12 questions to ask any AI data buyer before you sign

The 12 questions

Print these out for the first call. Each one comes with what a clear answer looks like, using what buyers have actually published.^6,7

1. Are you the buyer, or a broker? Most company data advertisers are brokers. A clear answer names the company that pays for the data.¹
2. Who pays you, and how much? Pewter says its licensing partner pays it at close; Jottings says it takes typically 20-30% of your payout.^8,9
3. What is your legal company name and address? micro1 names a Delaware corporation. Some advertisers give no company name, address or terms at all.^10,11
4. Which AI labs will receive our data, and will you tell us? Sellers often cannot see the end buyer, so ask early.²
5. What is the price, in writing, before anything leaves our office? Nyne's ads say the price is agreed first. SimpleClosure's CEO warns about buyers who want a full export before a real offer.^3,12
6. When and how are we paid? Look for a date in writing. Speeds vary, but it's usually around 3-6 weeks, sometimes longer.^{13,14,15,16,17}
7. Is the license exclusive, and for how long? micro1's is non-exclusive and perpetual. Polyshares says its licenses are exclusive by default. Troveo's exclusivity is time-limited.^4,18,19
8. Can you pass our data on to others? A sublicensable license, like micro1's, lets the buyer do that.⁴
9. Who removes personal details, by what method, and do we see a sample first? Polyshares uses an outside firm. micro1 lets you review scrubbed samples. In the Spirit case, a firm certifies the result to the court.^20,21,22
10. Do you need access to our systems? One law firm says a buyer should get none. Mercor's process links to your tools, so ask exactly what it can read.^5,23
11. How long do you keep our copy, and how will you prove it is deleted? Nyne's ads say 30 days for raw files; micro1's terms say 120 days. Models are not deleted.^24,4
12. Have you completed and paid for a deal before, and can we speak to that seller? SimpleClosure tells founders to ask this. micro1 said in court that it did 50+ data deals in 45 days.^6,25

Answers that should slow you down

A request for your whole archive before any number is mentioned. A website with no company name, no address and no terms. A domain registered last week. An example deal that also appears, word for word, on someone else's site.^12,11,26,27

None of these proves anything on its own. Plenty of honest brokers are new. But each one is a reason to ask for more in writing before you share a single file.²⁸

Answers that build confidence

Published terms you can read before the call, like micro1's data license terms. A named company and people, like Turing's Project Lazarus, which gives its company name and San Francisco address. An outside security audit, like the SOC 2 Type II report Mercor says it holds. And a willingness to put every answer above in the contract.^4,29,30