The short version: When a small business hires a freelancer for a project that involves AI, the hourly rate barely moves, but the job gets bigger. A new OpenAI and Upwork report, published November 10, found that posted hourly rates on AI-related Upwork jobs run only about 1.5 percent above comparable jobs without AI, while hired hourly contracts log about 41 percent more hours in their first 90 days, and fixed-price budgets run about 52 percent higher. Hiring freelancers for AI projects costs more because you are buying more work, not pricier work. The question to ask a bidder is no longer “what is your rate?” It is “how many hours, and what exactly do I get at the end?”
That is mostly good news, and worth saying plainly. The work clients still pay people for is finished work: the video, the integration, the campaign that actually runs. That work is growing. Below is what the full 24-page report found, the arithmetic it leaves out, three cautions from its own appendix, and how to brief and budget your next freelance job so the extra hours buy something you wanted.
What did the OpenAI and Upwork freelance report find?
The report, The Rebundling of Freelance Work in the AI Age, is the latest in OpenAI’s Work at the Frontier series, written by three OpenAI researchers and three from Upwork. OpenAI’s summary page carries the headlines; the full report, which we read on November 15, carries the numbers a buyer needs. It compares two things: what people do directly in ChatGPT, and what clients post as paid jobs on Upwork’s marketplace. The stakes are not small. Upwork’s 2026 Future Workforce Index, cited on OpenAI’s page, found that 38 percent of skilled US knowledge workers freelance.
Four findings stand out:
- People and freelancers do different work. Content writing makes up 14.5 percent of the sampled ChatGPT conversations and professional and business writing another 10.2 percent, against roughly 1.3 and 0.4 percent of Upwork job posts. Upwork postings lean instead toward graphic and presentation design, video and animation, software testing, mobile and web development, and digital marketing.
- Work that moved into ChatGPT lost freelance share. Of 63 work categories, 43 followed one pattern: categories common in ChatGPT tended to lose share of Upwork postings, and less common ones tended to gain. Content-writing postings fell at more than twice the rate of Upwork postings overall across two years, while video and animation postings grew by about a third relative to the marketplace. The other 20 categories went the opposite way, which is why the authors say this is not a simple substitution story.
- AI jobs bundle tasks in new combinations. AI-related posts list 3.35 distinct tasks on average against 3.19 for other posts. The bigger difference is which tasks travel together: 8.4 percent of the task pairs in AI-related jobs are combinations that appeared in fewer than 200 postings before generative AI, against 2.4 percent in other jobs.
- AI jobs carry bigger budgets, mostly through hours. Fixed-price budgets run about 70 percent higher raw and about 52 percent higher once you compare jobs in the same work category and month. Posted hourly rates run 9 percent higher raw and about 1.5 percent higher on the same comparison. Hours worked in the first 90 days run about 41 percent higher.
“AI-related” here has a specific meaning. A post counts if it asks for AI skills, requires AI tools, or involves improving output an AI produced, as judged by Upwork’s own GPT-based classifier reading the title, description and requested skills.
Why do AI freelance projects cost more if the rate is the same?
This is the most useful part of the report for anyone who hires. Hourly contracts let the authors separate two things a fixed price blurs together: the price of an hour and the number of hours bought. Once you compare like with like, the price of an hour on an AI-related job is almost the same as on any other job in its category. The number of hours is not.
The authors’ reading is that AI-related work means larger projects, not more expensive labor. They offer two explanations. AI may raise the payoff from experimenting and refining, so clients ask to see more alternatives and push harder on polish. And clients may be handing one freelancer a bigger bundle of work to own from start to finish. They compare it to the rebound effect economists know from energy: when something gets cheaper to do, people do more of it.
In plain terms, when a first draft becomes nearly free, nobody stops at one draft. The client who once wanted a single 30-second video now wants the video, three cut-downs for social and a version with captions. The chat assistant that once meant a script now means a script plus the wiring into the booking calendar. None of that is waste if you wanted it. All of it is hours.
What does that look like in dollars?
Here is the arithmetic the report does not do. The rate and hours below are our example, not figures from the report; only the percentages are theirs.
Take a five-person landscaping company that posts an hourly job: build a chat assistant for the website that answers service-area and pricing questions and books estimate visits. The owner expects 30 hours at $45 an hour, which is $1,350.
- Rate: the report’s adjusted rate gap of about 1.5 percent turns $45.00 into about $45.68. That is 68 cents.
- Hours: the adjusted hours gap of about 41 percent turns 30 hours into about 42.3 hours.
- Total: 42.3 hours at $45.68 is about $1,932, or $582 more than the owner planned. That is 43 percent over budget, and roughly 97 percent of the overrun is hours, not rate.
On a fixed-price job the same pattern shows up inside one number. A project that would carry a $1,000 budget without AI would, on the report’s averages, carry about $1,520 as an AI-related job in the same category. These are averages across a large marketplace, not a quote for your job, and a well-scoped project can land well under them. The lesson holds either way: comparing hourly rates between two bidders tells you almost nothing about what the job will cost. Comparing their hour estimates and what they promise to hand over tells you a great deal.
Three cautions from the report’s own appendix
The ChatGPT side is personal accounts only. The methodology appendix says the ChatGPT sample is 36,896 conversations from individual consumer accounts, sampled from January 2025 through June 2026 and classified by an OpenAI model without researchers reading them. Business and Enterprise workspaces are not in it. So “what people do in ChatGPT” here means what individuals do, and it says nothing directly about how companies use ChatGPT’s business plans.
It describes, it does not prove. The authors state that the price comparisons “do not identify a causal effect” and that neither Upwork nor ChatGPT represents the labor market as a whole. Content-writing postings falling while ChatGPT writing use is high is a pattern, not proof that one caused the other.
The money figures have edges. Price data runs through April 2026, the top 1 percent of posted prices is trimmed, and the hours figure covers only hourly contracts that led to a hire with hours recorded. That is a sensible way to keep a few giant contracts from skewing the averages, and it is also why the percentages describe the typical job, not the unusual one.
One check on an earlier number. In July we reported that freelancers who use AI earn 34 percent more per hour, from Upwork’s Future Workforce Index. That does not contradict today’s 1.5 percent. The July figure compared freelancers by whether they use AI. Today’s figure compares job postings by whether the job is AI-related, at the rate the client posted. A skilled freelancer can earn a premium while the average AI-related posting offers about the same rate as any other job in its category.
How should a small business hire a freelancer for AI work now?
1. Do the drafts yourself, and hire for the finish. The data shows where the market already moved: drafting moved into ChatGPT, and freelance postings shifted toward video, design, testing and development. Write your own first-draft website copy or job descriptions with AI, then spend freelance money on the parts that need a specialist’s hands. That is AI giving you time back and pointing your budget at real skill. We saw the same sorting from another angle in the World Bank data on which outsourced work AI absorbs.
2. Brief the outcome, not the task list. AI jobs often combine skills that used to sit with different specialists. The appendix found that 98.6 percent of the unusual pairings join two familiar tasks that were simply rarely asked for together. The report’s own examples of AI-related bundles: AI-generated video with editing and motion graphics, and chatbot building with app integration. So describe the finished thing you want in one sentence, and ask each bidder for one example where they delivered that same combination, not each piece separately.
3. Ask for an hour estimate and a cap. Keep the 41 percent in mind. If a bidder estimates 30 hours, ask what would turn it into 40, and agree that you will be told before the cap is crossed.
4. Put the number of versions in writing. The report’s best explanation for the extra hours is more alternatives and more refinement. Decide up front how many versions you want to see and how many revision rounds are included. Two rounds is a reasonable starting point for most small jobs.
5. Use fixed price when the deliverable is clear. A fixed price will run higher, about 52 percent on the report’s averages, but it puts the scope risk in the quote instead of in your inbox at the end of the month. Hourly suits work you cannot define yet.
Is this good news for small businesses?
Mostly, yes. The freelance market did not collapse into a chat window. The skills that turn an AI draft into something a customer sees, a finished video, a working integration, a running campaign, are the skills clients are paying more hours for. That matches what payroll data showed this fall, that small businesses using AI are hiring more, not less.
There is a quieter benefit too. If one freelancer can now own a bundle that used to need three people and two handoffs, a small business gets one person to call, one deadline and one invoice. The cost is that the bundle takes longer than any one piece used to. Budget for the hours, write down the finish line, and the extra time buys you something you can use.
Frequently Asked Questions
Do freelancers charge more for AI work?
Per hour, barely. The OpenAI and Upwork report published November 10, 2026 found posted hourly rates on AI-related Upwork jobs about 1.5 percent above comparable jobs in the same category and month. What rises is the size of the job: about 41 percent more hours in the first 90 days of hired hourly contracts, and fixed-price budgets about 52 percent higher.
Why do AI freelance projects take more hours?
The report’s authors suggest two reasons. AI makes it cheaper to try alternatives, so clients ask for more versions and more polish, and clients may hand one freelancer a larger bundle of tasks, such as AI video plus editing plus motion graphics, to own from start to finish. They describe the findings as patterns, not proof of cause.
Should a small business still hire a freelance writer?
For drafts, many owners now write their own with AI, and content-writing postings on Upwork fell at more than twice the rate of postings overall across two years. A specialist is still worth hiring where accuracy, voice or expertise matter, and for editing an AI draft into something publishable. Over the same period, video and animation postings gained share.
What should a brief for an AI freelance project include?
Describe the finished deliverable in one sentence, ask for an example of that same combination of work, request an hour estimate with a cap you are told about before it is crossed, and write down how many versions and revision rounds are included. If the deliverable is clearly defined, consider a fixed price so the scope risk sits in the quote.
Have your last few freelance projects come in over the hours you expected, and was the extra work worth it? Tell us in the comments.
