For Every $1 Cut in Freelance Spending, Companies Spend Just 3 Cents on AI

Something remarkable happened after ChatGPT launched in November 2022. Quietly, steadily, companies began redirecting their money. Not dramatically — not overnight. But dollar by dollar, quarter by quarter, budgets that once went to human contractors started flowing toward AI tools instead.

We have talked about this shift before on this site. Our coverage of the Brookings freelancer study documented the supply side — freelancers losing income, experienced professionals hit harder than newcomers. But a new working paper provides something we have been missing: the demand side. What are companies actually doing with their budgets?

The answer, according to economist Ryan Stevens, is startling.

The 1:33 Ratio That Should Get Your Attention

The paper, titled "Payrolls to Prompts: Firm-Level Evidence on the Substitution of Labor for AI," [Fact] uses data from a major U.S. corporate expense management platform covering Q3 2021 through Q3 2025. By treating the release of ChatGPT as a natural experiment and applying a difference-in-differences methodology, Stevens found that for every $1 a company spends on generative AI, it reduces outsourced labor spending by $33 [Fact].

Let that sink in. A ratio of 1:33.

[Estimate] This implies roughly a 97% cost saving when firms shift from human contractors to AI tools. To put it in personal terms: if a company was paying a freelance writer $3,300 for a month of content work, the AI subscription replacing that output costs about $100.

The study focuses specifically on companies with high exposure to AI-automatable tasks. [Fact] These "high-exposure" firms increased their AI spending by 0.8 percentage points more than low-exposure firms between the ChatGPT launch and Q3 2025. That might sound small, but across corporate budgets worth millions, it adds up fast.

Why This Study Matters More Than You Think

We have seen plenty of surveys asking executives whether they plan to use AI. We have seen freelancer platforms report declining earnings. But this paper does something different: it tracks actual corporate spending [Fact] — not intentions, not self-reported data, but real dollars moving through expense accounts.

[Claim] This makes it arguably the first firm-level empirical evidence that AI is genuinely substituting for human labor at scale, not just supplementing it.

The distinction between supplementing and substituting is critical. When AI supplements human work, workers remain employed but become more productive. When AI substitutes for human work, the work simply goes elsewhere — or disappears. Stevens's data points clearly toward substitution, at least for outsourced and contract labor.

The Two Sides of the Same Coin

Think of it this way. The Brookings freelancer study showed the wound from the perspective of the person being cut: freelancers on platforms like Upwork seeing 5% lower monthly earnings [Fact], with highly experienced professionals losing even more. That study tracked individual workers and their shrinking contracts.

This new research shows the same wound from the other side — the hand holding the knife. Companies are not just passively benefiting from AI; they are actively reallocating budgets away from human contractors. The spending data reveals a deliberate strategic shift.

Together, these two studies paint a complete picture. Supply side: freelancers losing income. Demand side: companies choosing AI over people. The mechanism connecting them is now empirically documented from both ends.

Who Gets Hurt

The occupations most directly in the crosshairs are those where outsourced, project-based work has been common. If your job involves producing discrete deliverables that a client can evaluate — articles, designs, code, edited documents, support responses — you are in the zone of highest impact.

Graphic designers face a market where companies can generate visuals for a fraction of what a human charges. Our data shows graphic design already carries significant AI exposure, and this study explains the budget mechanism behind it. See detailed data

Writers and authors are perhaps the most directly affected group, given that large language models are literally built to produce text. The 1:33 ratio likely hits content-heavy outsourcing hardest. See detailed data

Web developers, particularly those doing front-end and template work, face similar pressure. AI coding assistants can now generate functional pages and components that once required human contractors. See detailed data

Editors who worked on contract reviewing and polishing content are finding that AI both generates and self-edits, reducing the need for human review passes. See detailed data

Customer service representatives, especially those in outsourced call centers and chat support, are watching companies redirect budgets toward AI chatbots that operate around the clock. See detailed data

What This Means for You

If you work in any of these fields — especially as a freelancer or contractor — this research confirms what you may already feel. The shift is not hypothetical. It is happening in corporate budgets right now.

But there is an important nuance. [Claim] The study measures outsourced labor specifically. Full-time employees within companies appear less immediately affected, likely because switching costs, institutional knowledge, and employment regulations create friction that does not exist with contract labor. Freelancers and contractors are, in economic terms, the path of least resistance.

The practical takeaway is straightforward. If your income depends on producing work that AI can approximate at 3% of the cost, competing on price is a losing strategy. The workers who will thrive are those who move toward tasks that AI cannot replicate: building client relationships, exercising judgment in ambiguous situations, and combining domain expertise in ways that require genuine understanding rather than pattern matching.

[Claim] This does not mean these jobs will disappear entirely. But the volume of available contract work — and what companies are willing to pay for it — is being permanently recalibrated.

Sources

• Stevens, R. (2026). "Payrolls to Prompts: Firm-Level Evidence on the Substitution of Labor for AI." arXiv:2602.00139. https://arxiv.org/abs/2602.00139
• Hui, X. & Reshef, O. (2025). "The Short-Term Effects of Generative AI on the Online Freelance Workforce." Brookings Institution.

Update History

• 2026-03-23: Initial publication based on Stevens (2026) arXiv paper.

---

This analysis was generated with AI assistance. All factual claims are tagged with [Fact], opinions and interpretations with [Claim], and projections with [Estimate]. Source data and methodology details can be found in the linked papers. For detailed occupation-level data, visit individual occupation pages.