The AI Gig Economy and How to Get Paid to Train AI

The phrase covers different work. An AI trainer might compare two answers, rate an output against a rubric, write an example, label an image, check a model’s response, or record a task that shows how people move through the real world.

The AI Hustle episode that sent me down this rabbit hole focused on first-person footage of everyday tasks. It called this “egocentric data” and named Figure’s Index and Contracted AI.

Other projects are less physical. RWS TrainAI lists online rating, data collection, annotation, and search or ad evaluation. Outlier describes ranking AI outputs and improving model responses, while DataAnnotation describes reviewing answers, checking accuracy, refining prompts, and rating outputs.

Figure’s Index listing says accepted creators may wait for access, receive recording equipment, and get paid by the minute. It does not give one universal rate.

Contracted AI describes project-based work, with task availability depending on project demand and quality. Its terms say rates can vary by project and tier and that a submitted clip can be rejected without pay.

Contracted AI also says that ownership of a recording transfers after it is accepted and paid for, and that contributors cannot then request deletion. Read the company’s privacy policy and terms before recording. That is one company’s policy, not a rule for every platform.

DoorDash Tasks is another example. DoorDash describes short activities for Dashers, including taking photos and recording everyday tasks or speech. Its announcement says pay is shown up front and based on effort and complexity.

Projects at DataAnnotation, Outlier, Alignerr, CrowdGen by Appen, and RWS TrainAI can involve reviewing answers, scoring model output, annotating data, or completing other evaluation tasks.

The assessment, task queue, quality rules, and rate depend on the role. CrowdGen says contributors can see a project’s rate before applying, while task availability varies by project.

Some work uses expertise in a field you already know. Mercor’s Education & Training Expert listing describes a talent pool for educators, not an open task queue or guaranteed job. It says a match can take a week or several months, depending on client demand. OneForma lists data collection and annotation projects with requirements that can depend on location, language, and the individual project.

Use these names as starting points, not promises of available work. A directory, a talent pool, and a live project are three different things. For example, AI Gig Jobs can help surface openings, but the platform or client listing is where you should confirm the role, location, pay, and terms.

AI Gig Jobs says on its About page that general listings are scraped daily across 15 platforms. Its advertising page explains that sponsored placements are labeled. Its salary report averages posted maximum hourly rates on listings that disclose hourly pay. Those are advertised amounts, not workers’ realized earnings.

That directory can be useful without being an eligibility oracle. On October 11, 2026, its DataAnnotation page for U.S. listings showed no listings, while DataAnnotation’s own application page listed the U.S. among accepted countries. An empty directory page is not proof that a platform rejects a state’s residents.

In my own search as of October 11, 2026, a number of sites did not let me access work as a New York State resident. At that time, I was going through DataAnnotation’s onboarding and did not yet know whether I would be approved or have projects available.

That is my experience, not a rule covering every platform. As of October 11, 2026, DataAnnotation’s signup page lists the U.S., but does not make a New York-specific promise.

Outlier’s location policy says customers can set extra restrictions for particular projects. One life-and-hobbies video study checked on October 11, 2026, excluded New York State and listed an end date of November 14, 2026. That is one project, not a company-wide rule.

DoorDash’s March 19, 2026 announcement says the Tasks pilot is available only in selected U.S. places and excludes New York City, among other locations. That is not the same as excluding the entire state. Separately, one Alignerr remote software engineering role sought people in or around New York City. That does not mean every Alignerr project is open throughout New York State.

Before spending time on an assessment, open the exact listing and look for state and country rules, language or credential requirements, and residency checks. If a platform verifies your location, do not misstate it.

Some listings publish hourly rates, but the label does not tell you how much work will be available or how long each task will take.

The household recording task in the hero image is illustrative, not a live job listing. It does not show a pay rate. Check each project’s official listing for its current pay and unit.

As of October 11, 2026, DataAnnotation’s Generalist listing said “up to $25–$50 per hour,” depending on task difficulty and specialization. Its FAQ says compensation depends on the qualification track and project access depends on expertise and performance. Those are the platform’s advertised figures, not a promise that a new applicant will receive that rate or a steady stream of tasks.

The Outlier video study checked on October 11, 2026, advertised $25 per recorded hour for a specific 25-hour project and excluded several states, including New York. It had a stated November 14 end date, so recheck the listing before relying on it.

AI Gig Jobs describes its averages as posted maximum hourly rates and labels them advertised rates, not earnings. Its salary report includes listings with disclosed hourly pay and averages their posted maximums. That can help you compare advertised ranges, but it is not an estimate of what a typical worker takes home.

A useful reality check is one WIRED reporter’s test of three robot-video gigs. The reporter said the total for the week was $21.55. That is one person’s experience, not typical weekly income or a forecast for another platform.

When you test a gig, count all the time you put in, not just the minutes the app says are paid. Include screening, training, recording, uploading, corrections, rejected work, and time waiting for a project. Your own first payout is a more useful signal than a large headline rate.

If a project asks you to film your home, workplace, family, or routine, check what the company wants recorded and what must stay out of frame. Look for consent rules, personally identifying information, ownership, storage, and deletion rights.

Contracted AI’s terms require creators to get necessary consent from people who appear in recordings. The terms also say accepted, paid clips transfer to the company and cannot be deleted on request. A specific Outlier recording study required contributors to avoid personally identifying information. Do not assume one company’s rules apply to another project.

Check for application fees too. RWS says there is no fee to join its TrainAI community, and Appen says it will not ask people to pay to apply or start an Appen job. Verify every application through the platform’s own official site.

The AI gig economy is another way to sell time, data, judgment, or expertise through projects. If you found this while searching for an “AI side hustle,” remember that the label does not guarantee hours or a paycheck. Some opportunities are project pools, not live task queues.

It is different from building your own AI-enabled service and selling that to clients. If that second route interests you more, ACC’s guide to building your first AI side project covers a different path.

If you want to try a platform gig, pick one lane, check your location, read the pay and data terms, and track what you actually earn after all your time is counted. Do not plan around a rate card or a project you have not been offered.