DIRTY STATIC

AI. TECH. CULTURE.
THE FUTURE, UNFILTERED.

DIRTY STATIC / DISPATCH Culture11 NOV 2026 β€’ 10 MIN READ
A lone freelance writer at a concrete desk in a brutalist room, laptop glowing caution yellow, a tall stack of job postings shrunk to a small pile beside empty chairs, hazard stripes on the desk edge.
CultureINTELLIGENCE FOR THE MACHINE AGE

Writing Freelance Postings Fall 31 Percent In Year

OpenAI and Upwork compared ChatGPT use with freelance postings and found content writing at opposite ends of the two markets, with its share of postings down about 31 percent in a year. This feature explains what that figure does and does not measure, how it follows a steep drop recorded after ChatGPT launched, and why the work that remains is getting broader and larger in scope.

DS
DIRTY STATIC EDITORSReporting on what comes next
10MIN READ
Start reading ↓

On November 10, OpenAI and Upwork published a study with an awkward split at its centre. Content writing made up 14.5 percent of the ChatGPT conversations the researchers sampled, drawn from personal and work use between January 2025 and June 2026. On Upwork, where clients pay people to do work, content writing made up just 1.3 percent of postings in the first half of 2026.

The gap is widening on the hiring side. The share of Upwork postings held by content writing fell by roughly 31 percent between the first half of 2025 and the first half of 2026. People still need words, in huge volume. More of them are getting those words from a chat window instead of from a person they hire by the job.

The Bigger Story

The report is called The Rebundling of Freelance Work in the AI Age, and the second word of that title does most of the work. The researchers compared 63 work categories across the two platforms. In 43 of them, activities that were common in ChatGPT lost share on Upwork, or activities that were rare in ChatGPT gained share. In the other 20, the pattern ran the other way.

That mixed result is why the authors resist a simple story in which a chatbot eats a job category whole. Video and animation made up 1.0 percent of sampled ChatGPT conversations but 10.0 percent of Upwork postings. Mobile development made up 0.2 percent of conversations and 7.8 percent of postings. Clients still hire people for work that ends in a finished thing they can ship, sell, or show. They hire less often for a block of text on its own.

What clients do ask for, increasingly, is bundles. The mix of tasks inside AI related postings grew about 45 percent more diverse between early 2023 and late 2025, rising from 14.6 to 21.2 on the measure the report uses. AI related postings averaged 3.35 distinct task clusters, against 3.19 for other postings, and pairs of tasks that had rarely appeared together before AI accounted for 8.4 percent of pairs in AI related jobs, against 2.4 percent elsewhere. One example in the report asks for AI generated video, editing, and motion graphics inside a single posting, with one person expected to deliver the finished video.

The money follows the bundle, not the hour. After the researchers adjusted for category and month, AI related fixed price postings carried budgets 52.4 percent higher than comparable work without AI. The gap in advertised hourly rates was just 1.5 percent. Among hired hourly contracts, AI related work involved 40.6 percent more hours in the first 90 days. Read plainly, clients are not paying much more for an hour of a person. They are buying more hours, and more kinds of skill, inside one larger job.

A concrete marketplace hall of freelance work stalls, the writing stall with its shutter half down and glowing caution yellow while video and coding stalls stay open

Context

This is the second wave of the same shock, and the first wave was measured while it was still happening. Researchers at Imperial College London, Harvard Business School, and the German Institute for Economic Research analysed 1,725,587 freelance job posts from July 2021 to July 2023. Writing jobs fell by 30 percent within months of the launch of ChatGPT, a steeper fall than coding, which fell by 20 percent. Their paper, published in Management Science in 2025, also found that the jobs which remained grew more complex, and that more freelancers competed for each one.

Imperial opened its own account of that research with people, not percentages. Early in 2023, the team heard that freelance writers in Kenya were struggling to earn a living in a market that had once thrived, as students turned to AI for work they had once paid writers to do. That detail matters because it names the kind of writing that vanished first. It was priced by the piece, briefed in a line or two, and judged mostly on speed and adequacy. That is a near perfect description of what a language model does cheaply.

A second study, summed up by the Brookings Institution, followed the people rather than the postings. Xiang Hui, Oren Reshef, and Luofeng Zhou found that freelancers in work exposed to generative AI saw about a 2 percent decline in new contracts and about a 5 percent drop in earnings after the tools arrived in 2022. Those numbers look small beside 30 percent, until you notice who they describe. The harm was pronounced among experienced freelancers who offered higher priced, higher quality services. Exposure did not stop at the bottom of the market.

Set the three studies in a row and a timeline forms. First came a sharp fall in postings for text work, measured across close to two million jobs. Then came a quieter squeeze on contracts and earnings for the people who used to win that work. Now comes a fall in share for content writing on the largest freelance marketplace, alongside a rise in larger jobs that bundle writing in with video, code, automation, and integration. The standalone writing brief is not holding its ground in any of the three.

Different Perspectives

For a writer who sold commodity copy, none of this nuance softens the result. If your income came from product descriptions, short blog posts, and rewrites that a client could brief in one sentence, the client now has a tool that does a passable version at once and at almost no cost. Telling that writer that freelance work as a whole is being rearranged, rather than removed, is true at the level of the market and false at the level of their rent. Both things can hold at once, and honest coverage should let them.

Clients see the same shift from the other side of the desk, and their view is not mainly about cutting writers out. The report suggests that many clients still want a person, but they want that person to own an outcome. A campaign, a video, a working chatbot, a site that converts. Writing inside those jobs has not disappeared. It has stopped being the job title, and started being one task among three or four that a single hire is expected to carry. We saw a version of this bargain in Creatives Use AI Daily, Hide It From Clients, where the value creatives claimed was not the raw output. It was taste, judgment, and the final say.

The researchers themselves supply the strongest caution, and it deserves space rather than a footnote. The two datasets in the new report are not directly matched. The ChatGPT sample covered 36,896 conversations from consumer accounts, classed as personal or work related, with schoolwork and unclear cases left out. Upwork postings describe one marketplace, not the whole economy, and direct work arranged off platform leaves no trace in either dataset. The report says its estimates describe associations, not proven causes. A fall in share is also not the same as a fall of the same size in total jobs, since share depends on what every other category does too. The 31 percent figure is real, and it is narrower than the headline version of it sounds.

There is also a reading that favours adaptation, and the numbers give it some support. If AI related projects carry budgets about half again as large and involve about two fifths more hours, then a freelancer who can direct the tools, check their output, and deliver a finished bundle may end up with fewer, bigger, better paid engagements. The catch sits inside the same figures. The hourly rate barely moved. The gain comes from scope, which means the gain goes to people who can credibly sell scope, and that is a smaller and more senior group than the one that used to sell words by the piece.

Writing did not leave the freelance market. It stopped arriving as a job on its own, and started arriving as one task inside somebody else’s larger job.

Why It Matters

Freelance postings are worth watching precisely because they move first. OpenAI makes this point in its own summary of the research. Contracts are short, and clients rewrite their requirements with every new posting, so a change in what businesses want shows up in briefs long before it shows up in national employment counts. If that is right, the Upwork numbers are an early signal for staff jobs as well as gig work. The brief is changing under employees too. It is just changing in private, one team at a time.

That raises a question the data cannot answer yet, about how people learn the trade. Small writing jobs were never glamorous, but they were a first rung. They taught a new writer how clients think, how edits land, and how to hit a deadline with a stranger reading the result. If the first rung is now mostly handled by a model, the path from beginner to the senior person who can bundle, judge, and deliver gets harder to see. The market may still need that senior person in five years. It is less clear where she is supposed to come from, a problem we touched in Voice Actors Are Losing Jobs Already, where early work thinned out while demand for polish stayed.

For clients, the lesson runs in the opposite direction. A bundle is only cheaper than separate hires if the person inside it can actually judge every part of it. The report found clients pairing tasks that rarely sat together before 2023, which means some briefs now ask one freelancer to cover ground that used to take a small team. Some of those briefs will produce excellent work from a genuine generalist with good tools. Others will produce work that looks complete and fails quietly, because no one in the chain had the depth to catch the weak part. The savings are easy to count. The failures arrive later, and under a different budget line.

A writer assembling a large project from concrete blocks that carry video, code, and design tools, the seams between blocks glowing caution yellow

Closing Reflection

The cleanest way to read the 31 percent is also the least dramatic. People have not stopped valuing writing. They have stopped paying a stranger, by the piece, for the kind of writing a machine now drafts in seconds, while continuing to pay for judgment, integration, and finished work that someone will stand behind.

What stays open is who pays for the transition. In the studies above, the cost landed first on Kenyan writers losing student work, then on experienced freelancers losing contracts and earnings, and now on anyone whose whole offer was a block of competent text. The new work is real, larger, and broader, and it asks for skills that take years to build. Whether the market builds a path from the old first rung to that new work, or simply enjoys the savings and lets the ladder rot, is a choice being made right now in thousands of small briefs, most of them too minor for anyone to call it a decision.

Reporting draws on the OpenAI and Upwork report The Rebundling of Freelance Work in the AI Age and its summary, both published November 10, 2026, on the Imperial College London summary of the Management Science study of 1,725,587 freelance postings, and on the Brookings Institution summary of Hui, Reshef, and Zhou, checked November 2026.

Sources

β– 

END OF TRANSMISSION

Stay curious. Question everything.
FILED UNDERCulture
DS.
THE EDITORIAL DESK

Independent by design.

DIRTY STATIC tracks the people, tools, and ideas reshaping artificial intelligence. No hype without evidence. No future without questions.

CONTINUE THE DESCENT

MORE FROM DIRTY STATIC β†˜