AI remediation is one of the clearest examples of work that artificial intelligence is creating rather than destroying: people paid to check, correct and clean up what AI systems produce. In an essay published by The Conversation on 25 September 2026, Akhil Bhardwaj, an associate professor of strategy and organisation at the University of Bath School of Management, argues that the evidence so far points to limited large-scale job losses from AI, but a real change in how work is organised. The new jobs exist. The harder question he asks is whether anyone would want them.
This article looks at that question with the numbers behind it. It sets out where AI remediation work is growing, what it pays, what it does to the people doing it, and why the public sector and outsourcing markets may be next. It ends with what employers can do to make the new work worth having, because the answer to Bhardwaj’s question is not fixed. It depends on decisions organisations are making now.
Table of contents
- What AI Remediation Means
- The Evidence That AI Remediation Work Is Growing
- What AI Remediation Work Pays, and What It Costs
- Workslop: the Hidden AI Remediation Inside Every Office
- Why the Public Sector Faces an AI Remediation Bill
- The Subsidy Trap Behind Cheap AI Remediation Demand
- Will AI Remediation Become Offshore Piecework?
- Bargaining Power Decides Whether AI Remediation Is Rewarding
- Meaning, Not Just Money
- Signs Your Organisation Has an AI Remediation Problem
- How Employers Can Make AI Remediation Work Worth Doing
- AI Remediation FAQ
- References
What AI Remediation Means
Bhardwaj’s starting point is the sales pitch. AI systems are sold as tools that can write code, draft reports, analyse data and make routine decisions “faster and better than humans, all at negligible marginal cost”. In theory that frees people for higher-level thinking. In practice, he writes, AI agents produce “vast amounts of ‘almost right’ output”, with errors, hallucinations and “odd gaps in judgment that a real person has to catch.”
A job category with no single job title
The resulting work goes by several names: AI remediation, AI clean-up, post-editing, quality control and, less politely, AI janitor work. Bhardwaj notes that these AI remediation jobs “vary in their skill and remuneration levels”. At one end, AI companies are hiring experienced professionals in many fields to check whether their agents produce work of the right quality. At the other, freelancers take piecework fixing images, copy and video that an AI tool got nearly right.
Two kinds of AI remediation
It helps to separate two forms. The first is visible: jobs and gigs that exist specifically to fix AI output. The second is hidden inside existing roles, where staff spend part of each week checking colleagues’ AI-generated drafts. The second kind rarely appears in job statistics, but it may be the larger of the two.
The Evidence That AI Remediation Work Is Growing
The most concrete numbers come from freelance platforms, which can see demand shift in real time. The Guardian reported on 2 September 2026 on data shared by three of the biggest marketplaces.
| Platform | What was measured | Change | Period |
|---|---|---|---|
| Freelancer.com | Listings tagged “correct AI”, “AI hallucination” or “AI error” | Up 87% to 10,760 | August 2025 to June 2026 |
| Upwork | AI remediation gigs | Up 70% | Year on year |
| Fiverr | Searches for “AI cleanup” services | More than 20 times higher | 2023 to 2026 |
| Freelancer.com | Average bids per project | 54, up 8% | 2025 vs 2024 |
Where the demand sits
On Freelancer.com, which has more than 88 million users, the AI remediation listings were concentrated in graphic design, followed by video editing, proofreading and content writing. Working back from the Guardian’s figures, the platform had roughly 5,750 such listings in August 2025, rising to 10,760 by June 2026. Freelancer.com chief executive Matt Barrie said the work often comes from small firms and entrepreneurs who use AI for a first pass, then hit problems they lack the expertise to solve.
Growth alongside losses
AI remediation is growing in a market that is shrinking elsewhere. The Guardian cites a 2025 study in the journal Management Science which found that freelance listings for jobs exposed to automation fell 21% in the eight months after ChatGPT launched, and image-creation gigs dropped 17% after AI image generators arrived. More workers are also competing: the average Freelancer.com project drew 54 bids in 2025. So AI remediation is partly new demand and partly the only demand left in some creative categories.
The wider labour market
This fits the broad picture from the International Labour Organization. Its research brief on generative AI, drawing on surveys in Australia, Denmark, Germany, Korea, Kuwait, the UK and the US, finds that “large-scale job displacement remains limited”. Worker-reported time savings of a few per cent of working hours “have not yet translated into higher measured output, earnings or employment.” The main risks it identifies are growing inequality, fewer openings for younger workers, and “the reshaping of how work is organised”. AI remediation is a textbook case of that reshaping. For more on how the public sees this, see our coverage of the global fear of AI job loss.
What AI Remediation Work Pays, and What It Costs
Bhardwaj’s title asks how rewarding these jobs are. On the money alone, the evidence is mixed, and clients consistently underestimate the effort.
Clients expect quick and cheap
The freelancers the Guardian interviewed said clients often expect AI remediation to be fast and inexpensive, when it can take as much work as creating something original. Todd Van Linda, a Florida illustrator who charges $65 an hour, turned down one offer of about $500 to repair 13 to 15 AI-generated illustrations for a children’s book, on the assumption that each would take about 15 minutes. He says such jobs take anything from a few hours to days.
The arithmetic is simple. At $65 an hour, $500 buys about 7.7 hours of work. Spread across 13 to 15 images, that is 31 to 36 minutes each, double the client’s estimate but a fraction of the two to three hours per image that multimedia editor Nathan McConnell spent fixing AI-generated tarot cards before abandoning that project.
Income replaced, not added
For some freelancers AI remediation has replaced better work rather than adding to it. Kym Dunbar, an Australian writer and editor, told the Guardian that AI clean-up is now about 60% of her workload, while her monthly earnings from original writing have fallen to roughly a third of what she made during the pandemic. Lisa, a graphic designer in Spain, said that by 2025 about 90% of her logo and packaging requests involved cleaning up AI output, accounting for 60% to 70% of her income. She began turning them down.
Rates that hold, and rates that slip
The pay picture is not uniformly bleak. While he took AI remediation jobs, Van Linda would not cut his rate, which he says “doesn’t go down from there”. He has since stopped taking them. McConnell’s AI remediation requests grew from two or three projects in 2023 to eight to ten so far this year. Freelancers with a reputation can charge for the skill involved. Those without one face lowballing, which is where Bhardwaj’s concern about “endless, underpaid digital piecework” begins.
Most AI remediation is not advertised as a job at all. It happens inside existing roles, when someone receives an AI-generated report, email or analysis that looks finished and is not. Researchers at BetterUp Labs and Stanford named this “workslop”, defined as “AI-generated work content that masquerades as good work, but lacks the substance to meaningfully advance a given task.”
| Workslop measure | Finding |
|---|---|
| Survey | 1,150 full-time US workers (BetterUp and Stanford, 2025) |
| Received workslop in the last month | 40% |
| Share of received content that qualifies | About 15% on average |
| Time spent dealing with each incident | 1 hour 56 minutes |
| Invisible cost per affected employee | About $186 a month, or $2,232 a year |
| Cost for a 10,000-person organisation | About $9 million a year |
| Changed view of the sender | Roughly half see them as less creative, capable and reliable |
The figures hang together. $9 million across 10,000 workers is $900 each per year, close to 40% of employees each losing $2,232. That is AI remediation work nobody budgeted for.
How it feels to receive it
Bhardwaj quotes a retail director from the Harvard Business Review report who described spending “more time following up on the information [provided by AI] and checking it with my own research”, then more time “setting up meetings with other supervisors to address the issue.” This is AI remediation as a tax on the most experienced staff. He also notes that workslop “can also undermine trust in colleagues who are using AI agents”, which matches the survey’s finding that about one in three recipients are less willing to work with the sender afterwards.
Why the Public Sector Faces an AI Remediation Bill
Bhardwaj warns that governments risk being drawn into the same “AI repair” logic. The UK government, among many others, is experimenting with AI for summarising consultation responses, drafting correspondence and analysing sentiment in citizen feedback. Its guidance for civil servants encourages cautious use and warns that outputs can be misleading and must be independently verified.
Verification creates posts
If public bodies start producing large volumes of AI-generated letters and analysis, Bhardwaj argues, they will need new positions whose main job is to monitor, check and correct that output before it reaches citizens or shapes policy. That is AI remediation as a permanent line in the public payroll. His question is pointed: “How much of the apparent AI efficiency gain will be consumed by people silently tidying up after systems that are sold as automatic?”
The Dutch warning
He points to the Netherlands, where tax authorities used algorithmic risk scores to flag families for childcare benefit fraud. Tens of thousands of parents, many from minority backgrounds, were wrongly accused, forced to repay benefits they did not owe and pushed into debt or bankruptcy. The lesson he draws is that the failure was not only in the output but in “the absence of robust human supervision”. Any public-sector AI rollout, he argues, must invest heavily in exactly that supervision. Done properly, AI remediation in government is not a cost to minimise. It is the safeguard.
The Subsidy Trap Behind Cheap AI Remediation Demand
One of Bhardwaj’s sharpest points is economic. The tokens that generate AI output, the basic units of text that natural language processing systems such as large language models work with, are “heavily subsidised by debt-laden AI companies”, he writes. “The human time needed to check and fix it is not.”
Below-cost pricing builds dependence
Many AI companies, he argues, are deliberately pricing access below cost to build dependence. Once organisations have reorganised workflows around a particular model, the provider can raise prices, cut free tiers or change terms. So today’s cost-benefit analysis may look very different later. A process that looks cheap because generation is subsidised and AI remediation is absorbed by existing staff may become expensive on both sides at once.
Budget for the checking, not just the generating
For businesses, this is an argument for honest accounting. Any business case for automation should include the human checking time, priced at the rate of the people who actually do it, and should test what happens if model prices rise. That is part of a sound AI strategy, and it is often missing from vendor-led pilots.
Will AI Remediation Become Offshore Piecework?
The “hidden humans” behind AI are not new. Bhardwaj notes that chatbot moderation, data labelling and image annotation have been outsourced to low-wage countries for years. That work has taken a substantial mental toll on some workers, who review graphic violence and self-harm, often without legal protection.
A second wave of outsourcing
If workslop becomes normal, he suggests, a second wave may follow: routine checking and correction of AI-generated text shipped to offshore centres, “packaged as quality control, but experienced on the ground as endless, underpaid digital piecework.” AI remediation would then follow the same path as content moderation, with the least visible workers carrying the most tedious load.
Translation shows the pattern
Translation already works this way. In many agencies the standard process is machine translation plus “post-editing”: the AI does a first pass and freelance translators are paid less to fix it. Research cited by Bhardwaj finds post-editing is often paid worse than traditional translation, needs similar or greater cognitive effort, and is widely seen as tedious and deskilling. Those doing it also fear that even this role will eventually be automated.
| Type of AI remediation work | Typical task | Skill level | What the evidence says about reward |
|---|---|---|---|
| Expert quality control for AI companies | Checking whether agents’ work meets professional standards | High | Hiring of experienced professionals, pay varies by field |
| Freelance creative clean-up | Fixing AI images, video, copy and books | Medium to high | Demand up sharply; clients often lowball; some freelancers quit |
| Translation post-editing | Correcting machine translation | High | Often paid worse than translation, similar or greater effort |
| Internal workslop checking | Reviewing colleagues’ AI drafts | Varies | Unpaid, unbudgeted, damages trust |
| Moderation and labelling | Reviewing and tagging content for AI systems | Low to medium | Low wages, offshore, documented mental toll |
| Public-sector verification | Checking AI letters and analysis before release | Medium to high | Essential safeguard, adds cost to “automatic” systems |
Bargaining Power Decides Whether AI Remediation Is Rewarding
Bhardwaj’s most useful example comes from Hollywood. In 2023 the Writers Guild of America went on strike partly because writers feared studios would use AI to generate first drafts and hire them to polish the results cheaply.
What the writers won
The agreement set guardrails. AI cannot be credited as a writer, cannot be forced on writers, and cannot be used in ways that undercut human pay and recognition. In Bhardwaj’s words, the union “fought to stop a slide into AI-janitor work”, where creative professionals would spend their time tidying up after algorithms rather than originating stories. The result shows that AI remediation is not inevitably low-status. Its terms are negotiated.
Who lacks that power
Translators, freelancers in fields such as video games and many other workers have far less collective bargaining power, which makes it easier for AI-driven business models to “quietly make repair work as the new normal”, as Bhardwaj puts it. The freelancers in the Guardian’s reporting are each negotiating alone, one client at a time. Some hold their rates. Others accept whatever the platform market will bear.
Meaning, Not Just Money
Bhardwaj ends on a question economists rarely measure. The problem “is not only economic”, he writes, “it is also about meaning.” Most people want to feel their work contributes something that would not exist without them. AI remediation, done badly, removes exactly that.
The freelancers disagree with each other
The Guardian’s interviews show the split. Lisa called the work “just so soulless” and now focuses on human-made designs. Van Linda, who stopped taking clean-up gigs late last year, called it “extremely distasteful”. Dunbar, by contrast, finds it “interesting” and “challenging” and takes satisfaction in turning a weak draft into polished prose. The same task can feel like craft or like drudgery, depending on control, pay and respect.
How long the work will last
The freelancers also disagree on the future. Dunbar thinks much of the humanising work could dry up within five to ten years as models improve. McConnell argues that AI will still need to be “babysat” to catch errors an untrained eye misses. Bhardwaj’s view is that for as long as powerful AI stays cheap and subsidised, “more and more people will be diverted to cleaning up the mess these systems inevitably create.” For a related look at early-career effects, see our piece on computer science graduates and AI skills.
Signs Your Organisation Has an AI Remediation Problem
Because most AI remediation is invisible, many organisations only notice it when deadlines slip or trust breaks down. A few warning signs show up early, long before anyone calls it a problem.
Review cycles are getting longer, not shorter
If AI was introduced to speed up reports, proposals or code, but the time from first draft to sign-off has grown, the saving is being spent on checking. Measure the whole cycle, not just the drafting step. The workslop survey’s figure of almost two hours per incident shows how quickly the checking time adds up.
Senior staff are doing the fixing
AI remediation tends to flow upwards. The people best able to spot a plausible error are the most experienced, so they absorb the corrections. If managers and specialists report spending more time editing and less time on their own work, the organisation is paying senior rates for AI remediation it never planned.
Nobody owns the output
When an AI-generated document goes wrong and nobody is sure who wrote it, who checked it or who approved it, accountability has leaked. Every AI-assisted deliverable should have a named human owner who signs it off.
Suppliers quote less but deliver later
External suppliers are adopting AI too. Lower quotes paired with more rounds of revision can mean your own staff are doing the supplier’s AI remediation for free. Ask vendors how they use AI and who checks the output before it reaches you.
How Employers Can Make AI Remediation Work Worth Doing
Nothing in the evidence says AI remediation must be poorly paid or dispiriting. It says it tends to become so when nobody plans for it. Employers deploying AI can make different choices.
Name it, measure it, pay for it
Treat AI remediation as a real task. Track how much time staff spend checking AI output, put it in project estimates, and reward it in performance reviews. Hidden work is the work most likely to be undervalued.
Put experts where the risk is
Route AI output that affects customers, money or safety to people with the expertise to judge it, and give them the authority to reject it. This is what the Dutch case lacked, and it is what AI companies are paying experienced professionals to do.
Keep original work in the mix
Jobs that are entirely correction lose meaning quickly. Mixing AI remediation with original work, and giving people a say in which tasks AI handles, keeps skills alive and keeps staff. Good change management matters here as much as the technology.
Set rules for your own AI use
Agree internal standards before workslop spreads: when AI drafts must be disclosed, who checks them, and what “finished” means. Our guide to how AI is reshaping hiring shows what happens when both sides of a process automate without rules.
AI Remediation FAQ
What is AI remediation?
AI remediation is work that checks, corrects or cleans up output produced by AI systems, from fixing AI-generated images and copy to verifying reports, code and analysis before they are used.
Is AI remediation work growing?
Yes. Freelancer.com listings tagged for correcting AI errors rose 87% to 10,760 between August 2025 and June 2026, Upwork’s AI remediation gigs rose 70% year on year, and Fiverr searches for “AI cleanup” grew more than 20-fold between 2023 and 2026.
Does AI remediation pay well?
It varies. Experienced specialists and freelancers with strong reputations can hold their rates, but clients often expect clean-up to be quick and cheap, and translation post-editing is often paid worse than translation.
What is workslop?
Workslop is AI-generated work that looks finished but lacks the substance to move a task forward. A BetterUp and Stanford survey found 40% of US workers received some in the previous month.
Is AI causing mass job losses?
Not yet, according to the International Labour Organization’s review of the evidence, which finds large-scale displacement limited so far but significant changes in how work is organised.
Who wrote the original article?
Akhil Bhardwaj, associate professor of strategy and organisation at the University of Bath School of Management, in The Conversation on 25 September 2026.
References
The Conversation: AI is creating jobs as well as erasing them, but how rewarding are they?
The Guardian: Freelancers are getting buried with ‘soulless’ AI slop cleanup
Harvard Business Review: AI-Generated “Workslop” Is Destroying Productivity
CNBC: AI-generated workslop is destroying productivity and teams, researchers say
GOV.UK: Guidance to civil servants on use of generative AI
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
