Participants

Can I get paid to help train AI?

By the Inflight team · Updated

Yes. Ordinary people get paid to help train AI in three ways: labelling data, evaluating model answers, and taking recorded product tests whose de-identified recordings teach models how real users behave. Inflight pays US adults cash to PayPal for recorded website and app tests; with your separate consent, a de-identified recording also trains and evaluates AI models that behave like real users. Your name, email and raw audio are never part of released data.

The three kinds of paid AI work

KindWhat you doWhere
Data labellingTag images, text or audio to a specLabelling marketplaces
Model evaluationRate or compare AI answersProlific's AI studies, labs' own panels
Behaviour recordingUse a website or app normally, thinking aloudInflight

The first two reward speed and accuracy against a rubric. The third just needs you to be yourself, because what is wanted is how a real person actually behaves on a product, hesitation, wrong turns and giving up included.

What Inflight does with a recording

A study recording contains your voice, the study tab, and what you did on the product, with the time of each action. It is used to understand how people use products, and, with your study consent, to train and evaluate AI models that behave like real users. Models learn from behaviour, not from who you are. Your name, email, PayPal, identity results and the raw audio are never part of released data, and released data carries no identifier that links back to you. Password fields are masked and anything that looks like personal information you typed is removed before release.

You can withdraw a recording after the fact by email; it is removed from future training and benchmark sets and you receive a receipt. The full text is the recording and model-use overview.

What it pays

Every study shows its pay and expected time before you start, and a study done in good faith is paid the amount shown even if you got stuck or gave up. Pay is added to your balance when the study is approved, usually within two working days. The AI use does not change the pay; the consent is separate and the study pays the same either way.

Why this matters

AI agents that use websites on people's behalf are mostly trained on how other machines or expert annotators use them, which is not how people do. They over-complete tasks a real person would abandon and miss the places a real person gets stuck. Recordings of ordinary users are what teaches a model to predict what people will do, which is what product teams actually need to know.

Questions people ask next

Is my voice used to train a voice model?
No. Raw audio is never part of released data. What is used is what you did and a transcript with personal information removed.
Can I take studies without the AI consent?
Each study's consent screen states that study's use. A study that trains models asks for that consent explicitly.
Do I need to know anything about AI?
No. You are more useful if you don't, because the value is in how you normally behave.