The Future of Microwork: How AI Is Changing Gig Work

On September 30, 2026, Amazon shut down Mechanical Turk, one of the first and best-known microwork platforms, after a 21-year run. At its peak, it had more than 500,000 registered workers in 190 countries (Source: Gadget Review, 2026). Officially, Amazon said only that it made the decision after "evaluating its programs, tools, and services" (Source: Let's Data Science, 2026). Observers link the closure to AI: many of the simple tasks MTurk workers used to do are now handled by models. At first glance, this looks like the end of the format. But the same AI systems that took over part of the work have created demand for a different kind: they need people who record their voices, take photos, rate chatbot answers and check whether a model got something wrong.

The takeaway for workers is neither "microwork is dying" nor "everything will be fine." The mix of tasks is changing, and the people who understand that sooner than others will come out ahead. Below: what has already gone to machines, what has appeared in its place, how much it pays, which risks to keep in mind and how to prepare for what microwork will look like in a couple of years.

Which Microtasks AI Has Already Taken Over, and Why Those

AI is not replacing "simple work" in general but a specific type of task: the kind with one correct answer, where the input data is already available online. Recognizing a cat in a photo, retyping a receipt, transcribing clear English speech — modern models do these faster and cheaper than people. These were exactly the tasks that kept MTurk and similar platforms going in the 2010s. The table below is an industry estimate, not data from a specific study: how far automation goes depends heavily on the language and the quality of the source data.

There is another reason clients moved away from older platforms. In 2023, EPFL researchers asked MTurk workers to summarize medical research abstracts and found that between 33% and 46% of them used ChatGPT (Source: MIT Technology Review, 2023). It is a preprint that has not been peer-reviewed, with a small sample of 44 people, so it can't be generalized to all workers. But the conclusion for clients is simple: if a model answers instead of a person, there is no point paying for that work. Since then, clients have become much stricter about checking that the work was actually done by a human.

For workers, this means one thing: if a task can be done entirely through ChatGPT, it will most likely stop being posted soon. If it needs your voice, your camera, your location or your judgment, it will stay.

Which Tasks AI Has Created

To learn, models need data that doesn't exist online and judgments only a person can give. That has produced a new class of microtasks. They usually take a few minutes and pay more than clicks, but they also demand more attention.

1. Voice recording. Reading phrases aloud or describing a picture or a situation in your native language. Accents and languages with little data online are especially valuable: Swahili, Bengali, Tagalog, Arabic dialects. We explained why in our article on low-resource languages in AI.
2. Real-world photos and videos. A shop window in your city, a receipt, a handwritten note, first-person video in a kitchen. Stock photos don't work for this: models need ordinary conditions, bad lighting and many different countries.
3. Rating AI answers. Comparing two chatbot answers and picking the better one, flagging a mistake or rudeness, rating a translation. This is the foundation of RLHF, without which chatbots would not have become useful.
4. Checking AI's work. Confirming that an address found by an agent is correct, that a shop is still open, that a machine translation doesn't distort the meaning. The more companies rely on AI, the more they need this kind of human review.
5. Testing AI products. Trying out a new chatbot app and reporting where it misunderstood a request or gave a strange answer.

Who is entering this market also shows it is not a passing trend. In October 2025, Uber announced it would add a "digital tasks" section to its app for some drivers in the US, with the launch promised later that fall. The tasks: uploading restaurant menus and recording their own voice in different languages. Before that, the format was tested in India (Source: The Philadelphia Inquirer, 2025). In essence, these are the same microtasks, just of a new kind.

For more on what these tasks look like and where to start, see our guide to AI training tasks.

The Microwork Market in Numbers: Shrinking or Growing?

The answer depends on which part of the market you look at. The World Bank estimates that between 154 million and 435 million people do online gig work, or 4.4% to 12.5% of the global workforce. That figure covers all kinds of online platform work, including freelancing, not just microtasks. The upper bound also includes people for whom it is a side gig rather than their main income. Low- and middle-income countries account for 40% of the traffic to these platforms (Source: World Bank, "Working Without Borders," 2023).

Meanwhile, the AI data labeling market is growing. Technavio forecasts that between 2025 and 2029 it will grow by about $1.4 billion, at an average annual rate of around 21% (Source: Technavio, 2025). There is no contradiction here: money is flowing from old tasks to new ones. The total amount of work for people isn't disappearing, but it looks different.

Earnings are less clear-cut. Pay on MTurk was low even before: in a 2016 Pew Research survey, 52% of US workers said they earned less than $5 an hour, and only 8% made $8 or more (Source: Pew Research Center, 2016). New AI tasks usually pay more per task but also take longer. On RapidWorkers, for example, typical pay per task is $0.10–1, and the rate is set by the employer. What you actually earn per hour depends on how many tasks are available in your country and how often your work is approved. We did an honest time-based calculation in our realistic guide to earning with microtasks.

Here is what that looks like in practice. The numbers are illustrative, to show the logic of the calculation, not average rates on any platform:

The table shows the main point: the best-paying task is not always the most profitable. And if a photo or recording is rejected because you skimmed the instructions, your real hourly rate drops to zero.

Why These Changes Are Risky for Workers

The new tasks pay better, but the transition comes with real downsides. Articles about "the automation of the future" rarely mention them, yet they have the biggest effect on income.

Platforms close and change

The MTurk shutdown showed that even a platform run by one of the largest companies offers no guarantees. Registration for new workers stopped on July 30, and two months later the platform closed for good. Amazon also removed MTurk workers from its labeling services SageMaker Ground Truth and Augmented AI: the services themselves keep running, but clients had to find other workers (Source: Gadget Review, 2026). Workers who only worked on MTurk were left without the history and rating they had built up over years. Reputation doesn't transfer between platforms, so depending on a single one is the most costly risk. You can see how we compared platforms after the MTurk shutdown in RapidWorkers vs MTurk.

ChatGPT answers get you banned

Clients pay specifically for human data, so they add control questions, watch for copy-pasting and compare answers across different people. An answer generated by AI is easier to spot today than it seems. The result is a rejected payment and sometimes a ban. You can use AI where the instructions allow it, for example to translate the task itself. But the answer has to be yours.

Personal data becomes the product

A voice recording or a photo of your face is no longer a click; it is biometric data. Before taking these tasks, read who the client is, what the data is for and how long it will be stored. If the description is vague and the pay is too high for such a task, it is better to skip it. Honest clients never ask for ID documents, card details or passwords.

Location matters more than before

AI projects often need people from specific countries with specific languages. That is a plus for workers from countries with little data, and a minus for those where there are already plenty of workers. Trying to get around a restriction with a VPN almost always ends in a ban: the client needs data from that country, not just a person with the right IP address.

Unpaid time is growing

In its surveys of about 3,500 crowdworkers, the International Labour Organization singled out time that isn't paid: searching for tasks, qualification tests, messaging clients (Source: ILO). With AI tasks there is more of this time: instructions are longer and tests are stricter. What to count is not the pay per task but the money for the whole hour you spent on the platform.

How to Tell a Real AI Task From a Scam

Alongside real AI tasks, fake ones have appeared. "AI training" sounds modern and trustworthy, so scammers use it as bait. Check a task against these signs:

If at least two signs from the right-hand column match, skip the task and report it to the platform's support. We covered more scam warning signs in our guide to earning with microtasks.

How Workers Can Prepare for Microwork With AI

You don't need special courses for this. You need to understand what about you is valuable to an AI project and get your workflow in order. Here is what opens up access to the new tasks:

And what to do in practice:

1. Sign up on 2–3 platforms. The task flow on any single one is never constant, and the MTurk shutdown showed that the platform itself can disappear.
2. Fill in your profile honestly. Clients select workers for AI projects by country, languages and devices.
3. Start with simple labeling. Our data labeling guide for beginners will help you understand the logic of these tasks before you move on to better-paid ones.
4. Read the instructions to the end. In AI tasks, the requirements for lighting, background, recording length or answer format are strict, and work gets rejected over a single detail.
5. Count your earnings per hour. A $1 task that takes 20 minutes pays less than a $0.20 task that takes 2 minutes.
6. Don't let AI stand in for you. Your value to the client is that you are a person. That is exactly what AI cannot copy.

Who will come out ahead? People for whom microwork is a side gig of a few hours a week and who are willing to take slightly more complex tasks. Those who rely on fast clicks in high volume will have a harder time: there will be fewer of those tasks and more competition for them.

If You Worked on MTurk: What to Do Now

The platform is closed, but not all the money and deadlines are behind you. According to Gadget Review, approved earnings are paid out on the normal schedule, and clients have until October 30 to approve or reject submitted work. Anything not reviewed by then is approved automatically (Source: Gadget Review, 2026). What to do:

1. Check your payment settings in the Worker Portal. If your payout method is outdated, the money may get stuck.
2. Save your history. A screenshot of your number of completed tasks and approval rate won't transfer your rating, but it can serve as proof of experience, for example if a platform or client asks about it.
3. Write down which tasks you did best. Surveys, transcription, labeling — this determines which platform will suit you best.
4. Sign up on several platforms at once. You will have to build a rating from scratch on each one, and the sooner you start, the sooner better-paid tasks open up.

We compared the remaining platforms and how they differ in RapidWorkers vs Microworkers and RapidWorkers vs MTurk.

What Microwork Will Look Like in 2–3 Years

Nobody can give an exact forecast here. Below is our estimate, based on the trends in this article: the MTurk shutdown, the growth of the labeling market and new players such as Uber. It is not research data:

1. Fewer tasks, but each one pays more. Tasks that take a few seconds will keep moving to AI, while tasks that take a few minutes and involve real-world data will grow.
2. More checks that you are human. Platforms and clients will get stricter about filtering out bots and generated answers.
3. Country and language will matter more. Models need data from all over the world, and people from Africa, South Asia and Latin America will be in even greater demand.
4. Microtasks will show up in unexpected places. The Uber example shows these tasks will be built into apps where people already earn money.

Frequently Asked Questions

Will AI completely replace microtask workers?

No. AI replaces tasks with one correct answer, but it also creates demand for human data and judgments. Without people, models have nowhere to get new voices, photos from different countries or checks of their own answers.

Why did Amazon Mechanical Turk shut down?

Officially, Amazon only said it made the decision after evaluating its programs and services. Observers link the closure to AI: transcription, image labeling and some surveys are now done by models, and the labeling market has moved toward services with strict quality control. The platform stopped operating on September 30, 2026.

Is it worth starting to earn with microtasks now?

If you want a side gig with a flexible schedule, yes, especially with AI tasks. If you need a stable main income, no: the task flow depends on clients and on your country.

Which microtasks are least at risk from AI?

The ones that need something physical or personal: recording your own voice, taking photos in your city, giving your opinion in a survey, rating an AI answer, checking a place in person. These can't be generated without losing their point.

Can I use ChatGPT to complete microtasks?

Only if the instructions explicitly allow it. In most AI tasks, the client pays specifically for a human answer, and generated text will lead to a rejected payment.

What This Means for Workers

Microwork isn't disappearing, but the days when you could earn from thousands of identical clicks are ending. What's valuable now is what only a person has: a voice, a language, a location, experience and attention to detail. If you work on several platforms, read the instructions and aren't put off by tasks that take a few minutes, these changes are more likely to work in your favor.

On RapidWorkers, workers from 200+ countries already do regular microtasks and tasks from the AI Tasks section (currently in beta). If you want to give it a try, signing up as a worker takes a couple of minutes.