The new gig: humans hired to fix AI's bad drafts

A few times a week, Lisa opens a brief from a client, clicks the attachment, and sees a logo no human ever drew. The image is fuzzy, off-brand, full of melted letterforms. The client wants it print-ready by Friday. Lisa, a freelance graphic designer based in Spain, has become one of the people paid to make the AI version of a brand look like a brand.

"It's a shame we need to do it," she told The Guardian, "but someone has to clean up the mess."

She is one of a growing number of freelancers whose work now starts where the model's output ends. On Freelancer.com, the contract platform with more than 88 million users, job listings tagged with phrases like "correct AI", "AI hallucination" and "AI error" rose 87% to 10,760 globally between August 2025 and June 2026, according to internal data the company shared with The Guardian. Upwork's AI-remediation gigs rose 70% year over year, and Fiverr's keyword trend for related terms moved in the same direction.

For Lisa, the trend has already arrived. She estimates that 90% of her incoming logo and packaging requests in 2025 required cleaning up AI-generated content, and that work now accounts for 60% to 70% of her annual income.

The category has a name in freelancer circles: AI cleanup. It barely exists in mainstream coverage. It sits between the model and the finished product, in the gap that corporate AI announcements tend to skip. A designer might redraw a vector logo so it stops falling apart at print resolution. A video editor might re-cut a generated clip so the cuts actually match the script. A copywriter might rewrite an AI draft that has invented a quote, a statistic, or a product feature that does not exist. NBC News described the same work as humans being hired to make AI slop look less sloppy. Freelance Informer has called the gigs "AI F*ups." NDTV framed it as humans in demand to clean up AI's mess.

The work is concentrated. Graphic design leads, followed by video editing, proofreading and content writing. Pay is uneven. Clients who arrived by sending an AI prompt are often the same clients who treat cleanup as a small post-production step rather than a creative job, and they tend to bid the work low. Some outputs are so broken that freelancers throw the model output away and rebuild the asset from scratch, charging for what amounts to the original work while billing it as a fix. A few refuse briefs where polishing an AI version would mean attaching their name to an output they had no hand in making, a copyright concern that adds another layer to the pricing debate.

The freelancers who spoke to The Guardian are split on what the work feels like. Some find it tedious, repetitive, and creatively unfulfilling, a way to keep the rent paid while their own portfolio work gets pushed to evenings. Others enjoy the puzzle of reverse-engineering what a model meant to draw, and treat it as a new kind of craft. Lisa is closer to the first camp, which is why she used the word "soulless" in conversation with the reporter. The criticism is hers, and the people doing the work are the ones making it.

The bigger question sits underneath the income and the irritation. As the models improve, the cleanup layer narrows. The freelancers in the source piece are already asking how long the demand lasts. Some are turning jobs down, others are pivoting to "human-made" work as a selling point, and a few are raising their prices for cleanup as a premium service because it now requires undoing what an AI confidently produced. The Fiverr data point that did not make the headline, a measurable rise in keyword searches for AI-remediation terms, is the one to watch: it tells you what the buyer side expects, which is what the supply side will keep chasing.

For now, the gig exists, the platform data is up, and the freelancers are still taking the work. Whether cleanup becomes a durable labor market or a transitional rent that disappears when the next model release lands, the answer will be visible in the same search bar where the trend showed up.