I Went Looking for the Numbers on Creative Work in 2026

The two answers you already know, and why I stopped trusting both

There are two stories going around and you have heard them both. The first says the industry is finished. Streaming overspent, the strikes broke something, AI is coming for the rest, and anyone starting now has missed it. The second says none of this is new, every generation of creatives has been told the same thing, and the work will be there for people who are good.

I believed the second one for a while, mostly because I wanted to. Then I looked at what the statistical agencies publish, and neither story survives contact with it. The sector is not collapsing. It is also not fine. It is moving, and the totals hide where.

The losses are real, and they are in one place

Start with the number I found hardest to look at. The UK government's own economic estimates show the creative industries losing 70,000 jobs in the year to March 2024, a fall of about 2.8 per cent, with the culture sector down another 11,000.

What makes it worse is the year before. In that one, every single subsector grew by more than 4 per cent. So this is not a slow decline anyone could plan around. It is a turn.

In the United States the picture in traditional screen production is starker. Reporting on the Bureau of Labor Statistics figures describes motion picture and video production shedding 49,000 jobs across the decade to February 2026, a fall of 21 per cent, with exhibition down another 17 per cent. Writing for television and film reportedly fell 10.2 per cent in a single year, from 2024 to 2025.

What the strikes did, and what they only get blamed for

It is tempting to draw a line from the 2023 strikes to everything after them. I do not think the line holds.

The streaming companies turned from growth to profitability in 2022 and 2023, before the strikes, and content budgets came down as a result. The strikes landed on top of a contraction already underway. If you are deciding what to do next, this distinction matters more than it sounds: a pause caused by a labour dispute ends when the dispute does. A structural change in what gets commissioned does not.

The AI question, answered with measurements instead of feelings

This is the part I expected to be bleak, and it turned out to be the most interesting thing I read.

Gallup did the study I wanted somebody to do. They built an index of how exposed each occupation is to generative AI, based on how much of its work the technology overlaps with. Then they compared highly exposed artistic occupations against less exposed ones, using Bureau of Labor Statistics wage data, the Census Bureau's American Community Survey to capture freelancers, and their own panel, across 2017 to 2024.

The earnings estimates for the highly exposed group came out, in their words, slightly positive and not statistically distinguishable from zero. Hours worked went up from 2022. About one artist in four reports using AI frequently, against one in five across the whole workforce, and mostly for idea generation rather than for the work itself.

That is not the finding I expected. It is also not a clean bill of health, because the same period shows something else.

Where the displacement is actually landing

The categories getting hit are specific, and they are the ones where the output is text or a single flat image. Freelance writing and translation volume fell by around 30 per cent year on year. Graphic design fell around 17 per cent. Roughly a quarter of illustrators and over a third of translators report having already lost work to generative tools.

Hold those two findings next to each other. Aggregate artist earnings show no measurable hit. Specific freelance categories show a third of their volume gone. Both can be true, because the average across a large group of occupations will not show you what happened inside one of them.

If your work is commissioned copy, stock-style illustration or translation, the aggregate number is not about you. If your work involves being in a room, on a set, or responsible for a decision somebody has to sign off, it probably is.

What the wage data cannot see

There is a piece of research I keep coming back to. It is called CREAATIF, and it was run by Queen Mary University of London, the Alan Turing Institute and the Institute for the Future of Work, working through Equity, BECTU, the Musicians' Union and the Society of Authors. They surveyed 335 people who work in the creative industries.

Their finding is not about pay. It is about security. Freelancers carry the exposure and lack the protections salaried staff have, and creative work is being used as training data without consent or payment.

So Gallup and CREAATIF are not in conflict, which took me a while to see. Gallup measures whether artists are earning less. CREAATIF measures whether the work is becoming less secure. You can hold the same income and have a worse life, and anybody who has been freelance through a bad year knows exactly what the difference feels like.

Where the work is going, with a warning attached

The pattern reported is that hiring is strongest where craft meets engineering: real-time engines, virtual production, pipeline and tools work, and the roles that sit between departments. There are claims of demand outstripping supply for virtual production supervisors, and of creative job postings rising through late 2025 into 2026.

I could not verify any of it to the standard of the contraction figures. Most of it comes from training providers and from market research firms selling forecasts, and both make money when you believe demand is high. I am including it because the direction is consistent across several accounts. But take care, because you should not retrain on the strength of a school's prospectus.

What I would say with more confidence is narrower. The parts of this industry with a compliance deadline attached are hiring, because a deadline is not an opinion. Accessibility work has a legal requirement behind it in Europe since June 2025. Delivery, loudness and provenance work has specifications behind it. None of that is glamorous and all of it is checkable.

What I could not find out, which you should know before you use any of this

Every figure above comes from one of two places, the United Kingdom or the United States. That is not a choice about what matters. It is where the public statistics are good enough to quote. If you work in Nigeria, India, South Korea or Brazil, three of which have screen industries larger than the ones I have data for, this article is not describing your market and I would not want you to read it as though it were.

The sector definitions are also broader than the work most of us do. Government statistics on the creative industries include architecture, advertising, museums, publishing and software. A fall of 2.8 per cent across all of that tells you the direction and not the shape. A camera operator and a museum curator sit in the same total.

Games and music barely appear in what I found. Both employ enormous numbers of people, both have had their own upheaval, and neither is captured well by the screen production figures I have quoted. Since writing this I have found counted data for one corner of it, and it is on the site as Game Audio Jobs in 2026, read from 142 job postings rather than from sector totals. It covers game audio and nothing else, which is roughly how much of this gap good data actually closes. If somebody tells you the creative sector is up or down by a number, ask them which sector they mean.

And all of this is backward looking. The most recent complete UK figures end in March 2024. We are more than two years past that. Anything anybody tells you about right now, including me, is guessing.

So what would I tell my two friends

To the one deciding on the course: go if you want the skill, not if you want the job guarantee, because nobody can offer you the second one right now. Pick a programme that teaches you to plan, to record clean sound, to finish something and to hand it over, because those transfer to any tool. Do not pick one on the strength of its equipment list.

To the one thinking about a salary: the security question is real and the data supports taking it seriously. The precarity finding is better evidenced than the doom narrative and better evidenced than the reassurance. Taking a salaried job in this climate is not giving up.

To both of them, and to you: the useful question is not whether the industry is dying. It is which part of it you are standing in. Traditional screen production has contracted sharply. Text and flat-image freelancing is being displaced now. Work with a specification or a legal deadline behind it is steady. Those are three different weather systems and the forecast for one tells you nothing about the others.

I found this more useful than either of the stories I started with. I hope you do too.

Sources

• Gallup, AI Is Changing Creative Work, but the Arts Aren't Disappearing. Read directly. Uses Bureau of Labor Statistics employment and wage data, the Census Bureau's American Community Survey and Gallup's own panel, 2017 to 2024, with an index of occupational exposure to generative AI. The earnings estimates for highly exposed artistic occupations are described as slightly positive and not statistically distinguishable from zero.
• CREAATIF, Crafting Responsive Assessments of AI and Tech-Impacted Futures. Read directly. Run by Queen Mary University of London, the Alan Turing Institute and the Institute for the Future of Work during 2024, with 335 creative workers reached through Equity, BECTU, the Musicians' Union and the Society of Authors. Its findings concern job security, consent and the use of creative work as training data.
• ArtsProfessional, reporting the DCMS economic estimates for April 2023 to March 2024. Read directly. Gives creative industries employment falling from 2,457,000 to 2,387,000, a drop of 70,000, and the culture sector falling by 11,000, against growth of more than 4 per cent across all nine subsectors the previous year.

The freelance category figures come from reports I found in several places, rather than from any single publication.