Short answer: Freelancing with AI skills is one of the few parts of Upwork that is still growing. Upwork says work billed by freelancers with AI skills rose 55% year over year to about $112 million in its latest quarter, while total marketplace volume fell 10%. The money is going to people who can implement AI in a client’s business, not to people who only resell prompts or generated content.
What did Upwork actually report?
In its third-quarter results, Upwork reported gross services volume (GSV, the total clients spend on the platform) of $914.1 million, down 10.1% year over year. Inside that shrinking total, GSV from freelancers with AI skills grew 55% and reached an estimated annualized run rate of nearly $450 million.
CEO Hayden Brown said on the earnings call that clients “are increasingly turning to Upwork for the expertise to implement it, integrate it into their operations, and get more value from AI.”
| Metric (latest quarter) | Value | Change year over year |
|---|---|---|
| Total GSV | $914.1M | -10.1% |
| GSV from freelancers with AI skills | ~$112M | +55% |
| AI-skill GSV, annualized | ~$450M | n/a |
| GSV per active client | $5,418 | +8% |
| Contracts of $100,000 or more | n/a | +7% |
How should you read the 55% number?
Treat it as a direction, not a pay forecast. Three caveats matter:
- It counts profiles, not tasks. Upwork’s investor presentation defines the figure as all GSV from freelancers whose profile lists at least one skill on Upwork’s internal list of generative AI skills. A web developer who adds “ChatGPT” to their profile counts, even if the job wasn’t AI work.
- It is a new metric. Upwork said on the call that it is retiring its older measure based on job posts that explicitly asked for AI, so there is no long history to compare against.
- It says nothing about individual earnings. A growing pool can still be split among many more freelancers.
Which AI skills are clients actually paying for?
Upwork’s In-Demand Skills report and Fiverr’s Business Trends Index point the same way: implementation and video, not generic content.
| Skill | Signal | Source | What it measures |
|---|---|---|---|
| AI video generation and editing | +329% | Upwork | Client hiring demand |
| AI integration | +178% | Upwork | Client hiring demand |
| AI data annotation and labeling | +154% | Upwork | Client hiring demand |
| AI chatbot development | +71% | Upwork | Client hiring demand |
| Claude Code specialists | +938% | Fiverr | Searches, not orders |
| n8n AI automation | +125% | Fiverr | Searches, not orders |
| AI UGC video ads | +265% | Fiverr | Searches, not orders |
Fiverr does not publish the base numbers behind those percentages, so a large jump can start from a small number. An independent count by flconsole found that 27.6% of about 73,000 new Upwork jobs in one month mentioned AI or automation, but only about 3.5% were actual AI automation projects. Most “AI” jobs are ordinary work where the client expects you to use AI.
What does it cost to sell AI services on Upwork?
| Cost | Amount | Notes |
|---|---|---|
| Freelancer service fee | 0% to 15% per contract | Shown before you propose and locked for the contract (Upwork Help) |
| Connects (to send proposals) | $0.15 each | Each job lists how many it needs; Connects expire after one year (Upwork Help) |
| Freelancer Plus | $19.99 per month | 100 Connects a month and 0% fee on clients you bring (Upwork pricing) |
| Withdrawal and currency conversion | Varies by method | Check your payout method before quoting rates |
| AI tool subscriptions and API usage | Varies | Price these into fixed-price quotes |
Upwork’s own example: to take home $20 an hour on a contract with a 10% fee, you need to bill about $22.22.
Who is this a realistic option for?
| Good fit | Poor fit |
|---|---|
| Developers who can connect an LLM or agent to a client’s real systems (CRM, helpdesk, docs) | Anyone hoping to resell raw ChatGPT output as writing or design |
| Video editors who already have a portfolio and add AI generation to it | People with no samples who expect fast first contracts |
| Automation builders comfortable with n8n, Zapier or Make plus testing and handover | Anyone who can’t support a workflow after it breaks |
| Data people who can do careful labeling, evaluation or cleanup | Anyone needing guaranteed monthly income soon |
What are the risks?
- Prototype versus production. Clients pay for things that keep working. Budget time for error handling, monitoring and documentation.
- Data and confidentiality. Confirm what client data you may send to third-party AI APIs, and put it in writing.
- Proposal spend. Connects add up if you apply broadly. Track cost per contract won.
- Fast-moving tools. A workflow built on one model or API can change price or behavior. Say so in your scope.
- Platform dependence. Upwork’s overall volume is falling. Use it to win first clients, then build repeat work.
How can you start this week?
- Pick one lane from the skills table that matches what you can already do.
- Build one small, honest sample: a working support chatbot on public docs, an n8n workflow with a short demo video, or a before-and-after AI video edit.
- Read 10 to 20 real job posts in that lane and note the problems clients describe and the budgets they list.
- Write a profile headline about the outcome (“I connect your helpdesk to an AI agent that drafts replies”), not the tool.
- Quote with fees and tool costs included, and offer a small paid first milestone.
What to watch next
Upwork is also pushing work through AI agents. It says its MCP server’s weekly active users rose 20x since launch and that agents made nearly 50 million requests. If clients increasingly hire through AI assistants, clear service descriptions and verified reviews may matter more than proposal volume.
We have no affiliate relationship with Upwork or Fiverr. Figures are company-reported; we have not independently verified them.