A client can ask for an AI-generated video. Or the brief can include editing, motion graphics and a finished piece ready to use. Those are different jobs—even if both start with the same tool. New freelance research makes that difference worth watching before anyone promises cheaper projects or better pay.

Based on 5 reviewed sources · See sources

The project gets bigger. The hourly rate barely moves.

The joint OpenAI–Upwork report, published on 10 November, finds AI-related work combining more kinds of tasks. One example brings generated video, editing and motion graphics into one posting. Its price-and-hours analysis separates three measures that should not be collapsed into a single pay claim.

A fixed-price budget describes the proposed project. An hourly rate prices time. Recorded hours show how much hired work happened within the study’s window. More work at a similar rate can increase a contract’s value without making each hour more valuable. None of those measures alone tells a freelancer what they will take home.

OpenAI chief economist Ronnie Chatterji’s announcement summarises the price-and-hours finding below. It is the research sponsor’s explanation, rather than an independent assessment.

From the source · Ronnie Chatterji ·

OpenAI’s chief economist distinguishes larger project budgets and more hours from the small adjusted hourly-rate difference.

View the original post on X ↗
The detail

Three measures, three different questions

AI-related versus other Upwork work, after adjustment.

Three measures, three different questions. AI-related versus other Upwork work, after adjustment.
MeasureAdjusted difference
Posted fixed-price project budget+52.4%
Posted hourly rate+1.5%
Hours in hired hourly contracts, first 90 days+40.6%
Analysis through April 2026; controls for work subcategory and posting month. These are different samples, not three outcomes for the same jobs. Budgets and rates are posted values. The comparison does not establish a causal AI premium.Source: OpenAI–Upwork report, figure 2.4, page 10 · 2026-11-11

A smaller slice is not a count of jobs lost

Across 43 of 63 matched work categories, tasks common in ChatGPT lost share on Upwork; 20 moved the other way. That is a change in the marketplace’s mix, not a measurement of how many jobs disappeared. The report compares separate datasets rather than following the same clients from a conversation to a hiring decision.

Independent research supplies a reason to resist a simple “everyone wins” reading. Xiang Hui, Oren Reshef and Luofeng Zhou’s 2024 study found short-term employment and earnings declines for freelancers more exposed to early generative-AI releases. It studies a different period and outcome; it is not a replication of the new report.

Both questions matter to someone planning a creative team: what is being bought, and who is still being hired? An expanding brief does not answer the second. Nor should a European agency treat findings from this marketplace as a forecast for its local rates.

Decide what the extra exploration is for

Report coauthor Nadav Kunievsky, affiliated with Upwork in the paper, offers a hypothesis: easier experimentation may encourage people to explore more alternatives for an open-ended project. That could mean more human work even when individual steps become easier. His thread proposes an explanation; it does not demonstrate the cause of the observed hours difference.

For a commissioning marketer, the useful move is to specify the finished outcome, the alternatives worth exploring and who decides when exploration stops. If a brief grows from one asset into a campaign, make that change visible in its scope and budget. Otherwise a faster first draft can quietly become an unlimited revision promise.

Our visual feature, The image is ready. Is the campaign?, follows the checking and revision work that remains after generation. This research asks a related question at marketplace scale: whether the thing clients commission is changing along with the tool.

Sources

Written with AI assistance from the sources above. Analysis and practical implications are our interpretation. Our editorial approach.