The Degree in Question: What a College Credential Is Worth in the AI Era

A question that has two correct answers

Ask whether a college degree is still worth it and the answer is yes, on average, more than at almost any point in the data, and also no, for a meaningful and growing minority of the people who buy one. Both statements are measured, not rhetorical. The gap between them is the story, and it is where generative AI is doing its work.

The averaged case for the degree has not weakened. The college wage premium, what a bachelor's-or-higher worker earns over a high-school-only worker of similar age, sits near 80 percent as of 2024, and it has widened rather than narrowed over two decades: earnings for college-educated workers rose 6.3 percent from 2004 to 2024 while earnings for high-school-only workers rose 3.2 percent over the same period, according to Census Bureau research summarized by the Association of Public and Land-grant Universities. The Bureau of Labor Statistics puts the bachelor's median at roughly 66 percent above a high-school-only worker on its own measure, and the premium is not flat across a career: Census data show it around 69 percent for workers aged 22 to 27 and closer to 120 percent for workers aged 45 to 54, meaning the credential compounds rather than merely opens a door. Georgetown's Center on Education and the Workforce, in its canonical "College Payoff" analysis, puts median lifetime earnings for a bachelor's degree holder near $2.8 million against $1.6 million for a high-school-only worker, a gap on the order of $1.2 million across a working life.

None of that is in dispute, and none of it is new. What is new, and what this study is built around, is that the same years produced a second, contradictory body of evidence: a large minority of individual programs losing money for their own graduates, a historic collapse in public confidence in the institution that grants the credential, an enrollment pattern actively shifting away from the four-year track, and a labor market for the newest graduates that has quietly inverted its usual relationship with the overall economy. Averages hide distributions, and the distribution underneath the college wage premium has never been more unequal.

The degree still pays, on average, better than at almost any point on record. A meaningful share of degrees are quietly losing money for the people who hold them. Both are true in the same year.

Where the degree loses money

The clearest evidence that the average is masking a split comes from FREOPP's "Does College Pay Off?" analysis, which builds a return-on-investment estimate for individual degree programs rather than for the category "college" as a whole, netting tuition, financial aid, and forgone earnings against the additional lifetime income a specific program produces. On that program-level accounting, 28 percent of bachelor's degree programs carry a negative financial return, and about 31 percent of enrolled students are sitting in a program with a zero or negative return. The picture worsens, not improves, at the graduate level: nearly half of master's degree programs show a zero or negative return once the additional years of tuition and forgone earnings are counted. And the loss is not confined to students who can absorb it. FREOPP estimates roughly 29 percent of federal financial aid dollars are flowing into programs with a non-positive return, meaning public policy is subsidizing a share of the credential that does not pay for the people receiving it.

The pattern by field is stark enough that it functions as a map of where AI-era demand is concentrating rather than dispersing. Engineering, computer science, nursing, and economics degrees each add $500,000 or more in net lifetime value on FREOPP's accounting, a premium built on licensure, quantitative rigor, or acute, undersupplied demand. At the other end, degrees in art, music, philosophy, and psychology frequently leave graduates financially worse off than if they had entered the workforce directly after high school, once the cost of the credential and the years of forgone income are counted against the earnings the degree actually produces in those fields.

This is not a verdict on the intrinsic worth of a discipline; a philosophy degree can do things for a mind that a financial-return model cannot price. It is a statement about what the labor market currently pays for, and it maps closely onto the axis this study series treats as the organizing fact of the moment: generative AI is most exposed to cognitive, language-based, credentialed work, and the fields whose graduates earn the strongest premium are, not coincidentally, the fields where a degree still proxies a hard, specific, verifiable skill rather than a general credential.

A record collapse in confidence

The financial split has a public-opinion mirror, and it moved faster than the ROI data itself. Gallup has tracked confidence in higher education since 2015, when 57 percent of Americans expressed a great deal or quite a lot of confidence in the institution. That figure fell to 36 percent by 2024, a collapse of more than a third in under a decade, before recovering somewhat to 42 percent in 2025. A separate and longer Gallup series asks Americans how important a college degree is to getting a good job: 75 percent called it very important in 2010; by 2025 that share had fallen to 35 percent, a record low on a measure Gallup has run for over a decade. Pew Research finds the same direction from a different angle: the share of Americans who say the higher-education system is going in the wrong direction rose from 56 percent in 2020 to 70 percent in 2025.

The reasons people give for their skepticism are informative, because they track the ROI data rather than contradicting it. Asked why they doubt the value of a degree, Americans cite three things in roughly this order: political bias or a perceived agenda in the institution (38 percent), a lack of job-relevant skills taught in the program (32 percent), and cost or student debt (24 percent). The second reason, that college does not teach what a job now requires, is the one this study's thesis predicts. If AI is repricing the generic version of cognitive credentialing and rewarding the specific, verifiable skill underneath it, then "the degree does not teach what employers actually need" is precisely the complaint a rational public would begin to voice, whether or not most respondents frame it in those terms.

The counter to reading this purely as AI's doing is that the confidence collapse predates the current wave of generative AI by years and has multiple independent causes: rising tuition and student debt, well-publicized cases of political controversy on campus, and a broader decline in trust across American institutions generally, not just higher education. The confidence data is real and severe, but attributing all of it to AI would overstate a single cause among several genuine ones.

The share of Americans calling a college degree "very important" fell from 75 percent to 35 percent in fifteen years, a record low on a measure Gallup has run for over a decade.

Enrollment is voting with its feet

If the wage premium says the degree still pays and the confidence data says the public increasingly doubts it, enrollment behavior sits between the two, and it is beginning to show a genuine shift rather than a mood. After several years of decline, total undergraduate enrollment grew 1.2 percent in fall 2025, according to the National Student Clearinghouse Research Center's preliminary report, an apparent recovery. But the composition of that growth tells a more specific story than the headline number. Bachelor's degree enrollment, the traditional four-year track this study is centrally about, grew only 0.9 percent, the slowest of the segments the Clearinghouse tracks. Community college enrollment grew 3.0 percent, associate degree enrollment grew 2.2 percent, and certificate program enrollment grew 1.9 percent and has now grown 28.3 percent since fall 2021, the fastest-growing credential type in the American postsecondary system over that window.

Certificates are the segment most worth watching, because they are the closest thing the current system has to an unbundled, skill-specific credential, the kind this study's thesis predicts should gain ground as the labor market reprices skills over degrees. A certificate is typically shorter, cheaper, and built around a named, verifiable competency, closer to what employers describe wanting when they cite a lack of job-relevant skills as their reason for doubting the four-year degree.

The other side of the enrollment picture is a stock, not a flow, and it is large: the National Student Clearinghouse estimates roughly 37.6 million working-age American adults have some college experience but no completed credential at all, a population that paid some or all of the cost of college and captured none of the wage premium the degree confers. That population is itself evidence for the two-sided reading. It did not result from AI, and it predates the current wave of the technology by decades. But it is the clearest illustration in the data of what happens when the credential is purchased and not completed: the cost is real and the wage premium, which attaches to the finished degree, is not.

The strange market recent graduates are entering

The freshest and most AI-relevant piece of evidence concerns not whether a degree pays over a career, but what happens to its holder in the first few years after graduation, and here the labor market has done something unusual. The Federal Reserve Bank of New York's college labor market data show recent-graduate underemployment, holding a job that typically does not require a bachelor's degree, near 42 percent in 2025, the highest level since 2020. More striking is a reversal in the unemployment relationship: the unemployment rate for recent graduates aged 22 to 27 has risen to about 4.8 percent, running above the overall US unemployment rate of roughly 4.0 percent, an inversion that does not fit the degree's historical pattern, in which new graduates have almost always fared at least as well as the general workforce, if not better, in their transition years.

The complication, and the one that keeps this from reading as simple decline, is that recent graduates still meaningfully outperform their non-graduate peers of the same age: the New York Fed data put recent-graduate unemployment at 5.6 percent against 7.8 percent for same-age workers without a degree, a gap of more than two points that has not closed. The degree, in other words, is still the better bet than not having one at the same age and stage. What has changed is the comparison to the overall labor market rather than to the alternative of skipping college, and that is a narrower but real deterioration, concentrated in exactly the entry-level, routine-cognitive roles, junior analyst work, first-tier support, early-career coordination and documentation, that this series' companion study on the entry-level job market treats as the most AI-exposed layer of the white-collar ladder.

A methodological note belongs here rather than buried in an appendix. The New York Fed's primary college labor market page returned a blocked response during this research and the figures above were confirmed through the Federal Reserve's own blog commentary and secondary financial press reporting rather than the underlying dataset directly. The direction and rough magnitude are corroborated across multiple outlets citing the same Fed release, which supports treating the figures as established, but the study flags the secondary path here in the interest of full disclosure.

For the first time in the New York Fed's recent series, unemployment among 22-to-27-year-old college graduates has run above the overall US rate, even as those same graduates continue to beat their non-graduate peers.

What AI is actually paying for

If the degree's average value is holding and its distribution is splitting, the sharpest new evidence is what the labor market is willing to pay for on top of a credential, and it points at skills rather than degrees. PwC's Global AI Jobs Barometer, built from more than one billion job advertisements across 27 countries, found the wage premium associated with documented AI skills grew from 25 percent in 2024 to 56 percent in 2025 to 62 percent in 2026, a near-tripling in two years that is among the fastest wage-premium moves recorded for any skill category in the barometer's history. Lightcast, working from 1.3 billion job postings, finds a comparable pattern from a different data source: postings requiring an AI skill pay roughly 28 percent more on average, about $18,000 a year, than comparable postings without one, and postings requiring two or more AI skills carry a 43 percent premium, evidence that the effect compounds rather than saturates at a single named skill.

The more consequential Lightcast finding for this study's thesis is not the premium itself but a shift in what a job requires to be filled at all. The share of AI-augmented roles that specify a bachelor's degree as a requirement fell from 66 percent in 2019 to 59 percent in 2024, a five-year move of roughly seven percentage points in the direction of dropping the degree requirement even as those same roles demand named, specific AI competency instead. That is the unbundling this study's thesis describes in miniature: employers are not abandoning the requirement for skill, capability, or education broadly, they are increasingly specifying the skill directly rather than trusting a bachelor's degree to proxy it, a trend independently corroborated by the broader "skills over degrees" hiring shift this series covers in its companion study.

The tension in this section is that a documented AI skill and a college degree are not competing goods in most of the postings driving these numbers; they frequently sit together on the same job requisition, and the strongest-paying postings tend to specify both. The Lightcast and PwC data show employers rewarding a specific, verifiable competency more than they used to reward the credential alone, not that the credential has become worthless. Read against the FREOPP field data from earlier in this study, the pattern converges: engineering, computer science, and quantitative fields, the disciplines whose degrees already function as a specific, verifiable proxy for a named skill, are the same disciplines producing the strongest financial return and are best positioned to absorb the AI-skill wage premium on top of the degree premium, while a generic bachelor's degree in a field with a looser connection to a demonstrable competency is less able to capture either.

India: quantity climbing, employability lagging

The American data describes a mature system splitting along the lines of specificity and demonstrable skill. India's higher-education system is at an earlier stage of a related but distinct problem: rapid quantity growth colliding with a stubborn employability gap that predates the current AI wave and complicates any single-cause reading of it. India's gross enrollment ratio in higher education sits near 28 to 30 percent according to the All India Survey on Higher Education for 2023-24, with total higher-education enrollment around 4.3 crore, roughly 43 million students, a scale expansion that has continued for over a decade.

Set against that growth, the India Skills Report 2025 puts national graduate employability, the share of graduates judged by employers to be job-ready at the point of graduation, at roughly 54.8 percent, meaning close to half of India's graduating cohort is assessed as not immediately employable in the field they trained for. The State of Working India 2026 report finds a related and sharper figure at the youngest end of the labor force: roughly 40 percent of graduates under 25 who are in the labour force are unemployed, a concentration of joblessness specifically among the young and credentialed rather than spread evenly across the workforce.

This study tiers the India figures emerging rather than established, both because the India Skills Report and the State of Working India series are secondary, survey-based measures with their own methodology-dependent definitions of "employability" and "unemployed," and because the underlying reports were reachable during this research primarily through public summaries of the survey findings rather than the full primary methodology in every case. The direction the figures describe, quantity of credentials climbing faster than demonstrated employability, is the same dynamic the FREOPP and Lightcast data show for the United States in a different form, and it strengthens rather than weakens this study's central claim: the credential and the skill it is meant to proxy are drifting apart in more than one higher-education system at once, under more than one set of institutional conditions.

The limits of this reading

Several cautions bound the argument this study makes. The most important is timing: the college wage premium data, the Georgetown lifetime-earnings figure, and the FREOPP program-level ROI analysis are all measured against graduates who mostly completed and entered the labor market before generative AI reached its current scale of workplace adoption. None of these figures directly measure an AI effect; they measure a state of the labor market that AI is now acting on, and attributing the ROI split or the confidence collapse primarily to AI, rather than to tuition inflation, changing labor demand, or broader economic conditions that predate the current AI wave by years, would overstate what the data shows. This study treats AI as one force acting on an already-splitting system, not as the sole or even necessarily the primary cause of the split.

The New York Fed figures on recent-graduate unemployment and underemployment were confirmed through secondary reporting after the primary Fed page returned a blocked response, a limitation disclosed above and reflected in the study's tiering. The India employability figures are similarly secondary and survey-methodology-dependent, and are tiered emerging rather than established for that reason. The Lightcast and PwC AI-skills premium figures are drawn from job-posting data, which measures what employers say they want in a requisition rather than directly measured outcomes such as realized wages or completed hires, a gap worth naming: posting-language premiums and paid-wage premiums are correlated but not identical measures.

What survives this scrutiny is narrow and well-founded. The average value of a bachelor's degree, measured as a lifetime wage premium, has not fallen and by some measures has risen. A meaningful minority of individual degree programs, concentrated in fields with a weaker connection to a specific, verifiable, in-demand skill, run a negative financial return for the students who complete them, a fact obscured by every average that treats "college" as one product. Public confidence in the institution has fallen to some of its lowest recorded levels, for reasons that include but are not limited to AI. And the market is increasingly willing to pay a rising, compounding premium for a documented AI skill, on top of or in some cases instead of the credential that used to be the only proxy available for that same competency. The degree is not disappearing. The generic version of it is losing ground to the specific skill it was always meant to represent, and that migration is the fact worth tracking, not the aggregate wage premium alone.