Moving from MTurk to 50heads

Mechanical Turk has been the default place to get a quick human answer for nearly twenty years. It is flexible, cheap at the bottom end and familiar. It is also, for the kind of question we care about, hard work: you design a HIT, set qualifications, guess a fair reward, wait, then clean the data because a share of it came from scripts, farms or people clicking through.

50heads does one narrower thing. You ask one short question with a few options, and fifty verified people answer it on their phones in about ten minutes. If that is most of what you used MTurk for, this guide is for you.

What maps to what

The price, side by side

On MTurk you set the reward and Amazon adds its commission: 20%, rising to 40% on HITs with ten or more assignments. A fifty-assignment HIT at 10 cents a response costs $7.00 before you have thrown anything away.

On 50heads a Tier 1 answer is 12p, and fifty of them cost £6.00. The price covers verification, the head's pay and our margin. Heads get 70% of it. You see the total and the time before you post, and if fewer heads answer than you asked for, the rest goes straight back to your balance.

What you stop doing

Screening for bots. Every answer on 50heads is signed by the secure hardware in the phone that gave it. Emulators and modified apps can't answer at all. Tier 1 heads have one real mobile number each; Tier 2 heads have passed an ID and live-face check. You don't need attention checks, and you shouldn't add them: they make the question longer, and longer questions cost more.

Approving and rejecting. We run known-answer questions and timing checks on our side. If an answer is thrown out after it counted, you get the credit back automatically.

Writing reward maths. The price is worked out from the tier, how long the question is to read and the kind of answer. The estimate is free and never spends anything.

What you need to change

Most MTurk HITs that fit 50heads are one of four shapes:

1. Pick one of several. Names, headlines, thumbnails. This is the default question type.
2. A or B with images. Two pictures, one tap.
3. A 1–5 rating. Use sparingly; a choice usually tells you more.
4. A short free-text reason. Available from Tier 2, up to 200 characters.

Longer tasks (transcription, tagging hundreds of items, multi-page surveys) don't fit. Keep MTurk or a research panel for those.

Keep questions short. Heads answer in about five seconds, on a phone, between other things. If your HIT has a paragraph of instructions, the question isn't ready yet. Our guide to writing questions that get clear answers covers the rules we use ourselves.

Doing it in an afternoon

1. Sign in and add credits. Packs start at £20.
2. Take your most common HIT and rewrite it as one question with two to eight options. Run it through the estimator to see the price and the time.
3. Ask it to fifty Tier 1 heads and compare the result with your last MTurk batch.
4. If you call MTurk from code, swap the call for the 50heads API or add the MCP server to your agent. There is a quick start with curl, TypeScript and Python.

curl https://api.50heads.com/v1/billing/estimate \
  -H "Content-Type: application/json" \
  -d '{"draft":{"type":"single_choice","text":"Which name would you trust with your savings?","options":[{"label":"Northway"},{"label":"Keel"}],"n":50,"tier":1}}'

When MTurk is still the better tool

If you need someone to do work rather than give a judgment, or you need hundreds of steps from the same person, or you are running an academic study with its own consent flow, a general-purpose marketplace or a research panel will serve you better. We would rather tell you that now than have you find out halfway through.

For everything that is really one question with an answer you'll act on this afternoon, fifty heads are better than one.