Short answer: AI data buyers really do license copies of company records such as email, chat, files and CRM history, and several programs publicly pay people who introduce companies, which is why these emails are arriving. Before you reply, find out who signs the license and pays you, exactly what data they want, whether it's exclusive, and whether the sender is paid. None of the programs we track says it charges companies to apply, and you can apply to any buyer directly.
On this page
Why these emails are arriving now
Company records have become AI training data, and it's been in the news: Forbes wrote about old work Slacks and emails as AI's new training data (April 2026), MIT Technology Review reported AI labs buying data from dead startups (October 13, 2026), and Accounting Today reported that BizBuyData will pay accounting firms for data to train AI (October 16, 2026).
Many of these programs also pay whoever introduces a company to them. On their own pages, Handshake AI says it pays $10K to someone who introduces a business that becomes a partner, Scale AI says referrers can earn up to $50,000, and Turing's Project Lazarus advertises up to $10K for referrers (Handshake AI, Scale AI, Turing).
Fees like that give strangers a reason to email small firms in bulk. So the person writing to you may be the buyer, or may be someone who gets paid only if you sign with a buyer they haven't named yet. Your first job is to find out which.
Check which buyers fit your firm in about a minute, no email needed
Answer a few questions and see which programs' published rules your firm meets, with the reasons quoted from each program. Nothing is sent to any program unless you tick a box that names it.
Start with your firm type: law firm · accounting or CPA firm · architecture or engineering firm · any other business. Or browse every buyer we track.
We may earn a referral fee from programs we introduce companies to. Companies never pay DataWorth AI anything. Disclosure
What “your data” usually means
Buyers license a copy of your operating records, not your company. Mode lists email and calendar, documents and files, chat, calls and transcripts, CRM and sales, and billing and operations (Mode). micro1 lists SOPs, internal documentation, knowledge bases, CRM metadata, project histories and QA processes (micro1). Per Mode's site, Mode buys a filtered, de-identified copy and you keep your original records and systems.
| Firm type | Often in scope | What usually has to stay out |
|---|---|---|
| Architecture | Drawings, models and specifications; consultant coordination and RFIs; proposals and fee records | Design files you don't own under the owner-architect agreement; consultants' drawings |
| Law firm | How the firm runs: intake and matter workflows, templates with no client facts, billing and staff procedures | Information relating to any client's representation, unless the client consents |
| Accounting / CPA | Workflows, checklists, internal procedures and practice-management history | Tax return information, which federal law restricts for preparers |
| Engineering | Project histories, calculations, RFIs, submittals and change orders | Client information you can't reveal without consent; covered defense information |
Sources: Mode, micro1, ABA Model Rule 1.6, 26 U.S.C. § 7216, NSPE Code, 17 U.S.C. § 102. Firm guides: architects, law firms, CPA firms, engineers.
Keep health records, payment card data and children's data out of scope entirely: each has its own legal rules, and a buyer's de-identification doesn't change that. If an email asks for “everything”, or for client files, patient records or tax returns, that tells you the sender hasn't thought about what can legally leave your firm.
How deals and payouts generally work
Per Mode's site, the steps are: share your company's size, operating history and systems; agree on the value, terms, systems and date ranges before any export; then your team signs in to the approved systems, and Mode exports and de-identifies the agreed copy and pays under the agreed terms.
Published money. Mode shows its own estimate on its site, and we don't republish Mode's figures: See Mode's own estimate. micro1 publishes $100K to $2M+ for approved data packages (micro1). Final offers follow the buyer's review. micro1 cites $100K to $2M+ for approved data packages (micro1). Lunchbreak gives “a two-minute estimate, then a written offer after a confidential sample review” (Lunchbreak). Small deals are real too: Ooak Data says its deals so far “range from $20K, the smallest closed, to several hundred thousand” (Ooak Data).
What moves the number. Per Mode's site, Mode reviews the systems and dates in scope, years of history, record volume and quality, and ownership and access terms before making a fixed offer.
Exclusivity varies by buyer. Per Mode's site, the agreement defines its scope and any exclusivity. Labelbox Prism says its licenses are “exclusive by default” (Labelbox Prism); Scrimdata says it is “non-exclusive by default” (Scrimdata). Signing one exclusive deal can shut you out of others, so read that clause first. More: exclusivity explained.
Real buyer or middleman? How to tell
A buyer licenses the data itself and its company name goes on the agreement and the payment. A middleman is anyone between you and that buyer who is paid if you sign. Being paid isn't wrong in itself (we're paid by some programs, below), but you should know who pays whom before you share anything.
| Ask | A buyer's answer | Warning sign |
|---|---|---|
| Who signs the license and pays us? | “We do”, with a company name you can check on its own website | “My partner” or “my client”, not named until you commit |
| Do we pay you anything? | No | Any upfront, valuation, listing or “onboarding” fee |
| Are you paid if we sign? By whom? | A plain answer | Dodging the question |
| What do we sign first? | Usually an NDA, then a written offer or agreed scope | An exclusive agreement with the sender before you've seen any offer |
| Can we apply to the buyer ourselves? | Yes | “Only through us” or claims of special access |
| How long do we have? | Time to review the written terms | An offer that expires within hours or days |
None of the programs we track says it charges companies to apply or to get an offer. Every buyer in our directory is listed with its own public page, where you can apply or get in touch directly.
Questions to ask before you reply
- Which company will sign the license and pay us? Is that you?
- Are you paid if we sign, by whom, and does your fee come out of our price?
- Exactly which systems and date ranges do you want, and what is excluded?
- Is it exclusive? For how long, for which data, and does it cover only external AI training?
- Is your number a written offer or an estimate? What would change it?
- What happens before anything leaves our systems: NDA, agreed scope, sample review, signed license?
- Who removes personal and client information, and can we audit a sample first?
- When are we paid, and can any of it be taken back later?
- Have you already submitted our company's name to any program? Will you submit it anywhere without our written OK?
Get the answers in writing. For comparison, per Mode's site, the agreed scope lists the approved systems and date ranges before any export, and no passwords are handed over.
Red flags
- Any fee to you. Upfront, valuation, listing or onboarding fees.
- A clock. An offer that “expires” in a day or two, or a price that's “only good today”. Real buyers give you time to review written terms.
- No named buyer. The sender won't say who signs and pays, or the buyer isn't a party to the agreement.
- Access before paper. Requests for passwords, admin access, exports or “a quick sample” before an NDA and written terms.
- Exclusive to the sender. You're asked to sign exclusive representation before seeing an offer.
- Big number first, small number later. A large figure in the email, then a much lower written offer with little explanation. Compare any number to the buyer's own published range or estimate.
- “Everything”. No clear list of systems, years or exclusions, or a request for client files, patient records or tax returns.
- Unclear terms. No written answer on exclusivity, payment timing, clawbacks or deletion.
Confidentiality and scope basics
The duty to your clients, patients and staff stays with your firm. No buyer's de-identification process and no sender's reassurance moves that risk off you. In plain terms:
- Law firms: ABA Model Rule 1.6(a) bars revealing information relating to a client's representation without informed consent or an exception (Rule 1.6). Your state's version binds you. If client information leaves the firm without consent, your license is the one at risk.
- Tax preparers: federal law restricts disclosing or using tax return information (26 U.S.C. § 7216), and the penalties apply to the preparer.
- Medical and dental: HIPAA applies, and practice email and CRM usually contain patient information, so expect strict scope limits. HIPAA generally bars selling protected health information without authorization (45 CFR 164.502).
- Engineers and architects: the NSPE Code bars revealing client information without consent (NSPE Code), and design files may belong to the client or a consultant under your contracts.
- Everyone: email and chat are full of employee personal data, and California's CCPA exemptions for employee and business-to-business data expired December 31, 2022 (California AG). Client contracts, NDAs and software terms can also bar sharing (Frankfurt Kurnit).
The conservative scope is your firm's own operations (templates with no client facts, workflows, SOPs, training and internal procedures), not client files. Exclude high-risk records outright rather than relying on de-identification. Put the scope (systems and date ranges), the exclusions, deletion, sample audit and a no-publicity promise in the agreement.
Your first reply is a disclosure too. Keep it general. Don't send client names, screenshots of your systems, file samples or logins to anyone you haven't verified.
More: de-identification, employee data, risk checklist.
How referral credit works, and why you can go direct
- micro1: a referrer is credited only if the company signs up and applies through that referrer's link; applications made without it can't be credited later (micro1 referral terms).
- Scale AI: only one referral is paid per deal (Scale AI).
- Lunchbreak: the person introduced must know they're being introduced and have agreed to hear from Lunchbreak, and Lunchbreak says it contacts them about selling data to Lunchbreak and nothing else (Lunchbreak referral terms).
What that means for you. Whoever introduces your firm first may hold the credit, so don't let anyone submit your company unless you've chosen them. You don't need an introduction at all: you can apply on any buyer's own page. Ask any buyer whether a referral fee affects your price.
How we fit in. DataWorth AI is a referral directory, not a buyer. We may earn a referral fee from programs we introduce companies to; our Disclosure lists them. Companies never pay us. Our quiz needs no email and sends nothing to any program unless you tick a box that names it. You're free to use our list and apply directly.
Check which buyers fit your firm, then decide who to talk to
Answer a few questions and see which programs' published rules your firm meets, with the reasons quoted from each program. Nothing is sent to any program unless you tick a box that names it.
Start with your firm type: law firm · accounting or CPA firm · architecture or engineering firm · any other business. Or browse every buyer we track.
We may earn a referral fee from programs we introduce companies to. Companies never pay DataWorth AI anything. Disclosure
What could your company's records be worth?
Answer 6 questions. See the published range for a company like yours and which program fits. Takes about a minute.
Frequently asked questions
Is the person emailing me the actual buyer?
Not necessarily. Some senders are the buyer; others are paid by a program only if you sign. Ask which company signs the license and pays you, then check that company's own website and apply there if you want to.
The sender calls themselves a 'data broker'. Are they the buyer?
The label doesn't tell you. Ask who signs the license and who pays you. If it isn't the sender, they are a go-between, and you can contact the buyer directly.
Should I pay a fee to get an offer?
No. None of the AI data programs we track says it charges companies to apply or to get an offer. Treat any upfront fee as a red flag.
Is a low offer a sign of a scam?
Not by itself. Published ranges are wide and small deals happen; Ooak Data says its smallest closed deal was $20K. What matters is whether the written offer is explained and compares sensibly with the buyer's own published range.
If I go direct, do I lose anything?
Ask the buyer whether a referral fee affects your price. Going direct means no referrer is credited.
Can I talk to more than one buyer?
Yes, until you sign. Exclusivity starts with the agreement, and terms vary by buyer, so read that clause before you sign.
What should my first reply say?
Keep it general: ask who the buyer is, whether the sender is paid and by whom, and for the buyer's written terms. Don't send client names, files, screenshots of your systems or logins.
Does DataWorth AI charge anything?
No. Companies are never charged by DataWorth AI. We may earn a referral fee from programs we introduce companies to, and we say so on every page.
Sources
- Forbes: AI's New Training Data: Your Old Work Slacks And Emails (Apr 16, 2026)forbes.com
- MIT Technology Review: AI labs are buying data from dead startups (Oct 13, 2026)technologyreview.com
- Accounting Today: BizBuyData will pay accounting firms for data (Oct 16, 2026)accountingtoday.com
- Handshake AI data partnerships: license pagejoinhandshake.com
- Scale AI Data Partnerships page and FAQscale.com
- Turing, Project Lazarus pagelazarus.turing.com
- Mode / BizBuyData referral pagebizbuydata.com
- Mode company data partnerships page (FAQ)ai.mode.inc
- micro1 company referral pagemicro1.ai
- micro1 Referral Program Terms (last updated Oct 14, 2026)micro1.ai
- Lunchbreak: Refer and Earnlunchbreak.com
- Lunchbreak referral termslunchbreak.com
- Ooak Data: Partnerooakdata.com
- Labelbox Prismprism.labelbox.com
- Scrimdata: Partnerscrimdata.com
- ABA Model Rule 1.6, Confidentiality of Informationamericanbar.org
- 26 U.S.C. § 7216 (Cornell LII)law.cornell.edu
- 45 CFR § 164.502, including the sale-of-PHI rule (Cornell LII)law.cornell.edu
- NSPE Code of Ethics for Engineers (PDF)nspe.org
- 17 U.S.C. § 102, Subject matter of copyright (Cornell LII)law.cornell.edu
- California Attorney General: CCPA overview and FAQoag.ca.gov
- Frankfurt Kurnit: Selling Company Data for AI Trainingtechnologylaw.fkks.com
Referral-fee disclosure: we may be paid by a program if your company signs with it after using this site. General information only, not legal advice.
Compare programs
Best AI data licensing programs for companies (2026)Side by side from each program's own pages. Rankings follow published rules, not referral fees.
See every AI data buyer we track, from their own pages, with sources and dates.