the consequences of a low-probability event. Harvard also reported 49% of young adults described household finances as strained or lacking much security. That points to a larger context: a brief income interruption has different consequences for households with different savings, debt, benefits or family support. The poll does not establish which of those mechanisms explains any individual’s response. [24]
Public concern is not top-issue salience
[*REDACTED*]–[*REDACTED*] Dedicated items versus issue choice · 2026
Prompted and competing priorities answer different questions
In June 2026 [*REDACTED*]–[*REDACTED*] results from 4,531 U.S. adults, 75% were very or somewhat concerned about AI eliminating jobs in some professions, and 53% expressed concern about themselves or someone in their household being put out of work by AI. The prompts vary whose welfare and whether a bounded group of occupations is in view. In a separate October [*REDACTED*]–[*REDACTED*] core political item, only 3% selected AI as the country’s most important problem from the listed categories. These readings cannot be treated as contradictions. [26] [27]
Build a defensible political claim
A person can care deeply about AI displacement when asked directly and still rank cost of living, government conduct or another issue above AI when offered one choice. AI concerns may also be expressed under broader categories such as the economy, rather than under the AI label. The data establish prompted concern and comparative issue prominence, not that AI determined voting choices or any election result. A civic report should pair the poll’s response format with a separate record of concrete political, administrative or civic actions. Those are distinct forms of evidence and do not substitute for one another. [26] [27] [28]
Trust in institutions and trust in tools
Bentley–[*REDACTED*] and Quinnipiac · U.S. adults · 2026
Low trust does not mean zero use
Bentley–[*REDACTED*] reported that 27% trusted businesses at least some to use AI responsibly in 2026, down from 31% the previous year. Among ages 18–29, the corresponding proportion fell from 30% to 20%, while the share with no trust rose from 29% to 41%. Separately, Quinnipiac reported that 76% of U.S. adults said businesses were not doing enough to disclose how they use AI and 74% said government did not do enough to regulate AI use. These are perceptions of governance, not independent tests of compliance or safety. [21] [20]
Improve the object of trust
Separate confidence in a system’s answer from confidence that an employer or agency will deploy it responsibly. A worker may trust a tool with summarization yet object to the same tool determining a promotion. A resident may favor product availability while demanding disclosures on adverse decisions. Policies should make uses intelligible at the decision point: what data are used, which recommendations influence people, where a person reviews, and how an error can be appealed. Pilots should ask workers about those controls directly rather than interpreting adoption as consent, or low adoption as proof of low utility. [21] [20]
What workers say about their own experience
[*REDACTED*] for Groundwork Collaborative · Ages 18–75 · June 11–16, 2026; n=1,533
A defined labor-force population
[*REDACTED*] surveyed a probability sample of Americans ages 18–75 who were employed or marginally attached to the workforce. Roughly two-thirds chose the view that AI would worsen the worker experience by eliminating jobs and increasing pressure rather than improve it by freeing workers from repetition and helping with more challenging work. This was a choice between bundled alternatives; it is not two separate measurements of workers reporting actual discharge or employer-imposed work intensity. [*REDACTED*] reported a full-sample sampling margin of ±2.7 percentage points at the 95% confidence level for qualified respondents. [25]
Why the denominator belongs in the caption
The study does not represent all adults regardless of age or connection to paid work. It also includes a group of potential workers beyond those currently employed. Within similar surveys, distinguish employed-only questions from broader eligible-worker questions and do not apply a full-sample error estimate to a smaller subgroup. For employer evaluation, ask separately whether a tool saves time, whether work targets increased, whether workers can decline its use, and whether it changes the expected number of positions. Unlike a bundled sentiment item, that design produces actionable information about mechanisms. [25] [28]
Job quality may improve or deteriorate without a layoff
American Job Quality Study · Employee reports · January–March 2026
Users report both benefits and greater duties
[*REDACTED*] and its partners reported that 63% of employees who used AI in their jobs said it helped them work faster, 56% found more creative solutions, 47% could spend more time on interesting work, and 46% produced higher-quality work. Thirty-one percent said their employer asked them to take on more responsibilities because of their AI use. The reported effects concern 7,845 U.S. employees using AI at least a few times in the previous year, while the full survey included 15,482 employees. [23]
Measure who actually gained
In the full sample, 28% reported weekly or more frequent use. Rates were 40% for college graduates versus 17% without a college degree, and 37% for managers versus 25% for individual contributors. These comparisons show differentiated access or use, not that AI caused existing job-quality differences. Worker surveys should be paired with accepted work per paid hour, quality and rework, target changes, and total compensation. A shorter average task time can support more capacity, lighter workloads or higher targets; which occurred is more relevant to distribution than a generic speed metric. [23] [36]
Workers distinguish AI from management decisions
[*REDACTED*] Pulse and Brunswick · Commissioned U.S. research · November 2026
How respondents assign responsibility
[*REDACTED*]’s second Pulse wave fielded an online survey of 1,600 U.S. adults, including an employed oversample, from November 4 to 7, 2026. Its published worker findings report that 45% viewed business leaders’ decisions as a greater career risk than AI. Separately, Brunswick’s November publication reports that 92% of its 65 senior leaders surveyed expected new opportunities from AI, compared with 53% of more than 2,000 employees surveyed. These are commissioning organizations’ own descriptive studies, not job-flow estimates. [34] [35]
Read the evidence asymmetrically
Brunswick’s leader panel is small and restricted to senior leaders at firms with at least 250 employees. Avoid presenting the difference as a definitive national executive–employee gap. The broader diagnostic is more useful: management may frame AI as opportunity while workers are trying to learn whether workload, job security or bargaining power will change. An internal decision review can record whose forecast is being used, test it against staff experience, and name the manager with authority to grant a new role, training time or an appeal. Do not interpret an announcement’s rationale as evidence of what occurred. [34] [35] [28]
Classroom adoption and student confidence can diverge
[*REDACTED*], [*REDACTED*] and [*REDACTED*] · Ages 14–29; n=1,572 · February–March 2026
High use is not unqualified optimism
[*REDACTED*] and partners surveyed 1,572 Americans ages 14–29 from February 24 to March 4, 2026, using the probability-based [*REDACTED*] Panel. In that study, 56% of Gen Z K–12 students and 48% of Gen Z adults reported weekly AI use. Across the surveyed Gen Z population, 80% believed using AI was very or somewhat likely to make future learning more difficult. The sample includes minors and adults. Neither the usage estimates nor the expectation of difficulty measure an actual change in student attainment. [29]
Ask the educational question directly
Distinguish completion speed, aided performance and durable independent skill. A classroom rollout should specify which assignments allow AI, whether students must explain their reasoning, and which learning objectives will be measured without assistance. Consider age-appropriate uses, accessibility, data protection and educator judgment as companion criteria. Federal education guidance issued in August 2026 encourages local decision-makers to focus on instructional value, transparency and evidence rather than engagement alone. The right policy comparison is not unchecked adoption versus a blanket ban; it is the incremental learning and risk of particular uses under clear rules. [29] [31]
From sentiment to public action
Administrative guidance · Civil-society priorities · 2026
Observable actions beyond the poll
The U.S. Labor Department published an AI literacy framework in February 2026 as a program-design resource for workforce and education institutions. The Education Department’s August classroom technology guidance emphasized local decisions, evidence, and educator judgment. On November 18, the Harvard Institute of Politics listed a public workforce-transition forum with former Commerce Secretary [*REDACTED*] and former Indiana Governor [*REDACTED*]. [19] The federal releases and forum show that AI has entered policy practice and public deliberation; none constitutes a guarantee of jobs or a universal legal obligation to implement a particular program. The documents serve different audiences and purposes. [30] [31]
A separate channel of civic intervention
In a November 18 plenary update, the U.S. Catholic bishops’ task force chair called for dialogue with policymakers, technology leaders and communities around human dignity. Its contemporaneous investment guidelines name worker-support considerations, including retraining and human oversight in employment decisions. These are statements and stewardship principles from a civic institution, not government rules, measures of all Catholics’ views, or evidence of achieved worker outcomes. Policy analyses should label the action being observed: survey expectation, guidance, investment principle, proposed legislation, enacted rule, appropriated program, or measured result. [32] [33] [30]
Poll uncertainty and counter-examples
A survey claim audit · Dates, base, context · No pooled headline
Check what disagrees before generalizing
In the May 2026 [*REDACTED*]/[*REDACTED*] polling, the national-job question produced a different age ordering from some other surveys: 57% of adults under 30 forecast a decrease, compared with 65% of those aged 30–44, 70% of those aged 45–64, and 66% of those aged 65 and older. The poll drew 2,064 U.S. adults, and the published weighted base for the under-30 cross-tab was 415. The finding is compatible with Pew’s narrower conclusion that the youngest group showed the biggest increase between Pew’s own waves. Weighting and sampling help align a survey sample with its stated population but do not eliminate wording, mode, timing or coverage differences. [37] [15] [28]
Minimum disclosure for reproducibility
List owner and sponsor, target group, geography, field dates, recruitment and mode, weighted and unweighted subgroup bases when available, wording and bins, and uncertainty where reported. A question about all jobs cannot substitute for an item on one industry or a respondent’s own livelihood. Retain “unsure” where it matters; dropping respondents from the denominator can artificially inflate a percentage. Use percentage points for differences between shares. State whether an age contrast, a year-to-year trend, or a comparison of unlike instruments is under discussion. If cross-tab precision is not published, do not draw thin margins as if they were meaningful ranks. [28] [15]
An accountable public-opinion dashboard
Proposed operating tool · Five series, five denominators · Do not calculate one fear score
Preserve the questions as separate series
Track national outlook with the same twenty-year wording over time, personal concern among workers, institutional trust among adults, classroom experience among eligible students or parents, and support for specific actions among whichever public the decision will affect. Each row should carry its own population and field dates. Record sponsor, version, filter, field dates and possible trend breaks. For fixed cohorts, design or acquire a longitudinal sample rather than assuming an under-30 group is unchanged over successive waves. [15] [22] [29] [28]
Treat opinion as an input to decisions
If personal concern rises in a controlled workforce pulse, investigate roles and surfaces creating the worry, publish planned changes and define protected time for training. If parent confidence in a classroom pilot falls, review actual learning and privacy data instead of attributing the result to a generalized dislike of technology. If public expectations of job loss rise while headcount is stable, check vacancies, cohorts, labor hours and earnings before claiming fears are unfounded. The useful output is a decision, owner and date for a retest. It is not a composite score that mixes populations or makes opinion look like labor-market fact. [28] [23]
Browse: White Papers · Reports · Downloads
