AI Data Buyer Match: Published Acceptance Criteria by Company Size, Country, Status and Data Type

This table shows what 32 AI data buyer programs publicly state about the companies they accept: company status, size and history, country and language, and data type, each as the program's own claim with the date DataPacts checked it. Rows run alphabetically, there is no payout or fee column, and no program pays DataPacts.

As of 2026-11-09, six rows carry a statement about company status, ten a headcount number, ten a country or language rule and 29 name data types. A cell reads "not stated" when the program's pages say nothing on that point, which is different from "no" and from "yes". Published criteria are the first filter in AI training data licensing: they show where a stated rule is missed, never that a program will take the data.

DataPacts Editorial Team read each program page on November 9, 2026; how we source, date and check every figure sets out the rules.

What Each Buyer Program States It Accepts, Side by Side

Each row quotes a program's own pages as checked on November 9, 2026, with the source in the last column. The table compares 32 company data rows as of 2026-11-09. Cells are shortened quotations, and a program may change its page after the check date.

Kinds: CD company data program; RL RL-environment builder taking company data; CC custom capture for businesses with physical work; NV terms not verified. Every cell is the program's claim, not a DataPacts finding.

How to Check Your Company Against the Four Criteria

Read each row against four facts about the company: its status, its size and history, its country and language, and the data types it holds, in the order of the four subsections below. A missing claim reads "not stated" and is not a miss.

Company Status: Operating, Winding Down, Closed or Insolvent

Six of the 32 rows carry a statement on company status, and the other 26 say nothing, which is different from refusing closed companies. Four state a rule or a welcome:

• AfterQuery states that operating, winding-down and closed companies are welcome.
• Labelbox states that distressed and wound-down companies are eligible for Prism.
• Ooak Data states that wind-down deals are closed with a director, liquidator or administrator signing.
• Turing states that Project Lazarus takes active, winding-down or wound-down companies, with shutdown 90 days ago or less and more than 90 days ago as separate statuses.

The other two are weaker: Centific says winding-down companies "may move faster", a note on speed and not a rule, while micro1's referral page lists defunct partnerships separately while its program page states no status. Replay states no closed-company wording, so it is not grouped with these programs. Who signs for a company that has closed is covered under selling data from a shut-down or defunct company. This page records only what each program states.

Company Size and Operating History

Ten rows carry a headcount number, with floors from 5 employees on some Mercor listings to 50 on Mode's referral page and about 50 at Collinear AI, and three state no minimum or any size. The numbers, by floor:

• 10+: Abaka AI states 10 or more employees across departments for company data.
• 20+: Handshake AI's license and referral pages state 20+ full-time employees and 3+ years, Nyne states 20+ full-time US W-2 employees and 3+ years of workflows, and Replay's referral page states 20+ for a referred company.
• 30+: micro1 states 30+ full-time employees, and Mode's home page states 30+ (10+ accounting, finance and medical; 6+ legal).
• 40+: Scale AI states 40+ as a preferred starting point, case by case.
• About 50+: Collinear AI states teams of about 50+ with several years of human-written history.

Mercor's industry listings state a size on 23 of 25 listings: 5 to 2,000 employees on 20, 50 to 5,000 agents for call centers, a clinician or patient count for medical practices and "Any size" for imaging centers. Centific, Labelbox and SuperAnnotate state no minimum or, in SuperAnnotate's words, "Any size, any industry, anywhere."

Country, Language and Where the Data May Go

Ten rows state a country or language rule, and most of them put US companies or English-language records first.

• US first: micro1 states data primarily in English with US companies prioritized, then other Western markets; Mode's home page states the US as strongest fit, others case by case; Nyne asks for US W-2 employees and offers a country selector (USA, Canada, Europe, Other); Replay's referral page states English, US preferred.
• Listed regions: Collinear AI lists the US, Canada, the UK, the EU, Australia and New Zealand with English data; Turing names most regions, with strongest demand in the US, UK, Canada and EU; Mode's referral page lists the US, Canada, UK and Australia.
• EU processing: Ooak Data states that EU partner data is processed and stored in the EU through Ooak Data SAS.
• Anywhere: SuperAnnotate states "anywhere", and 18 of Mercor's 25 listings say "Anywhere".

micro1 states that agent trajectories are licensed only to buyers in the US and allied countries. That is a program's rule; separately, 28 CFR Part 202 (in force 2025-04-08) bars selling or licensing bulk US sensitive personal data, even de-identified, to buyers in or linked to China, Cuba, Iran, North Korea, Russia or Venezuela. Transfer and tax steps for a company abroad are set out in selling data to AI companies from outside the US.

Data Types Named and the Exclusions Programs State

Twenty-nine rows name the data types a program takes, and workplace records such as chat, email, tickets, CRM and documents appear in most of them.

• Workplace records: Abaka AI, Centific, Handshake AI, Mercor, Mode, Nyne, Polyshares, Scale AI, SuperAnnotate and Turing name chat, email, documents, CRM or tickets.
• Code: Abaka AI, Origin Lab, Patronus AI and pre.dev name codebases, source code or game projects.
• Physical work and footage: Asimov, Claru, Sieve and Toloka name first-person video, archive footage or recordings of physical work.
• Medical: Gradient Health and Shfa.ai name clinical records and EHR exports, and Labelbox requires IRB approval or qualified de-identification for medical records.

micro1's Data License Master Terms (updated 2026-11-08) exclude PHI, biometric identifiers, payment card or financial account data, children's data (under 18, or a higher age where the law requires) and export-controlled information unless the Order Form allows them (s.5.7). The legal limits behind such exclusions are in what company data cannot be licensed for AI training.

What a Match on Published Criteria Does Not Mean

Meeting every criterion a program publishes means only that nothing on its public pages rules a company out; it is not an acceptance, an offer or a price.

• Not an acceptance: each program decides after its own review, and no public record links these criteria to the companies a program accepted.
• "Preferred" and "case by case" are not rules: Scale AI states 40+ employees as a preferred starting point, and Mode's home page states other countries case by case.
• Criteria change: every cell describes a page as read on 2026-11-09.
• "Not stated" is not "yes": a page that is silent on status may still decline closed companies, and Replay's "One requirement" is a stated criterion, not a promise.

After the criteria, the choice among labs, vendors, marketplaces and brokers follows the steps in choosing a buyer for your data.

Programs Whose Own Pages State Different Criteria

Where a program's own pages give two thresholds, the table shows both and DataPacts does not pick one. Handshake AI, Mode and the intermediary Data Licensing Advisors each state different values on separate pages, as of 2026-11-09.

The buyer program changelog records dated changes over time. A company can meet one page and miss another, so a reader checks the page that governs the contract.

Intake Criteria Stated by Intermediaries and Marketplaces

Brokers, sell-side advisers, shutdown services and marketplaces publish intake criteria too, but they place or list data for a company and do not acquire it. Their thresholds are the intermediary's own statement of what it handles, so a cell here is not a buyer's rule.

Some thresholds describe labs: Marin Data states that most AI labs look for 20+ employees, 3+ years and several tools. Troveo's estimator slider, from 20 to 1,000+ employees, is an estimator range and not a rule. The table compares 22 intermediary and marketplace rows as of 2026-11-09, with IN for broker, adviser or shutdown service and MK for marketplace.

Listing rules and fees of marketplaces, among them Magma and Troveo, are compared on AI training data marketplaces.

Matching on Published Criteria: Benefits, Risks, Downsides and Alternatives

Checking published criteria first saves calls with programs that state a rule a company misses, but it cannot tell the holder whether licensing is right for the company.

Benefits

• It rules programs out before any NDA is signed.
• It shows where one program's own pages contradict each other.
• It shows where non-US or closed companies are named.

Risks

• A match is read as an acceptance.
• Criteria go stale after 2026-11-09.
• Contacting a program signals a possible sale.

Downsides

• 26 of the 32 rows state no status rule.
• Intermediary thresholds may describe what labs look for, not the intermediary's own rule.

Alternatives

• Ask the program in writing.
• Use an independent adviser or counsel.
• Send an enquiry for an independent introduction.
• Do not sell or license the data at all.

Whether to license is a separate decision, set out in should you sell or license your data.

Where the Buyer Match Sits Among the Seller Tools

The buyer match answers who a program says it accepts; the other seller tools answer what programs say they pay, how their contracts differ and what their calculators return. It is one of the tools for AI data sellers, and each neighbor covers one question:

• Payouts and referral fees: AI data partnership payouts and program terms.
• Contracts: the license clause matrix, which shows how buyer contracts differ, clause by clause.
• Calculators: buyer estimators side by side compares what each published calculator returns for one company profile.
• Programs: company data partnership programs explains how application works, and who buys AI training data lists every buyer, including those with no published criteria; brokers and advisers have their own page on AI data licensing brokers and sell-side advisers.
• Data: data types AI companies buy shows value and takers by data type.

How the Criteria Are Checked and Kept Independent

Each criterion is read from the program's own pages and dated; a cell changes only when the page has been read again. Rows are alphabetical and carry no score, no program or intermediary pays DataPacts, and the page on how DataPacts is funded states that no referral link appears in the tables.

Frequently Asked Questions

What is the lowest headcount any program states?

Abaka AI states 10 or more employees for company data, and Mercor's industry listings start at 5 employees, each as the program's claim checked November 9, 2026. Centific, Labelbox and SuperAnnotate state no minimum or any size, and Handshake AI's Data Partnerships page states no minimum while its license and referral pages state 20+ full-time employees.

Which programs state a rule for insolvent companies?

Only Ooak Data names insolvency officeholders, stating that wind-down deals are signed by a director, liquidator or administrator. Labelbox names distressed companies, and AfterQuery and Turing name winding-down or closed companies. How officeholders sell data is covered in how administrators, trustees and liquidators sell data.

Can a company outside the US meet these criteria?

Several programs name other regions: Collinear AI lists Canada, the UK, the EU, Australia and New Zealand, and Turing names most regions with strongest demand in the US, UK, Canada and EU. Ooak Data states EU processing, and 18 of Mercor's 25 listings say "Anywhere". micro1 prioritizes US companies.

Does meeting every stated criterion mean a program will buy the data?

No: criteria are a first screen, and each program decides after its own review of samples and terms. Under micro1's Data License Master Terms (updated 2026-11-08), the buyer has 30 business days to reject delivered data, otherwise it is deemed accepted (s.1.3), which is a separate step from a criteria match.