See if your data qualifies

Guides · Published

How to Sell Data to AI Companies in 2026: A Plain Guide

Learn how companies sell internal records to AI companies in 2026: what sells, how licensing works, what affects price, and the steps to take before you agree to anything.

Most companies sell data to AI companies by licensing a copy of records they already keep, such as support tickets, CRM history, project files or lab data, after names and IDs are removed. Deal sizes vary widely, and no honest number exists until a buyer has reviewed a sample. Smelted is one free route: it matches your company with buyers, and you approve each one before anything is shared.

Who buys company data

AI data companies and labs buy records to train AI agents on how real businesses work. Troveo says it has paid out more than 20 million dollars to over 7,000 rights holders licensing video, audio, gaming, robotics and business data (Troveo). The same page describes disclosed deals from about 10 million dollars for a company archive to more than 250 million over five years for a major publisher (Troveo). Those are large media deals. Treat them as the top of the range, not a typical offer for a mid-size company.

Marketplaces are another channel. OpenDataBay sells licensed datasets for AI training, and it states that from 1 October 2026 its marketplace prices are in USD (OpenDataBay). Marketplaces list ready-made datasets. Direct matching is a different model, where a buyer is introduced to a specific company's records.

What usually sells

Buyers look for records that show how work gets done. Records with years of history and clear structure tend to be most useful.

RecordsCommon toolsIndustries that hold them
Support tickets and help desk historyZendeskB2B software
CRM historySalesforce, HubSpotStaffing, B2B software
Engineering work and codeJira, GitHubB2B software
Placement and recruiting recordsBullhornStaffing and recruiting
Design and drawing filesAutoCAD, RevitEngineering and architecture
Lab and research recordsBenchlingScience labs and biotech R&D

SOPs and internal project files also come up often. Physical books and published works do not sell this way, because the copyright is not yours to license.

Selling usually means licensing, not handing over ownership

In most deals the company keeps ownership and grants a license to a copy. The contract sets what the buyer may do with the data, for how long, and what is excluded. Read those terms before you sign. Questions about contract language, customer confidentiality, privacy duties or industry rules belong with your own lawyer.

What affects the price

No honest figure exists before a buyer reviews a sample. Buyers weigh:

  • Volume and how many years the records cover
  • How clean and well-structured the records are
  • How unusual the data is for the buyer's work
  • Whether the data is fully yours or partly owned by clients
  • How much buyer demand there is for your industry right now

Smaller companies usually get less. Buyers who take smaller companies pay less, and that is true for any route, not only one.

How Smelted works

Smelted is a free matching service for US companies. The buyer pays Smelted's fee, and the company pays nothing. The steps are:

  1. Get a ballpark. Fill in the 1-2 minute form at smelted.ai, or book a 15-minute call with Chris Irving.
  2. Choose what is included. You decide which records are part of the offer and which buyers you meet. Names, contact details and IDs are removed before anything is shared.
  3. Meet 2-3 matching buyers. Smelted introduces you to buyers that fit your company, and you approve each one.
  4. Sign and get paid. Buyers' own pages say payment comes a few weeks to two months after signing. Plan on 1-3 months from first call to payment.

The company keeps ownership of its records. Smelted does not promise legal or compliance approval for any deal, so check terms with your counsel.

Who it fits

Smelted works best for owners, founders and C-level leaders at US companies with 50-500 employees (science labs 30-500), founded in 2022 or earlier, with years of records in tools like Zendesk, Jira/GitHub, Salesforce/HubSpot, Bullhorn, AutoCAD/Revit or Benchling. The core industries are staffing and recruiting, engineering and architecture, B2B software, and science labs and biotech R&D.

If your company has under 30 employees, the buyers who take smaller companies pay less. Talk to Chris Irving before you decide.

What we do not take

We decline law firms, healthcare and medical billing, accounting and tax, call centers, IT managed-service providers, education, government and defense. Physical books and published works are out of scope. Where a company only holds materials owned by its firm and not by clients, we can talk through that case first.

Before you agree to anything

Check these before signing with any buyer, whether it is Smelted or someone else:

  • Who owns each record, and whether clients have rights in it
  • What the contract allows the buyer to do, and for how long
  • Whether any customer agreement or privacy rule limits sharing
  • Whether you are already talking to another data buyer

FAQ

How do I sell my data to AI companies?

Start by listing your records and the tools that hold them, then decide whether you want to license a copy through a matching service or a marketplace. Remove names and IDs, confirm you have the rights to license the records, and review each buyer and contract before you agree.

How much do AI companies pay for data?

There is no fixed rate. Public deals range from modest archive licenses to very large media deals, as described above. For your company, the honest answer is the number a buyer gives after reviewing a sample, so ask for that before you commit.

What is AI data licensing?

AI data licensing means the owner grants a buyer permission to use a copy of data for a defined purpose, for a set time, under written terms. The owner keeps the original records.

How much is one person's data worth?

A single person's records rarely sell on their own. Buyers usually want large sets of business records with consistent structure, so value depends on the volume and quality of the set as a whole.

Sources