Upwork goes all-in on OpenAI across product, fraud, and ops

Upwork isn’t just dabbling in AI — the freelance marketplace is reengineering itself as what VP of Partnerships Adel Farahmand calls an “OpenAI shop.” After evaluating a crowded field of models, the company bet heavily on OpenAI’s stack, threading GPT-3.5, GPT-4o, and ChatGPT Enterprise through every layer of its business. The results are already showing up in the numbers. The Job Post Generator, built on GPT-3.5, slashes the time it takes to create a listing by 80%, and the clients who use it end up spending 9% more on the platform. That’s a direct line from a product feature to revenue lift.

Those early wins spawned a suite of AI-powered tools now bundled under an umbrella called Uma, Upwork’s “Mindful AI.” It pulls in the company’s own marketplace data and layers on OpenAI’s models to act as an intelligent concierge for both freelancers and clients. Features like Upwork Chat Pro give freelancers a GPT-4o-powered assistant to handle repetitive grunt work, while Proposal Tips helps them write pitches that actually stand out. For clients, a beta feature called Best Match insights runs on GPT-3.5 to cut through the noise and surface freelancers whose reviews and skills genuinely match the brief. The early data on Uma is promising: new clients who used an initial version spent 7% more in their first month. It’s a growth loop disguised as a utility.

But the AI infiltration goes deeper than customer-facing products. Upwork is also using OpenAI’s models to fight the inevitable wave of AI-generated spam and fraud that hits any large marketplace. They’ve deployed internal automations to sniff out low-quality and scam job posts, work that used to require tedious manual review. On the employee side, the company’s internal evaluation showed a staggering 98% of team members preferred ChatGPT Enterprise for everything from coding to QA. They’re now rolling out access company-wide, with staff building custom GPTs to automate sales scripts, crunch data, and optimize content.

What’s notable here isn’t just the adoption speed — it’s the breadth. Upwork is simultaneously using AI to make its own workforce faster, keep its marketplace cleaner, and build stickier features for both sides of its two-sided network. The 7% and 9% spending bumps aren’t earth-shattering, but they’re concrete signals that this stuff actually moves the needle on transactions. The question, as always, is whether those early gains compound or plateau once the novelty wears off. For now, Upwork is treating AI less like a feature and more like the operating system underneath its entire business.