What is AI training? A guide for professionals
AI training is paid, remote work where experts check and correct the answers of AI systems like ChatGPT, Claude or Gemini. An AI answer is not only text. It can also be a calculation, a table, a spreadsheet model or a chart. You don't need AI or tech skills. Platforms pay for your finance knowledge, typically $60-$150/h for experienced professionals.
You are hired for a specific project, or for several. Each project has its own rate, tasks and hours, and work can be busy one month and quiet the next.

Why does AI need experts?
AI systems learn from examples, mostly text from the internet. The internet can't teach an AI how a senior auditor tests revenue recognition or how an analyst builds a merger model. AI companies therefore pay professionals to:
- spot mistakes in the AI's answers
- write what a correct answer looks like
- write hard questions that show where the AI still falls short
As AI improves, its remaining mistakes get more specialized, and finding them takes expertise in the field.
What does the work look like?
Most tasks are done in your browser, in the hours you choose. You get a task, do it and submit it. A task usually takes between 15 minutes and a few hours.
| Task type | What you do | Example (finance) |
|---|---|---|
| Review an answer | Read the AI's answer and score whether it is correct, complete and clear. | The AI explains how a lease is treated under IFRS 16. You mark the error and explain it. |
| Write a model answer | Write the answer an expert would give. | You write how you'd calculate free cash flow from a set of statements. |
| Write a hard question | Create a realistic question the AI gets wrong. | A case where the AI confuses enterprise value and equity value. |
| Build a scoring checklist ("rubric") | List what a good answer must include. | "Must adjust for net debt" and "must state the discount rate assumption". |
| Compare two answers | Say which of two AI answers is better, and why. | Two summaries of a 10-K, where one misses a going-concern note. |
Example task
Question to the AI: "A company has an equity value of $500 million, $200 million of debt and $50 million of cash. It owns 80% of a subsidiary that it fully consolidates, and the non-controlling interest is worth $40 million. What is its enterprise value?"
AI answer: "$650 million: equity value plus debt minus cash."
Your correction: Incomplete. The company's EBITDA includes 100% of the subsidiary, so enterprise value must include the $40 million non-controlling interest: $500m + $200m - $50m + $40m = $690 million. Without it, the EV/EBITDA multiple comes out too low.
Example written by us. Real tasks are often longer and come with documents or spreadsheets.
You can't use AI tools for the tasks. You judge the AI's answers with your own knowledge and experience.
Who can do this?
Platforms mostly look for people with:
- several years of work experience in their field, often 2 to 5 years or more for senior roles
- a relevant degree, and for some roles a professional qualification such as CPA, ACCA, CFA or chartered accountant
- good written English (some projects also need other languages, such as German, French, Spanish or Japanese)
- a computer and a reliable internet connection
Most platforms accept people from most countries, but some projects are open to US residents only. See which platforms accept experts in your country.

How much does it pay?
Finance professionals usually see $60-$150/h, and specialists up to ~$250/h. Each project sets its own rate, and you work as an independent contractor. See Pay rates for finance for more.
Is AI training legit?
Yes. The large platforms on this site are real companies that reportedly work with the largest AI labs, such as OpenAI, Anthropic and Google DeepMind. Scam websites copy this work, though, and a real platform never asks you to pay. Read How to spot a scam before you sign up anywhere.
The market behind this work is large and well funded:
- Meta invested $14.3 billion in Scale AI, the company behind Outlier, for a 49% stake (June 2025).
- Mercor raised $350 million at a $10 billion valuation (October 2025) and reportedly paid contractors more than $2 million a day in May 2026.
- micro1 reportedly raised more than $100 million at a $4 billion valuation (September 2026).
Am I training my own replacement?
This is a fair question. AI will change finance work with or without your help. When experts train it, the AI in your field makes fewer of the mistakes you would catch.
Can I do this next to my job?
Yes, most people do. We recommend checking your employment contract first, to be safe. Read more on our Employer rules page.

Frequently asked questions
Will I talk to clients or AI companies?
Usually not. You work through the platform, which manages the project and is in contact with the client.
Can I use AI tools to do the tasks?
No. Platforms pay for your own judgment, and training AI with AI defeats the purpose of the work. In our experience, it is the most common reason experts are removed from projects or platforms. Some projects allow no AI tools at all, not even for grammar.
Is my work confidential?
Yes. You usually sign a confidentiality agreement for each project. Never use confidential information from your employer or clients in a task.
Your next step
Sources
- Pay ranges: aitrainer.work job pages, tracked 5 and 6 October 2026, and the platforms' own job listings (our method)
- Accessed 16 October 2026: TechCrunch, "Mercor quintuples valuation to $10B with $350M Series C", 27 October 2025
- Accessed 21 October 2026: CNBC, "Scale AI's Alexandr Wang confirms departure for Meta as part of $14.3 billion deal", 12 June 2025
- Accessed 21 October 2026: Forbes, "This 25-Year-Old Raised Over $100 Million For His AI Data Startup At A $4 Billion Valuation", 22 September 2026
- Accessed 16 October 2026: Yahoo Finance, 2 May 2026 (Mercor pays contractors more than $2 million a day)