How to Sell Your Company's Data to AI Labs

The short answer: how companies sell data to AI labs

The route this guide covers is licensing: you give an AI lab the right to use a de-identified copy of your business data for training, under terms you agree to. On SellMyData, the exchange that publishes this guide, the lab never gets access to your systems and receives only the de-identified dataset, never raw data. You export a copy yourself, so your own systems and records are unchanged.

Licensing means granting a buyer the right to use a copy of your data under agreed terms.

Why labs pay for this kind of data: Forbes, citing former OpenAI chief scientist Ilya Sutskever, reports that labs first trained models on the public internet (Reddit threads, Wikipedia entries, digitized books) and had exhausted all of it by late 2024. The same article reports that defunct startups are being liquidated for their Slack archives, Jira tickets and email threads, operational data that AI labs now treat as premium training data.

There are public deals on record. Reddit struck a deal that lets Google use posts from the site to train its AI models. In August 2026, Google won a bankruptcy auction with a $10 million bid for Spirit Airlines’ corporate data.

What kinds of company data AI labs want

The corporate data in the Spirit sale and the startup wind-downs described above is everyday operational data: messages, code, tickets and the records of how work gets done. You don’t need a special product to have it. If your team has used chat, email, a code host or a help desk for years, you already hold it.

The Spirit Airlines sale shows the range. Per court filings reported by Axios, it includes about 100 million emails and 500 million Microsoft Teams chats. TechSpot reported that it also covers 516 code repositories holding about 30 million lines of custom software, plus IT support tickets and customer-service workflows.

TechSpot also explained why this material differs from scraped web text: business records show how people make decisions, coordinate with colleagues, fix mistakes and navigate business systems. All of that could help train AI agents (software that carries out multi-step tasks on its own).

Our exchange accepts five types of business data:

Data type	Examples	What it can show a model
Messages	Team chat, internal email	How colleagues coordinate and reach decisions
Code	Repositories	How real software is written and fixed
Documents	Specs, policies, playbooks	How a business explains and records its decisions
Records	Structured entries from business systems	How work moves through business systems
Tickets	Support and IT tickets	How problems are reported, worked and resolved

Volume and years of history matter too. They are two of the four factors we use to estimate price (more on that below).

What is out of scope

We do not buy consumer personal data, health records tied to people, or payment card data. The Spirit sale drew a similar line: it does not include passenger profiles or frequent flyer information.

Selling vs. licensing: what you actually agree to

When you “sell” data to an AI lab through our exchange, you agree to license terms that set what the lab can do with a copy. Four terms are agreed per dataset:

• Scope: which data is included and what the buyer may use it for.
• Format: how the data is structured and delivered.
• Exclusivity: whether you can license the same data to other buyers.
• Term: how long the license lasts.

Nothing is licensed without your sign-off on the buyer, the terms and the price, so you can decline a buyer or a set of terms. What a license actually permits depends on the exact wording of the agreement, so have counsel review it before you sign.

Step by step: from estimate to payout

The route below is how the process works on our exchange. Other routes, such as the startup wind-downs and bankruptcy sale described above, work differently.

1. Get an indicative range. The homepage calculator (employees, years of history, data types, industry) shows an indicative payout range. It is not an offer.
2. Request an estimate. Answer a few questions before sharing any data. A person reviews every request and replies by email. There is no minimum headcount and no sales call to list.
3. You export the data yourself. You choose which tools and date ranges to export. We never write to your systems.
4. We clean and de-identify it, and you approve the result. People’s names, contact details and identifiers are replaced with consistent placeholders. Nothing moves forward until you approve.
5. We check rights with you before listing. This covers customer contracts, your privacy policy, employee communications and third-party content.
6. Buyers see a redacted sample, and you approve each deal. You sign off on the buyer, the terms and the price. After a sale, the lab receives the de-identified dataset, never raw data.
7. You get paid. You pay nothing up front. We take a commission only when a sale closes, and the target is payment within 7 days of a sale closing.

Check your rights before you sell

Before any data is listed, confirm that you have the right to license it. Four areas deserve a close look. Whether each one allows a sale depends on your specific agreements and where you operate, so treat this checklist as general information and review it with counsel.

Area	The question to ask
Customer contracts	Do your agreements limit how you can use data your customers send you?
Privacy policy	What have you told users and customers about sharing data or using it for AI training?
Employee communications	Do chats and emails contain staff’s personal details, and what have employees been told?
Third-party content	Does your data include material owned by others, such as vendor documents or licensed code?

Privacy policies. FTC staff have written that it may be unfair or deceptive for a company to adopt more permissive data practices, such as sharing consumers’ data with third parties or using it for AI training, and to tell consumers only through a surreptitious, retroactive change to its terms of service or privacy policy. The same post says any firm that reneges on its user privacy commitments risks running afoul of the law. This is staff guidance, not a rule, but it shows why your existing promises matter. How it applies to you is a question for counsel.

Employee communications. Workplace chat is about real people. As a privacy advocate quoted in recent coverage put it: “It’s not generic data. It’s identifiable people.” That is why de-identification and a specific check on employee communications come before any listing.

Our rights page describes what we check with a company before its data is listed. It complements a review by your own counsel; it does not replace one.

How de-identification protects people in your data

De-identification removes the link between data and the people it describes, so a lab can learn from the work while lowering the risk that the people in it can be identified. It lowers risk; it does not make re-identification impossible, which is why it works alongside rights checks and contract terms.

NIST, the U.S. standards agency, defines de-identification as a general term for any process of removing the association between a set of identifying data and the data subject (the person the data is about). Its guidance lists techniques including removing identifiers, transforming quasi-identifiers and generating synthetic data using models. A quasi-identifier is a detail, such as a job title or a city, that could single someone out when combined with other details.

In our process, people’s names, contact details and identifiers are removed and replaced with consistent placeholders before anything is listed. “Consistent” means the same person gets the same placeholder throughout a dataset, so a support thread or a chat conversation still reads as a coherent exchange between distinct people. That consistency is what keeps the data useful for training.

It also has limits. FTC staff explain that hashing (turning an identifier into a code) works consistently, so the same input always creates the same output, and that companies should not act or claim as if hashing personal information makes it anonymized. The same logic applies to placeholders: they keep data useful and lower risk, but contracts and checks still matter. In the Spirit sale, for example, the court filing says the buyer agrees not to attempt to re-identify users.

Our de-identification page covers what is removed before a lab sees anything, and the limits.

What your company data is worth

There is no universal price list for business data. On our exchange, price depends on four things: data type, volume, years of history and current lab demand.

Publicly reported figures give a sense of range, but none is a benchmark for your data. They involve different data, different sizes and different circumstances:

• Reddit and Google. The deal was signed on 22 February 2024, and Reuters reported it is worth around $60m per year. Neither company confirmed the price.
• Spirit Airlines. Google won the bankruptcy auction with a $10 million bid for the airline’s corporate data, which covers hundreds of millions of messages.
• Closed startups. One founder of a closed company said she sold every Slack message, internal email and Jira ticket as training data for “hundreds of thousands of dollars”. That is a single anecdote. Separately, a wind-down services firm told Forbes it had processed 100 deals selling shuttered startups’ workplace data in the past year.

For a number closer to your situation, the homepage calculator takes your employees, years of history, data types and industry and shows an indicative payout range. It is calibrated to publicly reported deals and other buyers’ estimators as of October 2026, and it is not an offer. Here is how the estimate works.

Staying in control after you list

Listing your data should not mean losing control of it. On our exchange, you approve every buyer, the terms and the price, and you can unlist and delete your data at any time.

Other safeguards:

• Buyers see a redacted sample first. The licensed lab receives the de-identified dataset, never raw data.
• No access to your systems. You run the exports; we never write to your systems.
• Limited handling of exports. Company exports are kept in a restricted Google Drive folder that only the founder can access, and they are deleted within 30 days of the de-identified dataset being approved. Details are on our security page.

How the law treats de-identified data (general information)

Legal labels for processed data vary, so don’t assume that any de-identified dataset, from any source, falls outside privacy law. What follows is general information about datasets of all kinds. Check the legal status of your data with counsel where you operate.

• Re-identification risk. The UK Information Commissioner’s Office (ICO) says that even if all identifying information such as names and addresses is removed, enough details about people may remain to potentially re-identify them.
• Pseudonymisation. This means replacing identifiers with placeholders. The ICO says pseudonymous information is still personal data and the law applies to it. That is why data handled with placeholders is best described as “de-identified”, not “anonymous”, why placeholders should be paired with a rights check and contract terms, and why you should confirm the legal status of any dataset with counsel where you operate.
• Undefined terms. The ICO published its final anonymisation and pseudonymisation guidance on 28 March 2025. The law firm Taylor Wessing’s summary of that guidance notes that “de-identified data” is not a term defined under UK law.

Rules in other countries may differ. Your counsel can tell you which ones apply to your company and your data.

Is selling data right for your company?

Selling makes sense when you have years of operational data, clear ownership of it and agreements that allow licensing. If any of those are uncertain, sort them out first, with counsel where needed.

Good-fit signals:

• Several years of history in everyday business tools such as chat, email, code hosting or a help desk.
• Clear ownership of the data and the accounts it lives in.
• Customer contracts and a privacy policy that, on review with counsel, permit this kind of use.

Reasons to wait:

• Your rights are unclear, or your contracts or privacy policy may not allow it.
• Your data is mostly consumer personal data, health records tied to people or payment card data.
• You have open legal questions. Talk to counsel before going further.

If the fit looks right, the low-commitment next step is to request an estimate from SellMyData. You answer a few questions before sharing any data, and a person reviews your request and replies by email.