An AI data broker finds companies that hold records of real work and connects them with the data companies that license those records for AI training. This page sets out how that business makes money, what it costs, when money arrives and where it goes wrong, using only figures the buyers publish. It is not about people-search data brokers, and it promises no income.
Already have a target? Qualify it against buyer rules in about 20 seconds, free.
Most writing about the data broker business model is about people-search and consumer data brokers: companies that compile information about individuals from public records, purchased lists and online sources, then sell profiles, lists or search access. That business is regulated in a growing number of places and is built on data about people who never agreed to be in it. This page does not advise on it in any way. If your plan involves compiling or selling information about individuals, stop here and talk to a lawyer.
The AI data broker is a newer and narrower role. It works with companies, not with profiles of people. A company that has operated for years holds records of real work: support tickets, process documents, project histories, code and the messages around them. Data companies such as micro1, Mode and Grepped license that kind of record, prepare it and supply it to AI labs. The broker's job is to find companies that fit those buyers' published rules and connect the two sides. Every model below starts from that job. The four roles in the market are described in what is an AI data broker; this page is about the money.
Every AI data broker business runs on one of three revenue models, or a mix of them. They differ in who pays, when, and how much of the deal you carry yourself. The further you move from introducing toward holding data, the more of the deal's risk becomes yours.
The buyer pays you for introducing a company that goes on to sign. As published, checked 7 October 2026: micro1 earn $50,000
per referred company, with no cap
; Mode Earn $50K per referral
on its site, and on X up to $55k or 6%
; Grepped Refer for another $10K
.
You never see the data and you do not negotiate. The buyer runs review, contract, export and payment.
The selling company pays you, under your own written agreement, for preparing it and taking it to buyers. No program publishes a standard rate, so the fee is whatever you and the company agree, usually tied to the close.
You owe duties to your client, you often see samples under NDA, and if a buyer also pays you a referral fee you are paid by both sides of the same deal.
You license or buy data from a company, then license it on to your own customers. In theory you keep the spread between what you pay and what you sell for.
In practice you hold the data, pay or commit before you are paid, and give warranties to customers on records you did not create. It is a data business, not a side project.
| Model | Who pays you | Published amounts | When money arrives | Capital at risk |
|---|---|---|---|---|
| Referral | The buyer program, under its referral terms | micro1 "earn $50,000"; Mode "$50K per referral" (on X: "up to $55k or 6%"); Grepped $10K | After the close plus program conditions. micro1: after onboarding plus a minimum revenue threshold | Your time and your tools |
| Success fee | The selling company, under your own agreement | None published; agreed case by case | As your agreement says, usually after the company is paid | Time, tools and the legal review of your own agreement |
| Resale | Your own customers | None published for resale. Company payout ranges show what buyers pay owners directly | After you deliver and your customer accepts | The purchase price, storage, security, de-identification, insurance and legal work |
Resale is the only model in which the broker becomes a data business. That changes five things at once, and each of them is a cost a referrer never carries.
Nobody publishes what independent AI data brokers or referrers earn, and we make no claim about any individual's income. What is published is how large the market is, what buyers pay companies, and what they pay per referral.
Two pieces of arithmetic come straight out of those figures. First, the market is concentrated. If four companies hold 75% of about $8.5B, the remaining quarter, roughly $2.1B, is shared by the other 46 or more. A broker does not get a share of that revenue. A broker gets a fee from a buyer's sourcing budget, one closed deal at a time.
Second, the referral fee is large next to the bottom of each company range. In the published numbers of all three referral programs, the referral amount is exactly half of the floor: $50K against micro1's $100K, $50K against Mode's $100K, $10K against Grepped's $20K. Keep that in mind when reading qualifying conditions such as micro1's minimum revenue threshold: a flat amount is paid only on deals that meet the program's own bar, not on every signature.
The archive market is the context, not the product. Forbes (16 April 2026), Fast Company and Gizmodo covered startups selling old Slack and email after shutting down. We do not sell a shut-down or wind-down list; every Data Asset Score shows a company's activity status instead.
Market figures are estimates from the sources named, not audited numbers. Buyer figures as published, checked 7 October 2026.
In this business, revenue is late and uncertain while costs are early and certain. Four cost lines matter, and every one of them grows as you move from referral toward resale.
The largest cost, and the one nobody pays back. Researching companies, checking fit, getting consent from someone authorized, writing the introduction and following up all happen before any close. Every introduction that does not close is unpaid. Put a value on your own hours before you put one on anything else.
Research, lists and record keeping. Our Data Asset Score plans are $99 a month (Basic, 5,000 lookups), $299 (Pro, 25,000 lookups plus company lists) and $799 (Scale, 100,000 lookups, lists and bulk CSV). A single sector list bought once is $249, with no monthly fee. The demo and the check against buyer rules are free. Add whatever you use to track introductions.
A lawyer's read of each program's terms, of any seller agreement you sign, and of your own setup: entity, tax, disclosure, and whether any registration rule touches what you do. Costs vary by place and scope, so get a quote before you start rather than after a problem.
Close to zero if you never hold data, which is the point of introducing rather than carrying. The moment you accept samples or files, you take on security, privacy and confidentiality duties, and an incident becomes yours. In resale it is the main cost line, not a footnote.
| Cost line | Referral | Success fee | Resale |
|---|---|---|---|
| Time per company | Research, qualifying, consent, introduction and follow-up | All of that, plus preparing the company and negotiating for it | All of that, plus acquiring, preparing and selling the data |
| Tools | Research and record keeping | Research, records and a secure place for samples under NDA | Research, records, secure storage and de-identification |
| Legal advice | Program terms, tax and disclosure | Your client agreement, conflicts and NDA duties | Licenses in and out, warranties, privacy compliance and possible registration |
| Capital | Running costs only | Running costs only | Payment or commitment to the company before your customer pays |
| Data handling | None, if you never take data | Samples under NDA | Full custody of the records |
Timing is the part of the model new brokers underestimate most. Every published referral amount is paid per qualifying closed deal, never per introduction. Here is the order of events, and which way money flows at each step.
You pick companies, score them and check them against the published rules. This is where most of your hours go, and nobody has agreed to anything yet.
Someone authorized at the company agrees, and you submit it through the program's own referral process. From here on, the buyer controls the pace.
The buyer checks fit against its rules and reviews a manifest and samples. Many companies stop here. An introduction that stops here earns nothing, however much work went into it.
Scope, price, exclusivity and liability are negotiated between the company and the buyer. Practitioners cite 60 to 90 days for a deal to close. That is the earliest point at which any fee can be earned.
micro1, for example, pays referral fees after onboarding plus a minimum revenue threshold. A close is necessary, not sufficient, and these conditions add time after it.
micro1's terms give it sole discretion over referral payouts. A success fee from a seller arrives when your own agreement says, usually after the company has been paid.
micro1's terms allow clawbacks. A fee that has arrived can still be reclaimed under the terms, so treat it as settled only when the terms say it is, and keep the tax on it aside until then.
Unit economics for a broker come down to three numbers: the fee per qualifying close, the cost of running the business for a period, and the number of closes in that period. The first two are published or knowable. The third is not, and we do not invent it.
No program publishes how many introductions turn into a close, and any win rate quoted for this market is a guess. So the example below shows the arithmetic at zero closes and at one close, the only two cases we can state without making up a conversion rate. It is illustrative: not a forecast, not an offer and not a description of anyone's earnings.
| Line | At a $10,000 fee (Grepped, as published) | At a $50,000 fee (micro1 or Mode's site, as published) |
|---|---|---|
| Published fee per qualifying close | $10,000 | $50,000 |
| Pro plan for 12 months, at $299 a month | $3,588 | $3,588 |
| Of which paid before the earliest close (60 to 90 days, about 2 to 3 months) | $598 to $897 | $598 to $897 |
| Result at zero closes | a loss of $3,588, plus your time | a loss of $3,588, plus your time |
| Result at one close that clears | $10,000 less $3,588 = $6,412, before tax, advice and your time | $50,000 less $3,588 = $46,412, before tax, advice and your time |
| Result at one close, later clawed back | a loss of $3,588, plus your time | a loss of $3,588, plus your time |
The same arithmetic shows why the plan should follow the work you actually do. Here is a year of each plan as a share of one published fee. The bigger the share, the more of a single close the tools would consume.
| Plan | Per month | Per year | Share of one $10,000 fee | Share of one $50,000 fee |
|---|---|---|---|---|
| Basic | $99 | $1,188 | 11.9% | 2.4% |
| Pro | $299 | $3,588 | 35.9% | 7.2% |
| Scale | $799 | $9,588 | 95.9% | 19.2% |
Under a flat fee, deal size does not change what you are paid. Read as a flat fee, micro1's published "earn $50,000" would pay the same on a $100,000 deal and on a $2M deal: 50% of the first, 2.5% of the second. Against Mode's published "$100K-$5M" the same $50K runs from 50% down to 1%, and Grepped's $10K against "$20K-$5M" runs from 50% down to 0.2%. Each program's own terms say how its amount is set for a given referral.
Under a percentage basis, such as the 6% in Mode's post on X, the fee moves with the deal: 6% of $100,000 is $6,000, and 6% only reaches $55,000 on a deal of about $917,000. Which basis applies to a given referral is set by Mode's own terms and site, not by a post and not by this page.
What the arithmetic says. With flat fees, a broker's lever is how well each introduction fits the published rules, not how large a deal might become. With zero closes the business is pure cost. Plan your finances as if the number of closes is zero, and treat a fee as income only once it has cleared.
A broker's pipeline has four stages. Each one has a job, an output and a record worth keeping, because the record is what shows when and how an introduction was made if a payout is decided months later.
Find companies likely to hold the records buyers want. Start from a segment rather than a hunch: a list ranked by Data Asset Score, or companies you already know, scored one at a time.
Output: a short list with a score, a grade, the data each company likely holds, its history and its activity status.
Compare each company with the published rules. micro1 lists 30+ employees (its referral posting says 30 to 200), US first, primarily English. Mode lists 20+ full-time US office employees, 10+ for accounting firms, 6+ for law firms, with several years of records. Grepped lists any vertical.
Output: a likely, possible or unlikely fit per program, with the facts behind it.
Get consent from someone authorized at the company. Tell them you may be paid if they sign, that they are not charged, and that the buyer decides everything. Submit through the program's own referral process, and never present yourself as the buyer's partner.
Output: a dated record of who agreed, when, and through which link.
Stay out of the negotiation, but keep your record current: review, agreement, close, program conditions, payout and the clawback window. Ask the company, not the buyer, how things stand, and only as often as is reasonable.
Output: a status per company and the date you will next look at it.
Sell Data to AI sells research tools for the first two stages: deciding which companies to look at and whether each one fits. The tools do not introduce anyone, contact anyone or close anything.
Scores any company 0 to 100 for the data AI buyers want, with a grade, the data it likely holds, its history and its activity status: active, winding down or acquired, parked, or unreachable. Built on our index of 102 million domains, 99.99% of the active internet, with domain history, across nine factor groups: history, scale, knowledge assets, operational systems, customer systems, organization, industry value, expertise and activity status.
Free demo, limited per day.
Ready lists for 20 US sectors, from labs and CROs to law firms, accounting firms and marketing agencies. Every company is verified active: the domain resolves, is not expired, is not parked and shows no error or placeholder page. Each list has a free preview with the top 5 visible, such as the labs and CROs preview, and the list hub shows live counts.
Buy a list once from $249 as a CSV snapshot, or keep lists current on Pro through the API as JSON, up to 100 companies per call, with paging. Scale adds the one-file bulk CSV export and new segments on request.
REST and JSON, with your key in an X-API-Key header. Score the companies already in your own records, or page through a sector list into your own tracker.
The key appears in your dashboard after payment, not by email. API documentation.
Free. Reads a company's public pages and compares team size, location, history and systems with the published rules of micro1, Mode and Grepped, quoting the sentences it relied on.
Qualify a target in about 20 seconds before you ask anyone for consent.
| Plan | Price | Lookups a month | Company lists | Bulk CSV | Suits a broker who |
|---|---|---|---|---|---|
| Basic | $99 a month | 5,000 | No | No | scores companies already known, one at a time. Choose Basic |
| Pro | $299 a month | 25,000 | Yes, through the API | No | works one or more of the 20 sector lists and wants them current. Choose Pro |
| Scale | $799 a month | 100,000 | Yes | Yes, plus new segments on request | runs a sourcing team or loads a CRM in bulk. Choose Scale |
| One-time lists | $249 for 1 list | None | Yes, as a CSV download | Yes, a snapshot with no updates | wants one sector list, or a few, without a monthly plan. 2 lists $449, 3 lists $599, 5 lists $899, each further list +$110, all 20 lists $2,490. Buy lists |
The conduct rules for referrers are on how to become an AI data broker. This is the business side: the order in which to set things up so costs stay small until you know whether your introductions close.
General information, not legal advice. Talk to your own lawyer before you take on any role in a data deal.
The phrase "data broker" carries legal weight in some places. Some US states have registration rules for data brokers, but those rules apply mainly to businesses that collect and sell personal information about consumers, which describes the people-search business rather than the referral role. Whether any of them reaches what you do depends on where you operate and whether you ever handle personal data. Ask a lawyer rather than assume either way.
Most of the legal work for a referrer is lighter than that: reading program terms properly, setting up to receive fees and pay tax on them, and making the right disclosures. It gets heavier with each step toward the data. A success-fee adviser owes duties to a client and may see samples under NDA. A reseller takes on licenses in both directions, warranties and privacy law, including laws such as GDPR, CCPA/CPRA and HIPAA where the records fall under them.
Deals involving companies in bankruptcy or winding down add a further question: who has authority to agree. The Spirit Airlines data sale is a reminder that a sale by a bankrupt company can need approval: as of 7 October 2026, approval had not been confirmed.
None of these depends on any one buyer. They follow from how the model works: someone else decides, someone else pays, and money arrives last.
Most of the work goes into introductions that do not close, and none of it is paid. That is the base case to plan around, not the exception.
Fit, price, acceptance and payout decisions sit with the buyer. micro1's terms state sole discretion over referral payouts.
A fee that has been paid can be reclaimed under micro1's terms. Spending it early turns a reversal into debt.
60 to 90 days to close, then the program's conditions. After the introduction, you control nothing in that sequence.
Programs can change amounts, rules and who qualifies. Every figure on this page is as published on a date, not a standing promise.
Three of the programs we follow publish referral amounts. A broker who relies on one carries that program's every decision as a single point of failure.
A success fee from the seller and a referral fee from the buyer on one deal puts you on both sides. Disclose, check the terms, and ask counsel.
Accepting samples or files turns a referral business into a data business, with duties you may not be set up to meet.
Quoting the top of a range, promising acceptance, or mass outreach to companies you do not know damages the trust every introduction depends on.
On the AI side, in one of three ways: a referral fee paid by a buyer program after a company you introduced signs and meets the program's conditions; a success fee paid by a selling company under your own written agreement; or the spread on data you license and resell. People-search data brokers, which sell profiles of individuals, are a different business that this page does not cover or advise on.
Nobody publishes what independent AI data brokers earn, and we make no claim about anyone's income. What is published is per closed deal: micro1 "earn $50,000", Mode "$50K per referral" (on X, "up to $55k or 6%") and Grepped "Refer for another $10K", as published, checked 7 October 2026. At the market level, a July 2026 market map by Deedy Das counts 50+ companies selling data and RL environments to labs, with about $8.5B in revenue and 75% of it held by Scale, Surge, Mercor and Handshake.
Finding companies that hold records of real work, checking them against buyers' published rules and connecting them with those buyers. The broker is paid by the buyer (a referral fee), by the seller (a success fee) or by its own customers (resale). The costs are time, tools, legal advice and, for anyone who holds data, data handling risk. Revenue arrives only after a deal closes.
Pick one revenue model, read every program's terms in full, take your legal questions to a lawyer, set a cost ceiling for the first 90 days, focus on one sector, qualify each target against the published rules before you contact it, and introduce only with consent from someone authorized at the company. Then review your own numbers at day 90.
After the deal closes and the program's own conditions are met, never for an introduction alone. Practitioners cite 60 to 90 days for a deal to close. micro1 pays referral fees after onboarding plus a minimum revenue threshold, at its sole discretion, and its terms allow clawbacks.
Some US states have registration rules for data brokers, but they apply mainly to businesses that collect and sell personal information about consumers. Whether any of them reaches your activity depends on where you operate and whether you ever handle personal data. Ask a lawyer. This is general information, not legal advice.
Because the reseller pays or commits before it is paid, holds records that contain personal and confidential information, gives warranties to its customers on data it did not create, and competes with buyer programs that already license directly from companies. Raw data is also the cheapest layer: practitioners say evaluations are worth about 10x raw.
No. Everything is company level: no contacts, named people, emails or phone numbers. A Data Asset Score is an estimate from public signals, not a valuation, not an offer and not proof that a company wants to sell. The free check compares a company's public pages with the published rules of micro1, Mode and Grepped.
Score a company in the free demo, qualify it against buyer rules for free, and choose a plan only when your volume needs one. Company-level data only, and no promise of any fee.
Disclosure: Sell Data to AI is independent and sells the research tools described on this page. It is not a partner, agent or representative of any buyer, and it never receives company data.