AI Is Quietly Erasing the Premium You Earn for Being Good at Your Job

Your degree, your years of experience, your hard-won reputation — a major new study says they're worth 7.8% less than they were before ChatGPT. That's not a forecast or a survey opinion. It's what 2.26 million real freelance contracts reveal once generative AI entered the picture. And the pattern points to something many workers have quietly feared: AI may not just automate tasks, it may be quietly erasing the premium you earn for being good at your job.

A 2026 working paper titled Human Capital, AI, and Labor Commoditization tracked 49,610 workers on Upwork across 21 quarters, from January 2021 through March 2026 — straddling the launch of ChatGPT. The researchers wanted to answer a deceptively simple question: when AI tools can do part of your work, does the market still pay extra for skill, credentials, and track record? The data says less and less. Here is what that means for your career.

The signal that pays your bills is fading

In normal labor markets, employers and clients can't directly see how good you are, so they rely on signals: your education, your experience, your reviews and ratings. Economists call this "human capital signaling," and it's the reason a senior professional commands a higher rate than a newcomer.

The study found that in job categories highly exposed to AI, the importance of these human capital signals dropped by 7.8% after generative AI became widely available. In plain terms, the things that used to set you apart started mattering less to the people hiring you. [Fact] When a client believes an AI tool can close the gap between a beginner and a veteran, the veteran's edge shrinks — and so does the price that edge can command.

This is the mechanism behind what the authors call "commoditization." A commodity is something where one unit is interchangeable with another, so buyers shop almost entirely on price. The worry is that AI is nudging skilled knowledge work toward that condition.

What makes this finding hard to wave away is the data behind it. This wasn't a survey of how workers feel or a forecast built on assumptions. The researchers observed 2.26 million actual contracts — real money changing hands for real work — across more than five years on a single large platform. Because the window straddles the launch of ChatGPT in late 2022, they could compare the same categories of work before and after generative AI became a tool anyone could reach for. That before-and-after structure is what lets them argue the shift is about AI rather than some unrelated market trend. [Fact]

Price is winning, and cheaper workers are taking the contracts

If credentials matter less, what matters more? Price. The study found the weight clients placed on price as a deciding factor rose from 1.1% to 1.8% in AI-exposed categories — a small absolute number, but a notable shift in how hiring decisions get made.

You can see the consequence in where the work actually went. Demand moved toward lower-cost workers, whose share of contracts jumped from 3.2% to 7.9%. That's more than a doubling. Meanwhile, the gap in demand between high-human-capital and low-human-capital workers narrowed sharply, from 10.3% down to 6.2%. The premium for being the more qualified, more experienced option got compressed.

Overall, workers in AI-exposed categories saw their contract volume fall by roughly 7% compared with workers in categories AI couldn't easily touch. [Fact] The work didn't vanish — but it redistributed, and it redistributed away from the people who had invested most in their expertise.

Not every job is equally exposed

Here's the part that should reframe the panic. AI exposure is wildly uneven across occupations, and the study measured it across 102 subcategories grouped into 12 categories. The average exposure score sat at 0.252 on a scale from 0 to 1.

At the high end, Legal Translation scored 0.80 — a field where large language models are genuinely strong at the core task. Accounting came in around 0.20, exposed but far from dominated. At the low end, Photography registered just 0.02, almost untouched, because the work depends on physical presence, taste, and human moments that a text model simply cannot produce.

The lesson is not "AI is coming for everyone equally." It's that the more your value lives in language-based, easily-specified, deliverable-as-text output, the more pressure you'll feel on your rates. The more your value lives in physical craft, judgment in messy real-world settings, trust, and relationships, the more your human capital still commands a premium. [Estimate]

What this means for your job

If you're working in a high-exposure field, the strategy is not to compete with AI on the commoditized part of the job — you'll lose on price. The strategy is to move your value to where the commodity logic breaks down.

That means leaning into the things this study shows AI is weakest at displacing: accountability for outcomes, relationship and trust, judgment under ambiguity, and the ability to integrate AI output into something a client couldn't assemble themselves. A legal translator who becomes the person who guarantees a translation is legally airtight in a specific jurisdiction is selling something a model can't. An accountant who becomes a strategic advisor rather than a data processor is doing the same.

It also means being honest about pricing. If the market is shifting toward price for the routine layer of your work, the answer is to stop selling the routine layer as your headline product. Bundle the AI-assisted output with the human judgment that makes it trustworthy, and charge for the judgment.

There's a subtler move available too. The study shows demand flowing toward lower-cost workers, which tells you that many clients now treat the first draft of a deliverable as cheap and disposable. If your reputation has been built on producing that first draft, the ground is shifting under you. But the same dynamic creates a new, higher-value role: the person who takes AI-generated material and is accountable for whether it's actually right, compliant, on-brand, and fit for purpose. That role barely existed five years ago. It is being created by exactly the commoditization this paper documents, and it rewards judgment that no model can underwrite on its own. [Estimate]

For workers in low-exposure fields like photography, skilled trades, and hands-on services, this research is quietly reassuring: the very things that made your work hard to scale also make it hard to commoditize.

The bottom line

The promise of human capital — that investing in skills, education, and experience pays off — isn't dead. But in AI-exposed corners of the labor market, the return on that investment is being squeezed, and the data now shows it in hard numbers rather than speculation. The workers who thrive will be the ones who relocate their expertise from what AI can replicate to what it provably cannot.

That's not a reason for despair. It's a map. The same study that measures the squeeze also shows exactly where the squeeze doesn't reach — and that's where your next move should be.

Sources

• Human Capital, AI, and Labor Commoditization (2026), arXiv: https://arxiv.org/abs/2606.21880

AI-Assisted Analysis Disclosure

This article was produced with AI assistance. An AI model synthesized findings from the cited research and drafted the analysis, which reflects the source data rather than the opinions of any individual. All figures are drawn directly from the cited working paper. Readers should consult the primary source for full methodology.