People usually search this after a queue that used to be reliable stops producing work. It is a frustrating position and the advice available is poor, because most pages answering it are affiliate lists ranking platforms the author never worked on.
I run the contributor side of a speech data company, so I compete with several of the platforms below and I am going to describe them fairly anyway. There is not much point writing a hit piece about companies that pay people properly. What is genuinely useful is understanding why the tasks dried up, what work is actually adjacent, and what each kind of work rewards, so you can pick on something better than a ranking.
Why the tasks went quiet
Almost always, this is scheduling rather than a judgement on you. Annotation work exists because a client commissioned a specific dataset. When that project finishes or pauses, the queue empties for everyone qualified on it, regardless of their approval record. A platform can be entirely healthy and still have nothing for you this month.
There are two other causes worth ruling out honestly. Some projects are restricted by country or by language, so a queue can be full for someone else and empty for you. And a falling approval rate does reduce what you are shown on most platforms, so if the drop coincided with a batch that went badly, that is worth checking before assuming it was scheduling.
What the big general platforms are good at
DataAnnotation.tech, Outlier, Appen, Mindrift, Toloka, CrowdGen and Prolific are real companies that pay real contributors. They are worth having in your set, and the frustration people express about them is nearly always about volume and communication rather than about being cheated.
Their strength is breadth. They run text-heavy work at scale, including writing and comparison tasks that pay comparatively well for English speakers, and they take on beginners readily. Their weakness is the same thing: because the work is broadly accessible, the pool of people who can do it is enormous, so both prices and task supply are volatile. If your only qualification is fluent English and care, you are competing with a very large number of people who also have that.
The four families of similar work
Almost everything adjacent to annotation falls into one of four groups, and knowing which one suits you matters more than knowing which brand is currently hiring.
- Text work: intent and sentiment classification, entity marking, ranking model answers, writing reference responses. Closest to what most people mean by data annotation. Largest pool, most volatile supply.
- Speech work: recording prompts in your own voice, transcribing audio, reviewing someone else's transcript, marking speaker turns, tagging accent or noise. Requires a specific language or accent. Smallest pool.
- Visual work: bounding boxes, segmentation, keypoints, object tracking across frames. Learnable in an afternoon, which is exactly why it is priced the way it is.
- Evaluation and red teaming: using a model as a user would and documenting where it fails. Rewards domain expertise and stubbornness, and increasingly pays better than labelling.
Speech work is the least crowded of them
I have an obvious interest here, so weigh it accordingly, but the structural point holds independently of who is making it. Visual and general text work can be learned quickly by nearly anyone, so supply of workers is effectively unlimited and prices track that. Speech work needs a language, dialect, or accent that you either have or do not, and no amount of practice adds one.
That scarcity is the whole reason the pay bands differ. If you speak a language outside the handful that dominate training data, or you have an accent that speech recognisers routinely mishandle, that is a commercial advantage rather than a footnote on your profile. It is also the reason a speech project can pay meaningfully more per task than a general labelling queue for work that is no harder.
What the alternatives pay
Comparing platforms on advertised hourly figures is a waste of time, because none of them pay by the hour and the figures circulating online are mostly invented. What can be compared honestly is a per-task price and an observed range. Ours, for the speech side: Tasks are priced individually and shown before you accept. Across current projects that works out to roughly 12 to 50 USD per hour, depending on the task type and how scarce your language is. It is an effective rate on per-task work, not a wage, and it is not promised for any particular week.
When you compare an alternative, ask for the same three things rather than a headline number: is the payout visible before you accept a task, what is the review and approval process, and what are the written payment terms and thresholds. A platform that answers all three plainly is a reasonable bet even if its rates are unremarkable. One that answers none of them is a bad bet at any advertised rate.
How many platforms to sign up for
Two or three, chosen so they do not overlap. The instinct after a dry spell is to sign up for everything, and it backfires: approval rate is tracked per platform and does not transfer, qualification tasks take unpaid time, and ten shallow profiles produce ten shallow records.
A better set is one broad text platform for baseline volume, one that specifically values whatever makes you scarce, and at most one more to experiment with. Keep each profile complete and current, because a brief can only match details you actually filled in. Languages at honest levels, accent and region, what you can record with, and what you are willing to do.
The one thing every real alternative has in common
Whatever you move to, the test is the same and it is not about brand reputation. Real platforms pay you and never charge you. No registration fee, no paid training, no starter kit, no deposit to release a withdrawal, at any point and under any name.
This matters more when you are switching than at any other time, because people looking for a replacement platform are exactly who fraudulent schemes target. If a site turns up promising better rates than the one that went quiet, and there is any payment required from you to begin, it is not an alternative. It is the thing the alternatives are being confused with.