How I Landed Two AI Rater Contracts Without a Tech Degree

AI rater work is the quietest well paid microtask rail in the ecosystem. I stumbled on it by accident in early 2024, and by the end of the year had cleared a little over 4,300 euros across two niche vendors. Nobody on the mainstream microtask forums talks about it, because the people who get in tend to keep it to themselves.

What the work actually is

In practice, you read a prompt, you read two or three model answers, and you rank them against a rubric. Sometimes you write a better version yourself. Sometimes you annotate which of two answers is safer, more truthful or more helpful. The task types I handled were: preference ranking, fact checking, safety labelling, and rewriting. None required a technical background. All required clear written English and the ability to justify a rating in two to four sentences without waffle.

Where the gigs actually come from

Not from the big mainstream platforms. The two contracts I landed came through niche vendors that act as middlemen between the AI labs and individual raters. Think of them as staffing agencies. In my case: one UK based vendor that handles preference data for a large lab, and one European vendor that runs safety evaluations for a different lab. Both were found by searching LinkedIn for roles tagged *Data Analyst Part Time* with keywords *preference data* or *RLHF*, then applying directly. For the broader playbook on finding quiet work nobody advertises, the Freelance hub has my full strategy.

  1. LinkedIn search: filter for part time, contract, remote; keywords *RLHF*, *AI annotation*, *preference data*, *rater*.
  2. Company pages: identify 8 to 12 niche vendors, follow them, apply to any rater role that opens.
  3. Screener: expect a 90 minute unpaid written task, typically 4 to 6 ranking exercises.
  4. Interview: a 20 to 30 minute call with a program manager. Non technical, focused on rubric thinking.
  5. Paid qualification batch: 2 to 5 hours of paid work scored against internal benchmarks. Clear it and you are in.

The money, stripped of hype

On my two contracts, the rates landed at 18 and 22 euros an hour respectively. Payment terms were net 15 for the first vendor and net 30 for the second, both via SEPA bank transfer with VAT invoicing on my micro entreprise status. Volume varied wildly; one month I logged 41 paid hours, the next month seven. Treat it as a high rate, low reliability rail and it works. Treat it as a salary replacement and it will break your budget.

For the comparison against other freelance rails I run in parallel, the Prolific vs Userlytics piece is the companion read. For the mindset behind layering several small income streams so that no single one has to carry you, the dividend apps article applies the same logic on a different rail.

The red flags that cost me time

I applied to eleven vendors before my first offer. Three were scams or MLM adjacent, two never responded, four put me through a screener and ghosted. The remaining two paid. Red flags I now filter out immediately: unpaid screeners longer than two hours, rates advertised above 40 dollars an hour for generalist raters, any vendor that asks for a test payment or a credit check, and any role that requires you to pay for a training course. Legitimate vendors pay the qualification batch, every time.

John's rare tip

Write one clean sample evaluation, then reuse it

Screeners almost always ask you to rank two model outputs and justify your ranking. I wrote one template evaluation, 180 words, structured as *what the prompt asked for, how each answer handled it, why I rank A over B*. Reusing that structure on every screener cut my screener writing time from 90 minutes to 40 minutes, and scored me above the bar on four of five applications.

What I'd avoid

Do not sign an NDA without reading the non compete clause. One vendor offered me a 24 euros an hour contract that would have locked me out of any other AI annotation work for 12 months post termination. I walked away, took the 18 euros an hour offer that had no such clause, and kept the option to stack a second contract six months later. The flexibility was worth the 6 euros.

Frequently asked

Do I need a technical background to rate AI outputs?

No, but you need clear writing in English and a patient eye for nuance. My background is sports trading, not machine learning, and I cleared the screener on the second attempt. What actually matters is the ability to justify a rating in one clean paragraph.

How long does it take from application to first paid task?

Three to six weeks in my experience. The onboarding flow runs through a recruiter, a written screener, a short interview, and then a paid qualification batch. Anything faster than three weeks is usually a lower paying programme you should be skeptical of.

Is the income stable enough to plan around?

Not monthly. Programmes ramp up and wind down in four to eight week waves. I have seen 30 hours a week offered and then zero for a fortnight. Treat it as a high rate, low reliability rail, not a salary.