The platforms this site lists are real companies with public role postings and a payment history you can check. They are not the only names you will meet when you search for AI training work, and some operators look almost identical to a real one until you are a week into unpaid “assessments”.
This page lists the warning signs that recur in contractors’ own accounts. It is not a blacklist of companies. Use it to spot a bad operator before you have spent hours on it.
Rejection is not a scam signal
Most “is this a scam” questions come from this. Being rejected after an interview, having an application sit unreviewed for weeks, or being unable to share your project name or your pay with anyone, including strangers online asking for proof, are all normal features of legitimate contract work, not evidence of fraud.
A platform’s NDA on project specifics protects its client’s data, not the platform’s reputation. If someone can’t show you a payment screenshot with a project name attached, that’s usually the NDA doing its job. Judge a platform on what it does, regardless of what a rejected applicant believes about it.
The math doesn’t add up
The clearest, most checkable red flag is internally inconsistent pay. If a posting advertises a per-task rate and a completion time, multiply them out: $6.25 a task at a 2–4 minute completion time works out to $93–$187 an hour, not the $25 an hour the same posting claims. Real numbers are consistent with each other. A posting that quotes two figures that can’t both be true is using the bigger one to get you in the door and the smaller one to actually pay you.
The check takes thirty seconds. Run it on every posting that states both a task rate and an hourly equivalent.
The rate drops after you’re in
A related pattern: an advertised rate that gets cut once you’ve started, sometimes more than once. Workers have reported per-task pay cut mid-project: first one reduction, then another, with the original rate never coming back. The same shape shows up as a media-review rate that starts at $100+ an hour for new hires and is quietly down to under half that within a week.
A single renegotiation with notice and a reason is normal business. A silent cut, especially more than one, is not something a legitimate platform has any reason to do once you’ve already started the work.
Pay gated behind vague “acceptance”
Some postings pay only if your submitted work is “accepted,” with no clear rubric for what accepted means and no path to find out why something wasn’t. Quality control is reasonable (every platform this site lists has it), but it should come with a stated standard you can check your work against, not a black box that happens to reject enough submissions to avoid paying for the time spent making them. If the quality bar is never written down anywhere, that absence is the point.
Hours of unpaid “training” before any real task
Multi-hour unpaid assessments, training modules or “certification exams” are common across this space, and most of them are legitimate screening the way a work sample is for any job. The signal to watch for is scale: assessments that run four to six hours, repeat across multiple roles at the same company, and never convert into paid work for a meaningful share of people who complete them. At that point the “assessment” isn’t screening for the job. Supplying the free labour is the job.
If a company’s assessment output looks suspiciously like the actual product it sells (translated text, labelled images, transcribed audio), be more skeptical, not less, that the exam is real screening.
The job you’re hired for isn’t the job you’re offered
A few contractors have described being accepted into a role (a general annotation or evaluation position) and then the only task that materialises is something else entirely, like submitting selfies or biometric data for a token payment. Being screened into a stated role and then routed to a completely different, more personal one is worth walking away from, regardless of the number attached.
A recruitment post with no name attached
Outside the platforms’ own job boards, you’ll see posts like this circulating on job-focused subreddits and Facebook groups: “Hiring: AI Training & Evaluation Roles, $20–50/hr, remote, first-world countries only,” with no employer named, no link to an application page, and a request to comment your location before anything else is discussed. This is a lead-generation pattern, not a job posting. The same account often reposts the identical ad on a schedule, sometimes mixed in with generic “advice” posts to look established. A real opening names the company and links to where you actually apply.
Email pretending to be a platform you already work for
The patterns above target people who have not been hired yet. There is a second kind aimed at people who have, and it is harder to spot, because by then a message about your pay is exactly what you are expecting.
Mercor’s documentation names the domains its mail uses: @mercor.com for staff, and anonymised @mercor.expert addresses for contractors, which can only send to and receive from @mercor.com. Check the sender’s domain on anything about your pay. The same test works on every platform: learn what a real message looks like while nothing is wrong, so a fake one has something to fail against.
Be most suspicious of anything that asks you to move money, re-enter bank details, or sign in through a link to set up a payment you were not told about. Real payment changes on these platforms happen inside the account you already log into, not through a link in a mail.
What a real platform looks like
Every platform this site lists posts roles on its own domain, with a specific title, a specific application flow, and no fee at any step. None of them ask for payment to “process,” “verify” or “unlock” your account, and none pay only if your work is subjectively “accepted” against no stated standard. Their screening steps are documented, even if unevenly: a timed AI interview, a domain assessment, a coding test, the kind of thing you’d expect to see written down somewhere before you sit it.
A platform missing from this list is not therefore fake. Much of the wider market is ordinary contract work with ordinary downsides: uneven hours, slow reviews, and pay that works out lower than advertised once you count the unpaid parts. The warning signs above describe the minority that is not, so check them before you give a new name a week of your time.
Before you commit real hours
- Do the rate math yourself if a posting states both a per-task fee and an hourly equivalent
- Never pay a fee, deposit, or “training cost” to apply, onboard, or get paid
- Be more cautious the longer an unpaid assessment runs and the more it resembles the company’s actual product
- Treat a rate cut after you’ve started as a reason to stop, not a reason to catch up
- Search the company’s name alongside “reddit” or “trustpilot” before investing more than an hour
Each platform page sets out how that platform screens and pays, with links to its own documentation, so you can check it yourself. When you are ready, browse the open roles from all of them.