Upwork Is Cutting 24% of Its Staff. Freelancers Doing AI Work Earn 34% More Per Hour.

Upwork is cutting 24% of its staff and says freelancers doing AI work earn 34% more. What the filings show, and what it means for one-person shops.

Kemal Esensoy·Modified on September 21, 2026

Upwork Is Cutting 24% of Its Staff. Freelancers Doing AI Work Earn 34% More Per Hour.
Behind the Business

In May, Upwork told the SEC it would cut about 24% of its workforce. In July, it published research saying freelancers doing AI work earn 34% more per hour. In November, it reported 77,000 fewer clients than a year earlier, and a bigger quarterly profit.

The Upwork layoffs and that 34% get passed around as one story: AI is killing freelancing, unless you learn AI, in which case you win. The filings tell a less comfortable story than either half of that.

My own marketplace history is on Fiverr, and I wrote all of it down last year. Everything below about Upwork comes from its filings and press releases, so it comes from a company with a stock price to defend. I'll point out where that matters.

What the Upwork Layoffs Actually Look Like

The 8-K from May 7 says "approximately 24%" of the total workforce, with $16 million to $23 million in restructuring charges, mostly severance. Plenty of headlines rounded that up to a quarter.

The number reached so far is lower. The latest 10-Q says Upwork had reduced its workforce by about 20% as of September 30, with more reductions due by December 31. So the accurate sentence today is: Upwork is cutting 24% and has cut about 20%.

The stated reason for the Upwork layoffs is "a more efficient operating model" and "profitable growth as the nature of work evolves." That's corporate for: the top line is shrinking and the margin is not allowed to.

Clients Are Leaving. The Quarterly Profit Isn't.

Here is Upwork's third quarter, reported on November 2:

Q3 2025 Q3 2026 Change
Gross services volume $1,016.7M $914.1M -10%
Active clients 794,000 717,000 -10%
GSV per active client $5,036 $5,418 +8%
Marketplace take rate 18.9% 20.0% +1.1 points
Net income $29.3M $31.8M +8%

Gross services volume, GSV, is everything that moved through Upwork. It fell by more than $100 million in a year. Net income went up anyway, helped by operating expenses dropping 15%. (Over nine months net income is still down 11%, so this isn't a victory lap.)

The 10-Q blames the client drop on "lower acquisition of new clients as well as lower retention of existing clients." Fewer people are arriving and fewer are staying. The ones who remain spend more, which the filing pins on a shift "toward high-spending clients" plus growth at Lifted, Upwork's enterprise arm.

Take rate is the row a freelancer should stare at. Upwork now keeps 20 cents of every marketplace dollar, up from 18.9, and it expects that number to keep rising through the rest of 2026. Part of the rise comes from how it charges you: on contracts since May 2025, freelancers pay 0% to 15% "depending on platform-specific supply and demand factors." The older contracts pay a flat 10%.

Fiverr has the same shape, only steeper. In its Q3 results, annual active buyers fell from 3.3 million to 2.4 million, down 25.8%. Marketplace revenue fell 24.7%. Spend per buyer rose 12.9% to $373, and the take rate rose to 28.2%.

So I'd stop saying the platforms are simply losing. They are losing small buyers and charging more on what's left. Fiverr also blames "AI-related disruption to traditional SEO and SEM" and expects "a steeper decline than previously expected in marketplace traffic through 2027." A marketplace that lives on Google traffic has the same problem as your website.

I Watched the Small Jobs Go in 2024

On Fiverr I was getting messages daily until the middle of 2024. Then they thinned out, and the ones that still came sounded like this:

"hey, I made a website with AI and wanted to know if you could fix xxx"

Followed by "it took me about an hour to get this far, it shouldn't take much to fix this problem." Then I'd mention the price and they'd be gone. By the time I wrote the Fiverr post in August 2025, I was at about two messages a week.

That's one inbox on one platform. It proves nothing about the market. But it's the same thing Upwork's CEO said in August, that "lower-complexity work continues to shift toward automation." The small fix and the simple landing page: those buyers can now generate something in ten minutes and only come back when it breaks.

What the 34% Number Actually Measures

Here is the sentence from Upwork's Future Workforce Index release:

"Freelancers performing AI work on the Upwork Marketplace earn 34% more per hour than those not incorporating AI."

Read it slowly. It compares two groups of freelancers on Upwork's own platform. It doesn't follow the same people before and after they picked up AI, so it can't tell you AI raised anyone's rate. Someone building AI integrations may have been expensive before any of this. The release gives no dollar figure and doesn't say whether experience or job category was controlled for. It also measures AI work, which is a different thing from having "AI skills" on a profile.

And Upwork has a stake in the answer. AI is one of its three named growth pillars, and the jobs were sorted into AI categories by an LLM-based pipeline.

The next lines of the same release are more useful than the headline. Generative AI and creative production work saw 90% more contract starts, while per-contract earnings fell 13%. Freelancers doing "more complex work with AI" saw earnings rise 45% year over year. AI-augmented professional services grew 72% in volume, with earnings up 22%.

Those are all people using AI, moving in opposite directions. My reading: the first group sells what the tool produces, and the second sells the decision about what to do with it. That's my inference, because the release doesn't define the buckets. Upwork itself only says lower-complexity AI execution "may become less lucrative as it scales."

The Math That Isn't in the Headline

Upwork says GSV from freelancers with AI skills reached an annualized run rate of nearly $450 million in Q3, up 55%. That's about $112 million a quarter, or roughly 12% of the $914 million total.

Work backwards and that slice was about $73 million a year ago. Which means everything else went from about $944 million to about $802 million. Down roughly 15% in a year.

That's my napkin math on an estimated run rate, and "everything else" includes Lifted, so don't quote it as an Upwork figure. The direction is hard to argue with though. The premium sits on about an eighth of the volume, and the other seven eighths are shrinking faster than the headline 10%.

More Sellers Are on the Way

The same report says 38% of US skilled knowledge workers now freelance, up from 28% a year ago, and 58% of full-time employees are considering it, up from 36%. That's a survey of 2,400 US workers from March and April 2026, run by a company that makes money when people freelance. Considering isn't doing.

Discount it as hard as you want. If even a part of it is real, sellers are arriving while the client count falls 10%, on a platform whose fee now moves with supply and demand.

I know what the bottom of that market feels like. When I started on Fiverr I felt lost among countless freelancers doing everything for less, partly because they didn't have German tax rates to cover. My first Fiverr Pro application got rejected, and I had to wait 180 days to try again. The fear that someone on Fiverr would do it for less was part of why my first website went for EUR 250.

What I'd Do With a One-Person Shop Right Now

I already made the move the platforms are making, by accident and for worse reasons. In 2023 the workload broke me, I realized my pricing hadn't changed since 2020, and I raised it. I went from 35-40 clients to 15-20. Fewer buyers, more from each one. I can't promise that works for you. It's only the same shape.

Sell the decision. If your deliverable is something a client can get from a chatbot after an hour of effort, expect the -13% bucket. If what they pay for is deciding what to build and answering for it when it's wrong, you're closer to the +45% one.

Stop treating AI as the differentiator. I run my whole one-person agency on Claude, and it's the reason I don't need to hire a junior. It lets me do more. Everyone else has the same subscription available, so it can't be why a client picks me.

Own the relationship. When I wrote that Fiverr post I was still helping people there, and I wasn't relying on it. A platform that raises its cut while its traffic drops is a side door. What survives is the client who knows you, and that takes work: clients can't value what they can't see.

Do the math on your own platform income. Pull last year's earnings and split them into small fixes and everything else. If most of the money is in the small fixes, you now know what you're exposed to.

What I Don't Know

In May, Upwork's CFO talked about "cyclical and near-term factors." By August, its CEO was talking about work shifting to automation. Fiverr points at search traffic. I can't separate those three from the outside, and I haven't seen anyone do it cleanly. Upwork guides Q4 revenue at $169 million to $177 million, against about $198 million a year earlier, so it isn't expecting a quick rebound.

Upwork is also opening the marketplace to AI agents as a new class of client. Its MCP server launched in August, agents have made nearly 50 million requests, and the company estimates that at more than $3 million in annualized Connects revenue. I have no idea yet what it means for a one-person shop when the buyer on the other side is a piece of software.

I can't tell you what marketplace income looks like in 2028. What I can do is look at what you sell and tell you which parts a client can already get from a chatbot. Let's talk if that sounds useful.

Find these posts useful? Mark Wunderlandmedia as a preferred source on Google — my articles will then show up more often in your Search results, AI Overviews and AI Mode.

Set as preferred source

About the Author

Kemal Esensoy

Kemal Esensoy

Kemal Esensoy is the founder of Wunderlandmedia, a one-person practice for web development, SEO and AI consulting, founded in 2018. Before that he worked as a freelance web developer and designer and taught web design courses to more than 3,000 students.