Upwork says AI freelancers earn 34% more. Its own filings disagree.

Upwork published its Future Workforce Index 2026 on July 14, and one number is doing the rounds: 34%. Freelancers doing AI work on its marketplace earn 34% more per hour than freelancers who don't. The same report says 38% of skilled US knowledge workers now freelance, up from 28% a year ago, and that 58% of full-time employees are considering it, up from 36%.

Those figures are accurate as far as they go. They also describe something different from what a person about to open an Upwork account will hear.

Three months earlier, on May 7, the same company told its investors that gross services volume for Q1 2026 was $987.1 million, essentially flat year over year at -0.1%. Active clients fell 3% to 784,000. Revenue rose 1%, to $195.5 million. In the same release, Upwork announced a restructuring that cuts roughly 24% of its own workforce, with pre-tax charges of $16 million to $23 million.

Read the survey and the filing together and the 34% stops looking like a raise. The buyer money on the platform didn't grow. The number of sellers did. A premium inside a pot that isn't getting bigger is a sorting mechanism, and it tells you who is taking a larger slice and who is paying for it.

The premium is real, and it sits in one place

The index breaks earnings out by the kind of AI work, and that breakdown is where the story lives.

Complex AI-augmented work saw earnings rise 45% year over year. AI-augmented professional services, meaning domain experts folding AI into a field they already knew, grew 72% in volume with earnings up 22%. Nick Bloom, the Stanford economist on Upwork's own advisory council, put it plainly in the release: the value is concentrated in complex work where people apply expertise, judgment, and business context on top of AI.

Then there's the third tier. Generative AI and creative production contract starts grew 90% year over year. Per-contract earnings in that category fell 13%.

That's the whole argument in two lines. The category adding the most new work is the only one where pay went down.

The tier that's growing fastest is the tier being sold

Ninety percent growth in contract starts is not a mystery. It's the output of every "make money with AI" funnel pointing at the same door: write the prompt, generate the asset, deliver the file. It needs no domain expertise, which is exactly why it clears so fast and prices so low.

The 34% premium and the 13% decline aren't in tension. Averages hide their own composition. If the top tier is pulling 45% more and the entry tier is shedding 13% per contract, an aggregate that says "AI freelancers earn more" is true and useless at the same time, because nobody earns the average. You land in a tier. We ran the rate side of this in more detail in what Upwork and Fiverr actually report, and the shape hasn't changed: the money follows judgment.

Tool access is the part that got commoditized. When the tool was the moat, knowing Midjourney was a service. Now the buyer has the same subscription you do.

The filings say what the survey can't

A marketplace with healthy demand shows volume up, clients up, take rate holding. Upwork's Q1 2026 has volume flat, clients down 3%, and revenue divided by GSV working out to 19.8%.

The staff cut is the part worth sitting with. This is analysis rather than fact, so weigh it accordingly: a company that believed its own survey's growth story does not cut 24% of its people in the same quarter. Upwork's investor deck, as reported by Investing.com, puts AI-related GSV above $300 million on an annualized basis and growing more than 40% year over year. Against roughly $3.9 billion of annualized GSV, that's under a tenth of the marketplace. The AI line is growing fast off a base too small to move the total, which is precisely what "GSV flat" already told us.

The two documents have different jobs. The index sells the marketplace to sellers. The 8-K answers to the SEC. When they disagree, the one with legal liability attached is the one to read.

Fiverr ran the same sort, earlier and harder

If Upwork's numbers are ambiguous, Fiverr's Q1 2026 results from April 29 are not.

Marketplace revenue fell 13.6% year over year, to $67.1 million from $77.7 million. Annual active buyers dropped 17.8%, to 2.9 million from 3.5 million, which is more than 600,000 buyers gone in a year. Take rate held flat at 27.7%.

And yet annual spend per buyer rose 15.4%, to $356 from $309, while clients completing projects over $1,000 grew 18%.

Fewer buyers, each spending more, concentrated in bigger projects. Fiverr is more exposed to small commodity gigs than Upwork is, so it hit the wall first. Its own full-year 2026 guidance calls for revenue of $380 million to $420 million, a range whose top end is still 3% below last year.

Two platforms, two sets of audited numbers, one pattern: the low end is being absorbed and the high end is consolidating.

What could prove this wrong

The index surveyed 2,400 US skilled knowledge workers in March and April 2026, with a stated margin of error of 2% at 95% confidence. A 10-point jump in freelance participation in twelve months sits far outside that margin, which leaves two options: either something extraordinarily fast happened to American knowledge work, or the survey is measuring a softer thing than the headline implies. "Do you freelance" catches anyone who took one paid side project last year, and that's a real answer to a real question. A third of American knowledge workers did not walk off payroll in a year.

The rest of the caveats cut against my read, and they're worth stating:

• Flat GSV is not flat freelancer income. Upwork is one platform. Work that starts there often moves off it, and that migration would look identical to decline in these numbers.
• A 13% drop per contract isn't proof hourly rates fell. Contracts may simply have gotten smaller and faster, which is what you'd expect when AI compresses delivery time. Same rate, less time, less money per contract.
• Self-reported survey earnings run high. That applies to the 34% as much as to anything else here.

None of that rescues the entry tier. Volume up, price per unit down, and a buyer pool that shrank at both platforms is the same conclusion under every reading above.

Who this is for

If you're deciding whether to start selling AI services on these platforms this quarter, the data says something specific: the door with 90% more traffic through it is the one where pay per job fell 13%. That's where the funnels point, and it's where you're competing against everyone who watched the same video.

The tier that pays is the one where you bring something the buyer can't prompt for. Domain knowledge, accountability for an outcome, the judgment to know when the model's output is wrong. Upwork calls that person an "AI Orchestrator", which is marketing language for an old idea: the expert who now works faster. The 45% is going to people who were already experts in something before AI showed up.

If that isn't you yet, the honest read is that the platform premium won't cover the gap. It rewards what you knew before you got there. Related reading on the earnings side: the median side hustle earns $200 a month and what AI data labeling actually pays.

Sources

• Upwork's Future Workforce Index 2026, July 14, 2026: the 34% premium, the 38% and 58% survey figures, the 45% / 22% / -13% earnings splits, the 90% and 72% volume growth, and the methodology.
• Upwork Reports First Quarter 2026 Financial Results, May 7, 2026: GSV of $987.1M, active clients of 784,000, revenue of $195.5M, GSV per active client of $5,138, and the 24% restructuring.
• Fiverr Announces First Quarter 2026 Results, April 29, 2026: marketplace revenue of $67.1M, 2.9 million active buyers, $356 spend per buyer, 27.7% take rate, and full-year guidance.
• Upwork Q1 2026 slides: AI growth accelerates amid flat revenue, Investing.com, May 2026: the $300M annualized AI GSV figure and the 19.8% take rate. Single-source reporting on the investor deck rather than the filing itself.

This is not financial advice.