OpenAI and Upwork find AI freelance projects span more tasks
A joint OpenAI-Upwork study found that AI-related projects on Upwork combined more kinds of tasks and were associated with larger fixed-price budgets and longer hourly workloads, while the researchers cautioned that the results do not represent the wider labor market.
OpenAI and Upwork published a joint study finding that AI-related projects on Upwork combined a wider mix of tasks and were associated with higher fixed-price budgets and more logged hours than other projects on the marketplace.
Through April 2026, posted fixed-price budgets for AI-related work were about 70% higher in a raw comparison and 52% higher after the researchers controlled for work subcategory and posting month. Among hired hourly contracts with positive recorded hours, AI-related work logged about 41% more hours during the first 90 days after hire on the same adjusted basis. The adjusted difference in posted hourly rates was about 1.5%. The authors described these as associations, not evidence that AI caused the differences.
The full research report also compared activity patterns across ChatGPT and Upwork. OpenAI has separately published cross-country data on ChatGPT work use. The joint report’s ChatGPT sample covered 36,896 conversations drawn from monthly samples between January 2025 and June 2026. A separate analysis compared ChatGPT activity in the second half of 2025 with Upwork category shares from the first half of 2023, then tracked Upwork share changes through the second half of 2025.
Of 63 matched work subcategories, 43 paired lower ChatGPT activity intensity with an Upwork-share gain or higher intensity with a share loss. The other 20 moved in the opposite direction, making the relationship uneven across categories. The category-share analysis measured shifts in the mix of Upwork work, not the marketplace’s aggregate growth.
The effective number of task clusters represented in AI-related postings rose from 14.6 in the first-half 2023 average to 21.2 in the second-half 2025 average, an increase of about 45%. AI-related posts averaged 3.35 task clusters each, compared with 3.19 for non-AI posts. Among postings with at least two clusters, historically uncommon task pairs accounted for 8.4% of pairs in AI-related posts and 2.4% in other posts.
Upwork classified AI-related posts with an internal GPT-powered system that considered titles, descriptions, requested skills, taxonomy and other context. The ChatGPT and Upwork records were not linked, neither dataset represents the labor market as a whole, the proprietary datasets were not independently reproduced, and the report did not publish precision or recall metrics for the classifier.
More news

OpenAI releases cross-country data on ChatGPT work use

OpenAI reports 130,000 ChatGPT users and 95,000-plus Codex users at Oracle

OpenAI study finds workers repeat ChatGPT tasks tied to other occupations
