OpenAI and Upwork: Task Variety in AI-Related Freelance Jobs Up 45%

There is a popular story: AI can write texts, so copywriters will soon be out of work. A new study by OpenAI and Upwork shows a more complex picture. Demand for freelance work is indeed shifting, but not toward "people are no longer needed" — rather toward different jobs and different combinations of skills.

What Was Compared

The authors used two sources. The first is what people do in ChatGPT (January 2025 – June 2026). The second is job postings and orders on Upwork, where clients look for contractors. The work was divided into 63 categories, and the researchers looked at how the share of each category in job postings changed.

Writing has a noticeably larger share in ChatGPT, while video and animation are more prominent in freelance jobs. In other words, what people willingly do themselves with the help of AI and what they still hand over to specialists are different sets of tasks.

The main pattern: in 43 of 63 categories, what is popular in ChatGPT was losing share on Upwork, while the less popular grew. In the remaining 20 categories the picture is the opposite. Growth in ChatGPT usage alone does not explain how demand for freelancers is changing. OpenAI also points to the "capability overhang": ChatGPT is used for only a portion of what the models can already do.

AI Jobs Became More Complex and Broader

By the end of 2025, clients were asking for AI to be applied in a much larger number of areas: chatbot development, video production, marketing automation, integration into apps. The number of tasks in a posting is roughly the same as in regular jobs, but the combinations differ: combinations that previously almost never appeared in a single brief are now more common.

An example from the report: in one posting a client asks for AI video generation, editing and motion graphics. The freelancer must bring these skills together and deliver a finished video. Previously, such an order would have been split among three contractors or at least three separate projects.

By the numbers: from the start of 2023 to the end of 2025, the set of tasks in AI postings became 45% more diverse. This is equivalent to moving from roughly 15 equally likely tasks to roughly 21.

Money and Scope

Hourly rates in AI projects and regular jobs are similar. However, fixed budgets for AI projects are higher, and so are the hours of work. The data covers the period up to April 2026, and the comparison was adjusted for job type and month of posting. The authors cautiously note that budgets may be higher simply because the scope is broader: the client is buying not "make an image" but an entire process.

Limitations

This is an observation of job postings, not proof of a causal relationship. Work inside companies may change without being reflected in new job postings at all. The data concerns a single platform and the use of ChatGPT specifically, not all AI tools. Moreover, the study was conducted by OpenAI itself, and its conclusions should be read with that caveat in mind.

What to Do in Practice

For freelancers: stop selling a narrow service and start packaging bundles. Clients increasingly look not for an "editor" but for a person who will take the result from generation to the final product. This justifies a fixed price rather than an hourly one.

For studios or agencies: check which of your services overlap with what clients already do themselves in chatbots (for example, basic copy), and shift the emphasis to where integration, quality and accountability for the result are needed.

For product managers and founders: if you hire contractors for AI tasks, budget for greater scope and a fixed price, and in the brief clearly describe the final result rather than a set of separate operations.