Upwork AI Budgets Run 52% Higher as Hourly Rates Barely Move

If you buy freelance or contractor capacity, the OpenAI and Upwork study published this week says the money in AI work sits in project scope, while hourly rates have barely moved. AI-related fixed-price postings on Upwork carry budgets about 52% higher than comparable non-AI work after controlling for subcategory and month, the posted hourly rate gap shrinks to roughly 1.5% under the same controls, and hired hourly contracts log about 41% more hours in their first 90 days. Clients are paying for bigger bundles of work at roughly the same hourly price. Plan contingent spend by the bundle of tasks you need delivered, and treat a single-skill rate card as the wrong unit.

The demand mix is moving too. Between the first half of 2025 and the first half of 2026, web development lost 1.98 percentage points of Upwork's posting share, the largest drop of any category, while mobile development gained 2.27 points and video and animation gained 1.38, per Exhibit 2.1 of the report. Read all of it as directional. Both authors sell something here: OpenAI sells ChatGPT and Upwork sells freelance work.

What the OpenAI and Upwork Study Measured

The study compares what people do inside ChatGPT with what clients post on Upwork, and it measures postings and posted prices, not delivered outcomes. "The Rebundling of Freelance Work in the AI Age" has six authors, three from OpenAI and three from Upwork, and is the latest report in OpenAI Economic Research's Work at the Frontier series; Let's Data Science dates its publication to November 10, 2026. The first report in the series, from July 2026, looked only at ChatGPT messages. We covered it as AI crossing job boundaries.

The inputs, per the methodology appendix of the PDF:

• 36,896 sampled consumer ChatGPT conversations from January 2025 to June 2026, classified by an OpenAI model (GPT 5.6 Luna) into Upwork's task taxonomy.
• Upwork marketplace postings, with "AI-related" assigned by Upwork's own internal LLM classifier, and more than 3,000 activity descriptions grouped into 400-plus standardized tasks.
• Posted fixed-price budgets and posted hourly-rate midpoints through April 2026, with the top 1% of prices trimmed.

A task bundle, in the report's terms, is the set of distinct tasks a single job posting asks for. The headline finding is that AI-related postings bundle more of them, and bundle combinations that were rarely bought together before.

Which Freelance Categories Are Losing Share

Writing, web and data-handling work are losing share of Upwork postings, and visual production, mobile, QA and some legal work are gaining. Here is the H1 2025 to H1 2026 change in share of all Upwork postings from Exhibit 2.1 of the report:

Category	Change in Upwork posting share	Share of ChatGPT conversations
Mobile Development	+2.27 pp	0.18%
Video & Animation	+1.38 pp	1.02%
QA Testing	+0.58 pp	0.08%
Corporate & Contract Law	+0.49 pp	1.29%
Data Entry & Transcription	-0.30 pp	0.37%
Content Writing	-0.57 pp	14.45%
Web & Mobile Design	-0.68 pp	0.15%
Web Development	-1.98 pp	0.46%

Content writing is the cleanest case. It accounts for 14.45% of sampled ChatGPT conversations against 1.27% of Upwork postings, and over roughly two years starting in H1 2023 the report says content-writing postings fell at more than twice the rate of Upwork postings overall. Video and animation went the other way, growing about a third relative to the marketplace.

The pattern is not uniform, and the authors say so. Of 63 matched subcategories, 43 (68.3%) fit the simple story where activities common in ChatGPT lost Upwork share and uncommon ones gained it. Six heavy-ChatGPT categories gained share anyway, and 14 light-ChatGPT categories lost it. Web development fits that second pattern: hardly anyone does it in ChatGPT, and it still had the biggest share decline from H1 2025 to H1 2026.

One caution that changes how you read the table. These are shares of a marketplace that is shrinking. Upwork reported third-quarter 2026 gross services volume of $914.1 million, down 10.1% year over year. A category that gained share may still have fewer postings in absolute terms than a year ago.

---

Why AI Projects Post Bigger Budgets

AI-related projects post bigger budgets because they bundle more work, and the hours data supports that reading more than the rate data does. Exhibit 2.4 of the report gives three numbers that belong together:

• Fixed-price budgets are 70.3% higher raw and 52.4% higher after controlling for subcategory and posting month.
• Posted hourly rates are 9.2% higher raw and 1.5% higher adjusted.
• Hours worked in the first 90 days of hired hourly contracts are 20.1% higher raw and 40.6% higher adjusted.

The authors' interpretation is that AI work comes with more labor per contract and larger projects, "rather than primarily with a higher price for each hour of worker time." AI-related postings average 3.35 distinct tasks against 3.19 for non-AI postings, and the effective number of task clusters in AI-related work rose from 14.6 in early 2023 to 21.2 in late 2025, about 45% more spread.

The bundles are also unusual. Task pairs that appeared in fewer than 200 pre-AI postings make up 8.4% of pairs inside AI-related jobs versus 2.4% inside other jobs. In generative AI and creative production, pairings had roughly 60% lower historical prevalence than comparable non-AI jobs. The example the authors use is a single posting asking for AI-generated video, editing and motion graphics, tasks clients rarely bought together before. If your marketing team is buying output from tools like Runway or Synthesia, that is the shape of what you are now buying.

Inside AI-related postings, the mix moved toward finished output. Between 2023 and 2025, social media campaign management gained 1.68 points of task share, motion graphics 1.50 and AI-generated video 1.46. AI-generated written content fell 1.07 points and designing training programs fell 1.54. Building chatbots and ML models stayed the largest cluster but dropped from 21.8% to 17.8%.

What the Study Cannot Tell You

The study cannot tell you what AI work actually costs you, because it measures posted prices and has a commercial author on both sides. Let's Data Science notes the data is limited to Upwork and ChatGPT, so it suggests a change in project scope and does not show whether AI replaces freelancers. The authors write that the two datasets "are not directly matched, and neither represents the labor market as a whole."

Four limits matter for a budget decision:

1. A posted budget is what the client typed, before negotiation, scope changes or cancellation. It is not a paid invoice.
2. The 1.5% rate gap and the 41% hours gap come from different populations. The hours figure covers only hired hourly contracts with recorded time, so do not multiply the two into a cost estimate.
3. "AI-related" is a label from Upwork's own classifier. Postings it tags may differ from the non-AI set in ways the subcategory controls miss.
4. The adjusted figures come from regressions with month and subcategory fixed effects. They describe an association, and the report does not claim a causal effect.

Earlier independent work on the same marketplace was less comfortable for freelancers. Washington University's Olin Business School summarized research by Hui, Reshef and Zhou finding that after ChatGPT's release, writing freelancers on Upwork saw monthly jobs fall 2% and monthly earnings fall 5.2%, with image workers down 3.7% in jobs and 9.4% in income after image generators shipped. The new report cites that paper and does not rule out displacement in specific segments.

Upwork's own survey data also complicates the budget story. Its Future Workforce Index 2026, reported by Staffing Industry Analysts, found generative AI and creative contract starts up 90% year over year while per-contract earnings for generative and creative freelancers fell 13%, while per-contract earnings for freelancers on more complex work rose 45%. Demand for AI work can grow while the price of the simpler AI tasks falls, so a blanket AI premium in your budget would overpay for the easy work.

---

What Upwork's Own Numbers Say

Upwork's financials show the same rotation the study describes, with the low end of the market shrinking faster than the AI end grows. In August, PYMNTS reported that Upwork had identified about 10% of its GSV as more exposed to AI automation and that low-complexity jobs were being automated faster than expected; it cut full-year revenue guidance to $730 million to $750 million. CEO Hayden Brown said then that "the constraint is not access to AI technology, but access to people who understand how to apply it."

By the third quarter, GSV from talent with AI skills on their profiles grew 55% to $112 million, according to the earnings call, an annualized run rate of nearly $450 million per the company's results release. Upwork changed the metric that quarter. The old one counted postings where clients explicitly asked for AI; the new one counts freelancers who list AI skills, because Brown said AI needs have become "much more de facto." The two figures are not comparable, and a count based on profile skills will rise as more freelancers add AI to their profiles. On the same call the company's CFO said headwinds to overall growth, including changes in search traffic and low-value work rotating off the platform, would continue into the first half of 2027.

This matches research outside freelance markets. Stanford Digital Economy Lab work across 41 countries found that when firms adopt AI, the junior share of the workforce falls mainly because senior employment grows, with junior employment largely holding. We saw the in-house version in Workday's skills data, where basic AI skill demand fell as demand for building with AI rose.

How to Re-Tier Your Contingent Labor Spend

Re-tier contingent spend around the deliverable you need, then price it as a fixed bundle with acceptance criteria. Single-skill hourly requisitions sit in the categories where the study finds postings thinning, often the same work people now do directly in ChatGPT.

This Week:

1. Pull the last 12 months of freelance and SOW spend from your vendor management system or marketplace accounts and tag each engagement by category. Flag spend in content writing, web development, data entry and basic design, the categories losing share.
2. For each flagged engagement, ask the budget owner one question: was this work that an employee with an AI assistant could now produce, or a finished deliverable that needs a specialist? Write the answer next to the line item.

This Month:

1. Rewrite your two largest recurring freelance requisitions as bundles. A "blog writer at $X an hour" line becomes "a monthly campaign package: copy, short video, editing and scheduling, fixed fee, with named acceptance criteria." This mirrors the 3.35-task bundles the study found in AI postings.
2. Budget AI-integration and AI-video work at a fixed-price premium, and require a task breakdown with every quote so you can see whether a higher number reflects more scope or a higher rate. The study's data says it should be mostly scope.
3. Ask procurement to stop approving AI-related hourly contracts without an hours cap. The 41% hours gap suggests open-ended hourly AI work runs longer than the requisition assumed.

Before Your 2027 Budget Locks:

1. Move savings from content and web line items into a smaller pool for integration, QA and production work, and have one owner per bundle. Your org design needs someone who can scope a multi-skill project, or the bigger bundles land on nobody.
2. Treat any vendor's "AI skills premium" figure, including this one, as a negotiating anchor for the vendor. Compare it with your own invoice history before you accept a rate card built on it.

The Bigger Picture

The 2023 Upwork research and this report read as two stages of one shift. In the first months after ChatGPT, writing and image freelancers lost jobs and earnings. Three years on, the demand that remains is concentrating in projects that combine several skills, and posted budgets follow the scope. AI lowers the cost of individual tasks, while the cost of stitching them together stays with whoever owns the project. Contingent-labor programs built on single-skill hourly requisitions were designed for the earlier market. The same pressure shows up inside engineering, where AI coding agents moved the bottleneck to code review.

The evidence is early and written by two interested parties, and the price data ends in April 2026. Pull your own spend by category this week and check whether your web and content line items are falling the way Upwork's are.

Continue Reading

• AI Crossed 44% of Job Boundaries. Is Yours Next?
• Workday Finds Basic AI Demand Fell as Building Skills Rose 51%
• AI Coding Agents Lift Pull Requests 35.7%, Harvard Study Finds
• Canva Reorg Brings Product, Growth and Platform Into One Org
• EXL Acquires Aimpoint Digital; Ask About Staffing in Your Next SOW
• AI Created a 62% Wage Gap Between Two Types of Workers