An AI data broker is any intermediary between a company that owns work records and the AI labs that want them. It is not a people-search site that sells profiles of individuals. Here are the four roles you will meet, and how each one gets paid.
Search for "data broker" and most results are about companies that sell information on private individuals. The AI training data market uses the same words for something else.
AI data broker (also AI training data broker): a company or person that sits between a business that owns records of real work and the AI labs that train or evaluate models on such records. Depending on the role, it licenses the data itself, represents the seller, lists datasets for buyers, or only introduces a company to a buyer program.
People-search or consumer data broker: a company that collects personal information about individuals from public records, purchased lists and online sources, and sells profiles or search access to others. The people in those profiles usually have no relationship with the broker.
If someone offers your company a "data broker for AI" service, they mean your own documents, tickets, code, messages and process records, not lists of people. Your company signs the agreement, agrees the scope, and gets paid.
The privacy questions still overlap. Employee messages, customer emails and client files contain personal information, so de-identification, consent and confidentiality come up in every serious AI data deal.
| Question | People-search / consumer data broker | AI training data intermediary |
|---|---|---|
| What is traded | Profiles of individuals: names, addresses, phone numbers, relatives, history | A company's work records: SOPs, tickets, CRM notes, code history, documents, messages |
| Where it comes from | Public records, purchased lists and online sources | The company that owns it, under a signed agreement and an agreed scope |
| Who buys | Marketers, background checks, anyone searching a person | AI labs and the data companies that supply them |
| Who gets paid | The broker | The company that owns the data, plus any intermediary under its own terms |
| Consent | Usually none from the person profiled | The company agrees in writing; employee and customer privacy still has to be handled |
| Rules that apply | Consumer privacy and data broker rules in many places | Contract law, privacy law (GDPR, CCPA/CPRA), sector law (HIPAA, GLBA), client confidentiality |
General information, not legal advice. Talk to your own lawyer before you sign.
"AI data broker" is used loosely for all four. They do different jobs and answer to different people, so ask early which one you are dealing with.
Data companies that license company records, prepare them and deliver them to labs. micro1, Mode, Grepped and Miro Advisory publish programs for businesses.
Advisers who act for the company: scope the dataset, approach buyers and negotiate terms. They work for you, so you pay them.
Platforms that list datasets so several buyers can see them. Some focus on archives from companies that have shut down.
People or sites that introduce a company to a buyer program and then step out. They never touch the data and do not negotiate.
A buyer program signs with your company, takes an agreed copy of the records in scope, prepares it and supplies it to labs. It runs discovery, the contract, export and de-identification. micro1's page (checked 7 October 2026) describes four steps: a short assessment, an evaluation of fit, a discovery stage with more questions about your data and privacy requirements, then agreement and onboarding. It says pay depends on dataset size and volume, workflow complexity, data quality, domain expertise and uniqueness, and that companies keep ownership of their underlying data. Mode's page says it buys "an agreed copy", the originals stay with the company, and it de-identifies before onward delivery.
How it is paid: by the labs it supplies. It pays your company the license price set in your agreement.A representative works for the seller. The useful work is preparation and leverage: list what you have, write a manifest, agree exclusions, approach more than one buyer and compare offers on terms, not only price. Practitioners advise never sending a full dataset before a price is agreed. Share a manifest and samples, and get more than one offer. A representative should not need your raw data to do this. If one asks to hold or host it, you are adding a second party with access and a second contract to read.
How it is paid: by your company, as a flat fee or a commission on the deal. Ask in writing whether it also takes money from any buyer.A marketplace lists datasets so several buyers can look at them. The best-documented example in 2026 is the closure market, where startups that had shut down sold old Slack and email archives. Forbes covered it on 16 April 2026, and Fast Company and Gizmodo followed. Troveo cites about $5,000 per code repository and roughly $10,000 to $100,000 per archive deal in that market. A running company licensing a scoped copy is a different case: the staff, clients and contracts are all still active.
How it is paid: listing or transaction fees under the platform's own terms.A referrer sends a company to a buyer program through a tracked link and plays no further part. micro1's referral terms (updated 24 September 2026) show what the role involves. The referrer needs consent from an authorized person at the company before referring it. The company must not already be a lead in micro1's system. Referrers may not present themselves as micro1's agent, partner or representative, or make promises about outcomes. Payouts are gross amounts, at micro1's discretion, and can be withheld or recovered.
How it is paid: by the buyer program, after a close. micro1 publishes "Earn $50,000" per company, Mode "Earn $50K per referral", Grepped "Refer for another $10K".Some labs take offers themselves. Google has an intake at contentpilot.google.com, and OpenAI has a data partnerships page. Labs tend to want large or unique datasets and buy through their own procurement: NDA, master services agreement, dataset evaluation and a purchase order. Nobody takes a cut, but nobody scopes, prepares or de-identifies the data for you either. That is why most companies of 20 to 1,000 people meet a buyer program first. Our page on who buys company data for AI covers when the direct path makes sense.
How it is paid: the lab pays your company directly. There is no referral fee.The market is young. A July 2026 market map by Deedy Das counts 50+ companies selling data and RL environments to labs, with about $8.5B in revenue. Will Depue (July 2026) sees labs on a path to more than $100B a year of data spend by 2030. Both are estimates, not audited figures.
Who pays an intermediary tells you whose interests it serves. Here is the payment logic for each role, using only what the programs publish.
| Role | Works for | Who pays it | How it is paid |
|---|---|---|---|
| Buyer program | Itself and the labs it supplies | AI labs | Revenue from the labs it supplies; its margins are not published. It pays your company the license price. |
| Seller-side representative | Your company | Your company | A fee or commission agreed with you in writing; amounts vary, so get the terms before any work starts |
| Marketplace | The platform | Sellers, buyers or both | Listing or transaction fees under its own terms; read them before you list |
| Referrer | Neither side in the deal | The buyer program | A published referral fee, only after a deal closes, at the program's discretion |
| Direct to a lab | Not applicable | Not applicable | No intermediary; the lab pays your company directly |
Last checked: 7 October 2026. Role descriptions are general.
Company payouts are what a company might receive. Referral fees are what the buyer pays a referrer. Neither is an offer or an average.
| Program | Company payout (as published) | Eligibility (as published) | Referral fee (as published) |
|---|---|---|---|
| micro1 Enterprise Data Partnership | "$100k+ qualified", "$500k+ large-scale", "$1M+ highly unique"; referral page: "$100K-$2M+ for approved data packages" | 30+ employees, mature operations, documented processes, primarily English; US prioritized, then other Western markets | "Earn $50,000" per company; terms default to 7.5% of the license price when no amount is listed |
| Mode company data | "$100K-$5M" | 20+ full-time US office employees; accounting firms 10+; law firms 6+; several years of records the company owns | "Earn $50K per referral" (site); "up to $55k or 6% of the transaction, whichever is greater" (its posts on X) |
| Grepped | "$20K-$5M", "get paid in 7 days" | Any vertical; also pays individual professionals for expertise | "Refer for another $10K" |
| Miro Advisory | Operating datasets "$100K-$1M+", private codebases "$10K-$1M+" (indicative) | Businesses; software companies | Not published on its site |
Last checked: 7 October 2026. Sources: micro1.ai/data-partnerships, micro1.ai/company-referral and micro1.ai/legal/referral-terms (updated 24 September 2026); data.mode.inc; grepped.ai; miroadvisory.com. Full comparison: buyer programs compared.
None of this is a judgment about any named company. It is how incentives work in any deal with a middle party.
A buyer program wants your data at a price that works for its lab customers. It can be fair and still not be on your side of the table.
micro1's referral terms tie payouts to the referred company reaching revenue milestones, keep timing at micro1's discretion, and allow clawbacks. A referrer has no say in your price or terms.
Referral fees come from the buyer. micro1's terms also bar referrers from sharing a payout with the referred company, so a cash-back offer is a red flag.
The same steps run whether you apply directly, through a referral link or with a representative. Each step shows where each role stops.
You apply to a program yourself or through a referral link. A referrer's part ends here. A representative may already be preparing your manifest.
Signed before the buyer sees samples. Check who is bound, for how long, and whether it covers the buyer's downstream customers.
The buyer looks at your manifest and samples and decides whether to go further. micro1's FAQ says referred companies go through an initial AI interview to assess fit.
Scope, price, payment structure, exclusivity, indemnities, deletion and termination. This is where a representative, if you hired one, earns the fee, and where your lawyer should read every clause.
Only the systems and date ranges in scope are pulled. Exclusions agreed in writing (client folders, HR, legal) stay out.
The buyer removes or replaces personal and confidential details. micro1's page says companies can review representative samples before prepared data is used. Ask any buyer how you can check this step.
The buyer confirms the data meets the agreed criteria. Acceptance often triggers payment, so the criteria should be written down before export.
One-off or in milestones, as the agreement says. Any referral fee is paid by the buyer after this point, under its own terms.
None of these depends on who the buyer is. They come from how deals with a middle party work.
Once a buyer holds the full dataset, your leverage is gone. Share a manifest and samples first, and agree price and terms before export.
Published ranges such as "$100K-$5M" show what a program can pay its best cases. Your number depends on what your data holds and how unique it is.
A referrer is paid by the buyer and has no role in your terms. A buyer program is the other side of your contract. Only a representative you pay is on your side.
Exclusive or long time-limited terms rule out a second buyer for the same data. If you grant them, price them, and check whether resale is allowed.
Law, accounting, M&A and healthcare firms often owe clients confidentiality. Those duties can rule out material no matter what a buyer will accept.
Under the programs described here, the buyer pays any referral fee and your company is not charged. Someone asking you for money to "get you in" is offering something else.
The same checklist works for a buyer program, a representative, a marketplace or a lab. It does not depend on reputation or on anything said in public.
Sell Data to AI is an independent guide, and on the map above it is a referrer. Some links on this site are referral links: if your company signs with a buyer through them, the buyer may pay us a fee. You are not charged, and we never see your data.
We are not a broker in either sense. We never receive, store, view or transfer company data, we do not negotiate with labs, and we are not a partner, agent or representative of any buyer. For the referrer's side, see data referral programs and how to become an AI data broker.
An AI data broker is any intermediary that helps company records reach AI labs for training or evaluation. In practice that covers four roles: buyer programs that license company data and prepare it for labs, seller-side representatives who act for the company, marketplaces that list datasets, and referrers who introduce a company to a buyer program and step out. It is not the same thing as a people-search data broker.
No. People-search and consumer data brokers collect and sell profiles of individuals, usually without any relationship to those people. AI training data intermediaries deal with a company and its own work records, under a written agreement with that company. The privacy questions overlap, but the business is different.
It depends on the role. Buyer programs earn from the AI labs they supply. Seller-side representatives charge the company a fee or commission agreed in writing. Referrers are paid by the buyer program under its published terms: micro1 publishes "Earn $50,000" per company, Mode publishes "Earn $50K per referral" and Grepped publishes "Refer for another $10K", as checked on 7 October 2026. Referral fees come only after a deal closes and are at the program's discretion.
Not under the programs described on this site. The buyer program pays any referral fee from its own side. micro1's published referral terms also forbid referrers from sharing a payout with the referred company, so a referrer should never offer you a rebate or cash-back.
Read its published terms, check what exclusivity and resale rights it asks for, check who carries the indemnity if de-identification misses something, ask for a sample contract before you share data, and have your own lawyer review it. Compare more than one program before you send any data.
No. Sell Data to AI is an independent guide. It never receives, stores, views or transfers company data and never negotiates deals. Some links are referral links, so a buyer may pay us a fee if a company signs through them. We are not a partner, agent or representative of any buyer.
See which buyer programs your company fits by their own published criteria, then compare their ranges and terms side by side. Nothing you enter leaves your browser.