Amazon Mechanical Turk shuts down on 30 September 2026, after 21 years online. The site paid ordinary people small amounts of money to do small computer jobs: label a photo, check an address, write a one-line summary. Amazon founder Jeff Bezos once called it “artificial artificial intelligence” — because humans quietly did the work computers could not.
Today, real AI can do many of those same jobs. So the obvious question is: did AI kill Mechanical Turk? The answer is more interesting than a simple yes, and it tells you a lot about how AI actually changes work.
What was Amazon Mechanical Turk?
Mechanical Turk was a website where companies posted tiny tasks and paid strangers to finish them. Amazon called each task a HIT — a Human Intelligence Task. Workers, known as Turkers, picked the ones they wanted. Most paid a few cents. A busy day might add up to a few dollars.
The site launched in 2005. TechSpot reports that more than 500,000 workers from over 190 countries signed up over its lifetime.
The name came from an 18th-century chess machine called The Turk. It looked like a clever robot. A human chess player was hidden inside the cabinet.
In one line: Mechanical Turk sold human work like a vending machine — post a task, get a human answer back in minutes.
Companies used it to build the raw material AI needs. If a computer can tell a cat from a dog, it is because thousands of people first tagged thousands of pictures by hand. That job is called data labeling — marking up examples so software can learn the pattern.
Why is Amazon Mechanical Turk shutting down?
Amazon has not given one clear reason. It said only that it regularly reviews its tools, programs and services. But three things clearly weakened Amazon Mechanical Turk: newer rivals paying much better rates for expert work, AI that can now handle simple tasks by itself, and growing doubt about whether the answers were still coming from humans.
The wind-down was already visible. The platform stopped taking new customers on 30 July 2026 and added no new features. Two related Amazon services — SageMaker Ground Truth and Amazon Augmented AI — close on the same day, 30 September 2026. Together, that means Amazon is walking away from the business of collecting human data entirely.
Were the humans secretly using AI?
Some were. In 2023, researchers Veniamin Veselovsky, Manoel Horta Ribeiro and Robert West at EPFL in Switzerland ran a test on Mechanical Turk. They asked workers to summarise short research abstracts. Then they checked typing patterns and used software that spots machine-written text. They estimated that 33% to 46% of the workers used a chatbot to do the job.
That number matters more than it looks. Buyers paid for human answers so they could train AI on them. If the answers came from a chatbot, they were training AI on AI. The information gets thinner each round, like a photocopy of a photocopy.
In one line: The platform built to sell human judgement started returning machine judgement, and buyers could not easily tell the difference.
Workers were not doing this for fun. They were paid per task. A tool that finishes a task in ten seconds instead of two minutes multiplies their hourly pay. The pricing model made the shortcut almost inevitable.
Did AI take these jobs, or did the work move somewhere else?
Mostly, the work moved. AI swallowed the cheap end — sorting simple images, tidying text, basic yes-or-no checks. But the demand for human input did not fall. It shifted to harder work: people with real expertise checking what AI produces, catching its mistakes, and showing it better answers.
That work now sits with a different set of companies. Scale AI, Surge AI, Mercor, Prolific, Toloka and Clickworker all hire people to train and test AI systems. The pay looks nothing like Mechanical Turk.
Mercor’s chief executive said in late 2025 that the company pays out more than $1.5 million a day to people training AI, with contractors averaging around $85 an hour. Doctors, lawyers, finance specialists and software engineers are in demand, because only they can judge whether an AI answer in their field is right or quietly wrong.
The wider market is growing, not shrinking. Mordor Intelligence values the AI data labeling market at about $2.32 billion in 2026, rising to roughly $6.53 billion by 2031.
In one line: The human layer under AI did not disappear — it got smaller, choosier and far better paid.
This is the same pattern showing up across many roles. As we covered in AI is not one job anymore, the work does not vanish in one clean cut. It splits apart, and the pieces that need human judgement become the valuable ones.
What happens to Mechanical Turk workers now?
They have to move to other platforms, and not all of them will earn the same. Prolific and CloudResearch run paid research studies and surveys, which is the closest match to old Mechanical Turk work. Clickworker and Toloka handle high-volume small tasks. Surge AI and Mercor pay far more, but screen for real professional skills.
Two honest points are worth making here.
- The switch is not automatic. A worker who spent years labelling images for cents cannot simply apply to review medical answers at $85 an hour. The new work asks for credentials most Turkers do not have.
- Reputation does not transfer. Turkers built approval ratings over years. That score is what unlocked better-paying tasks. On a new platform, everyone starts at zero again.
For anyone thinking about this kind of work, the useful lesson is to sell judgement, not clicks. Our guide to freelance AI skills clients actually pay for covers where the money currently sits.
Why this matters even if you never used the site
Mechanical Turk was quietly part of modern science. Psychology and social science researchers used it for years to recruit study participants cheaply. Thousands of published papers rest on data gathered there. When the platform closes, that particular pipeline ends, and researchers move to Prolific and similar services.
There is a bigger point too. Every AI tool you use rests on human work you never see. Someone marked up the examples. Someone rated which answer was better. Someone flagged the harmful reply so the system learned to refuse it.
The closing of Mechanical Turk does not mean that human layer is gone. It means the cheapest version of it is finished. AI ate the pennies-per-task work, and the money moved upstairs to people who know something the machine does not.
Frequently Asked Questions
When exactly does Amazon Mechanical Turk close?
The platform shuts down on 30 September 2026. It already stopped accepting new customers on 30 July 2026, so it has been running in maintenance mode since then with no new features.
Why is Amazon shutting down Mechanical Turk?
Amazon has not given a specific reason, saying only that it regularly evaluates its tools and services. Industry coverage points to stronger competitors, AI handling simple tasks, and doubts about data quality on the platform.
Did AI replace Mechanical Turk workers?
Partly. AI now does much of the simple work the site was built for. But demand for human input has grown at the expert end, where people check and correct AI answers for far higher pay.
Where can Mechanical Turk workers go instead?
Prolific and CloudResearch cover paid research studies. Clickworker and Toloka handle everyday small tasks. Surge AI, Scale AI and Mercor pay much more but screen for professional expertise in fields like medicine, law and software.
Is Amazon leaving the human data business completely?
It looks that way. SageMaker Ground Truth and Amazon Augmented AI, its two other human-data services, close on the same date as Mechanical Turk. That removes Amazon’s remaining tools for collecting human-labelled training data.
