LA RIVOLUZIONE ALGORITMICA

Are Illustrators Losing Their Jobs to AI?

written by Francesco D'Isa
Are Illustrators Losing Their Jobs to AI?

A few days ago, Fumettologica picked up a survey by the British Association of Illustrators (2026) with a telling headline: “More and More Illustrators Are Losing Work Because of Artificial Intelligence.” Accustomed as I am to apocalyptic headlines about AI, particularly in the creative industries, and given comic artists’ notorious hostility towards the technology, I admit I wondered how much truth there was to it. Then I thought perhaps I was being overly suspicious. After all, an impact on employment is more than plausible, and the story has been making the rounds in various international publications, with even more dramatic headlines (“almost half of illustrators lost work to generative AI”).

The first thing I did, naturally, was read the report, which, once you get past the sensational headline, actually says very little. It merely records that 46% of the 549 respondents to a questionnaire circulated by an association campaigning on copyright issues reported having lost at least one job because a client had opted for AI. There are some fairly obvious problems here: the sample is not particularly large, losing a job is not the same as losing your job, and, above all, nobody explains how the respondents established that AI was actually responsible for the loss.

The report also states that 98% of respondents want prior consent to be required for model training and that 86% would refuse to license their work for training under any circumstances. These are all perfectly legitimate positions, but one begins to suspect that the people most inclined to respond were those already opposed to the technology. There are other interesting figures: 62% of respondents had received AI-generated images in briefs or mood boards, while 36% had been asked to redraw or refine them. These figures are more robust than the 46%, because they concern something respondents have witnessed firsthand, the arrival of generated images in creative briefs, rather than requiring them to speculate about its causes. The new technology is affecting work, and even many illustrators who reject it end up working with it, albeit indirectly.

Alongside the monopoly held by those who control these tools, I believe the employment question deserves close scrutiny and cannot simply be dismissed because of one sensationalist study. In other words, the issue needs to be taken seriously. That is why I decided to do some research, and what emerged was a picture considerably less apocalyptic, but one that offers little comfort to those who dismiss the problem as nonexistent or negligible. As so often happens, things are not that simple, and I hope a little clarity might prove more useful than another alarmist headline.

What the Surveys Say

Surveys conducted by professional associations are the most frequently cited sources for measuring how many illustrators have actually lost work, but they are also among the most susceptible to bias. They collect voluntary responses in a climate of mobilisation against these technologies, where those who have been affected, or feel affected, are more likely to participate. They also ask respondents to attribute the loss of commissions to a particular cause, even though nobody can know for certain why a client has simply stopped calling.

Comparing these surveys also reveals just how much the results depend on the way the questions are phrased. The same British association, for example, had recorded a 32% rate of respondents reporting the loss of at least one job the previous year, using a sample ten times larger (Association of Illustrators 2026). The jump to 46%, rather than reflecting an actual increase, might simply be the result of a drastic change in the sample.

In Japan, a much larger survey was conducted among nearly 25,000 creative professionals, again in an environment hostile to AI. The study found that 12% of freelancers had experienced a decline in income because of generative tools (Freelance League of Japan 2026). Once again, whether AI was actually responsible was left to the respondents’ judgment. The figure is more modest, although it is difficult to establish whether the differences reflect local circumstances or the survey methodology.

In Germany, the Illustratoren Organisation surveyed 856 illustrators in 2025, three-quarters of whom were freelancers. Of these, 63% said they could feel the effects of image generators on their work. Two years earlier, the figure had been 23%, but the question at the time referred to effects that were “clearly attributable” to AI. In the same survey, the distribution of revenues for 2024 showed no significant differences compared with the previous survey. As for usage, 7% reported using AI to create images for commissioned work, while 20% used it exclusively to generate ideas (Illustratoren Organisation 2025). Here, too, however, the respondents are people who are still working, and stable revenues among those who remain tell us nothing about those who have left the profession.

Self-reported AI usage also varies considerably between surveys: between 4% and 12% of respondents admit to using it in surveys conducted by professional associations (Society of Authors 2024; UK Comics Creators Research Report 2026), compared with 36% of professionals working in video games (GDC 2026) and 86% of creative professionals, predominantly designers, who responded to Creative Boom (Creative Boom 2026).

The variation is enormous, and it is likely that each survey draws from its own particular audience. But there is also a problem that may lead to underestimation: people who have already left the profession no longer respond to these surveys. In short, the interviews suggest that the phenomenon exists, but to understand how widespread it is, we need to look elsewhere.

What the Platforms Reveal

One study examines a major stock-image platform that has allowed AI-generated images since 2023. AI remains prohibited, however, in editorial images depicting real events and brands, making it possible to compare categories where AI was permitted with those where it was not.

Samuel Goldberg and Tai Lam (2026) observe an explosion in production, which increased by 136%, and an 82% rise in total sales. Meanwhile, sales of images created without AI fell by 28%, and the number of artists not using AI declined by almost a third.

In smaller niches, production grows even faster, and sales by artists who do not use AI fall more sharply. Images depicting people hold up better than other categories. The authors suggest a possible legal constraint: selling a portrait of someone requires a signed model release, and even a generated face might resemble a real person closely enough to require one. However, they cannot rule out other explanations, such as production costs or customers’ preference for authentic photographs.

In short, AI-generated work has partly displaced non-AI work, but it has also expanded the market. Generated images sell in large numbers, although each individual image has a lower chance of finding a buyer, even as the average quality of what gets sold has increased. The study does not tell us how much those producing images with AI actually earn.

On Pixiv, a major Japanese illustration platform, Sueyoul Kim, Ginger Zhe Jin and Eungik Lee (2026) observe an average 10% decline in illustrators’ uploads relative to those of comic artists following the launch of NovelAI, an AI image-generation program, in October 2022. Comic artists were less exposed because producing a comic requires a consistency of style that image generators at the time could not guarantee.

The decline reaches 14% among those linking to websites where they sell their work, through subscriptions, art-book sales, paid commissions and so on, and 15% among the most productive 1%, while the least active half actually increases its output by 7%.

Among illustrators who were already active, fewer than 0.5% adopt AI as their primary tool. Almost all AI-generated output comes from new accounts.

Other studies have focused on the most prominent creators: photographers on a website whose photographs had been selected among the best submitted (Peukert et al. 2024), and featured artists on DeviantArt (Lin 2024). They find declines ranging from 21% to 38%, which, according to Kim, Jin and Lee, risk overestimating the average effect.

These studies have limitations, too. Each examines a single platform and measures uploads, listings or preferences, while income remains invisible. An illustrator posting less on DeviantArt, as Lin himself shows, may simply have started publishing elsewhere. The observation periods are also relatively short, ending between 2023 and spring 2024, when AI models were considerably less capable than they are today.

In April 2018, a design contest platform introduced a $20 automated logo generator, far less capable than today’s tools. Over the following six months, contests for simple logos with prizes below $110 fell by 25%, while those for more expensive logos declined by 5%. Other design contests, meanwhile, increased by 10%.

Less skilled designers were more likely to leave the field, while more experienced designers shifted towards more complex assignments (Lysyakov and Viswanathan 2023). In these segments, namely standardised, low-cost work, the picture looks remarkably similar to the one suggested by Fumettologica’s headline.

On Upwork, following the arrival of AI generators, image-related services experienced a 2.1% decline in assignments and a 5.2% fall in monthly earnings (Hui, Reshef and Zhou 2024). On another platform, listings for image-creation jobs fell by 17% relative to manual tasks (Demirci, Hannane and Zhu 2025), although the authors’ graphs suggest that the decline had begun before the introduction of AI generators.

For now, national statistics have been unable to detect the effect. In Denmark, Anders Humlum and Emilie Vestergaard (2026) combined a survey on AI usage with official income records. Across eleven occupations, they found no appreciable effects on earnings or working hours and ruled out changes greater than 2%. The study, however, concerns chatbots rather than image generators.

In the United States, the earnings of artists most exposed to language models had not declined through 2024 (Makridis 2026). But these are averages drawn from large populations of workers, all of whom were employees in the Danish study, and freelancers can easily disappear into the statistics. In other words, the absence of measurable effects tells us relatively little about illustrators.

Who Benefits?

The companies selling these tools are the ones most eager to talk about their benefits, which hardly inspires confidence. Nevertheless, some evidence also emerges from independent research.

On Mihuashi, China’s leading platform for commissioned artwork, the spread of NovelAI in autumn 2022 coincided with a 121% increase in orders and a 56% rise in total revenue for anime-style character illustrations, the category closest to what NovelAI could produce. Meanwhile, the average price per artwork collapsed by 64% (Zhang, Yuan and Xiong 2023).

Demand also increased because lower prices attracted private customers who had previously been unable or unwilling to commission artwork. The study, however, rests on a rather weak foundation: it compares only one AI-exposed category, anime characters, with a single other product category, desktop wallpapers.

And there is no guarantee that individual artists benefit. They receive more orders, but at much lower prices. On Mihuashi, moreover, most of the subsequent orders go to artists who were already registered on the platform, while little is known about the earnings of newcomers producing work with AI.

On Pixiv, 1.75% of new accounts using AI have a commercial profile, compared with 2.33% of already established illustrators and 1.51% of newcomers who do not use AI (Kim, Jin and Lee 2026). This mainly tells us that newcomers sell less than those who were already active.

Among contributors supplying AI-generated images to stock agencies, an industry survey of around 160 contributors found that 63% earned between $1 and $100 per month, while only approximately 15% made more than $250 (Microstock Life 2024). However, the survey does not compare these figures with the earnings of people uploading traditional photographs or illustrations.

Among established sellers, a study of Fiverr found that designers who added an AI-based service to their offerings gained visibility, with a stronger effect among those charging more for it than for their traditional services. The study, however, measures likes and comments rather than revenue (Zhu et al. 2026).

But how does the creative process itself change? Who benefits most from using AI?

In two experiments published in Information Systems Research (Hou et al. 2025), involving design students and professional graphic designers, AI improved the ideation phase for everyone. During execution, however, it primarily helped less experienced participants and actually slowed down professionals, because it prompted them to reconsider their established working methods more frequently. In practice, AI narrowed the gap between those with more training and those with less.

Annual interviews conducted between 2021 and 2025 with seventeen Chinese digital painters reveal a similar process. After an initial period of scepticism, many began relying on AI for references, backgrounds, colour palettes and drafts to show clients. Some gained speed and even found inspiration.

By 2025, however, their experiences had diverged. Some had developed a hybrid workflow, combining hand-drawn sketches with manual finishing, which allowed them to retain a sense of ownership over their drawings. Meanwhile, five of the fourteen reported declining pay, fewer opportunities or unemployment. One of them now produces three or four concepts per day instead of one per week, for roughly the same pay as before (Meng et al. 2026).

Complicating matters further, almost no study manages to measure undisclosed AI use. In a KPMG survey involving more than 48,000 workers across 47 countries (KPMG and University of Melbourne 2025), 57% admitted to concealing their use of AI.

Experiments by Joel Carnevale and colleagues (2026) also show that disclosing AI use damages the reputation of even an established author, while explicitly stating that one does not use it offers no advantage over remaining silent.

Stigma comes at a cost. On YouTube, when a prominent comment beneath a video begins claiming that it was made with AI, views fall by an average of 10%. In an experiment using identical videos, participants were 23% less likely to choose a video labelled as AI-generated than those who saw it without the label (Fong, Lam and Natan 2026).

Among illustrators, where the stigma is particularly strong, silence is the most rational choice. Studies of online platforms can identify AI use only when it is disclosed or detected by an algorithm. On Pixiv, moreover, only those who use AI as their primary tool are counted as adopters. The tiny proportion of illustrators identified as AI users should therefore be understood as a minimum estimate. Anyone who uses the technology secretly, or only for drafts and sketches, is counted among the “purists.”

closeup of female dressmaker sketching and drawing dresses with graphic tablet. Horizontal shape, side view

A Familiar Story, at Least in Part

This is not the first time a new technology has disrupted the labour market.

When photography entered bourgeois drawing rooms, the first to lose work were portrait painters who made a living producing small, inexpensive images in high demand. Commissioned painting, meanwhile, survived for considerably longer.

In his Salon of 1859 (Baudelaire 1859), Baudelaire dismissed the photographic industry as a refuge for failed painters. Meanwhile, a caricaturist like Nadar became one of the most important photographers of the century, putting the eye he had developed through drawing to use in a new medium.

A century later, photography replaced much of the illustration used in magazines, while many illustrators adopted it to work more quickly. Norman Rockwell, for example, based his paintings on staged photographs and projected those photographs onto canvas using a balopticon, a projector he described as “an evil, inartistic, habit-forming, lazy and vicious machine.”

He was so ashamed of using it that he would hide it whenever he heard someone approaching (Gurney 2009). The practice was widespread among illustrators of the time, although they were reluctant to admit it, much as some are today with AI.

With the arrival of desktop publishing in the mid-1980s, phototypesetters disappeared within a few years, while layout artists and graphic designers who adopted the new tools found themselves doing alone what had previously required several people.

None of these transitions was painless, and the fact that something similar has happened before does not make the current upheaval any less serious. If anything, it makes it more troubling.

Yet I cannot think of a single case in which refusing a technology protected those who resisted it for very long.

The Luddites opposed the use of machines to drive down wages and quality more than they opposed the machines themselves. Hobsbawm described their acts of destruction as “collective bargaining by riot,” a means of exerting pressure over wages in the absence of recognised trade unions (Hobsbawm 1952).

British print workers, whose resistance succeeded in slowing the introduction of electronic typesetting for years, were defeated between 1986 and 1987, after Murdoch moved his newspapers to Wapping and dismissed more than five thousand striking workers (TUC, n.d.).

The few victories have generally come through negotiation. During the 2023 strike, Hollywood screenwriters secured an agreement preventing AI from being imposed on writers, while leaving them free to use it with the production company’s consent. They also ensured that AI-generated material could not deprive writers who reworked it of credits or compensation (Writers Guild of America 2023).

Against Pitting Workers Against Each Other

Putting the pieces together, this is roughly the picture that emerges.

Clients are the winners on whom the studies most consistently agree. Emerging artists with some training who use these tools without hesitation probably benefit as well, as do professionals who have yet to establish themselves and adopt AI to produce more work in less time, provided they are willing to accept lower prices.

There are also artists, although no study has yet counted them, who stop supplying just one part of the creative process and instead become the authors of an entire project, such as a video game or an illustrated book. AI helps them in areas where they lack expertise, such as coding or music, while leaving visual direction in their hands.

An indirect indication comes from video games. Since January 2024, Steam, the leading online store for PC games, has accepted AI-created titles provided their developers disclose its use. Since then, debut games by small independent developers have increased faster than previous trends would have predicted, and among those declaring AI use, image generation is the most common application (Jo et al. 2025). The study, however, does not tell us how well these games sell.

Beginners who oppose AI lose twice over: the small commissions through which they once learned their trade disappear, and they refuse the assistance that could compensate for their lack of experience.

I suspect that established professionals who remain attached to a working method they refuse to change also lose out, unless they have a distinctive, recognisable style, although I have found no data on this.

The losses are concentrated primarily in standardised, interchangeable work, which, despite lacking originality, is not necessarily of poor quality. And which, in any case, has been and remains the foundation of many people’s incomes.

Stigma may offer a small advantage to individual illustrators working by hand, given that the label “made with AI” reduces views and the likelihood of being chosen. But the profession as a whole loses out, because stigma encourages people who use the tools to conceal that fact, distorting the data and dividing colleagues into opposing camps, while the greatest advantage goes to those commissioning the work.

Faced with this complex and nuanced situation, the demands that make the most sense to me are those that bring all professionals together, such as collective bargaining. After all, the most consistent signal across the studies is downward pressure on compensation, from the collapse in prices on Mihuashi to the painter delivering a week’s worth of work in a single day for the same pay.

In short, this is a deeply complex and uncertain situation, neither apocalyptic nor rosy. One that makes me want to say: illustrators of the world, unite, whether you use AI or not.


Bibliography