How to Sell Your Data to AI Companies from Canada: Process, Buyers, and the Law

If your business has years of video, photography, specialist documents, or recordings of skilled work, an AI company may want to license it. This is a real market: AI labs and robotics companies are paying for data they cannot scrape. But it is also young, mostly private, and full of hype.

This guide covers how the process works, who is buying, and the Canadian rules that apply, as of October 10, 2026. It is general information, not legal advice. Anyone licensing data should have a lawyer review rights and contracts.

Step 1: Work out what you actually have

Start with an honest inventory. For each type of data, note how much there is (hours, images, words), where it came from, who created it, and who appears in it. Buyers value data that is not already online, so unpublished material counts for more than anything you have already posted publicly.

Buyers are explicit about what they want. OpenAI, for example, says it is looking for large-scale datasets that are not already easily accessible online, in any format, and particularly data that expresses human intention, such as long-form writing or conversations rather than disconnected snippets. It also says it does not want datasets containing sensitive or personal information, or information that belongs to a third party.

Source: OpenAI, Data Partnerships

Then check it against what buyers pay. Our breakdown of what AI companies pay for training data lists every public price point we could verify. If the rough value is small, stop here. That is a perfectly good outcome.

Step 2: Confirm you own it and can license it

This is where most datasets fail. You need a clear chain of title: you created the data, or you hold rights that let you license it for AI training. Watch for:

• Freelancer and agency work where the contract never assigned copyright to you
• Stock footage, music, or images you licensed for your own use, not for resale
• Client material you hold but do not own
• Content written by employees under terms that do not cover this use

Canadian copyright law has no specific exception for text and data mining. The federal government consulted on copyright and generative AI but, as of October 2026, has not changed the law, and Canadian news publishers are suing OpenAI in Ontario. Assume you need clear ownership or permission for every creative work in a dataset.

Source: ISED, consultation on copyright and generative AI

Source: BNN Bloomberg, publishers' lawsuit against OpenAI proceeds in Ontario

Step 3: Deal with personal information

If people can be identified in your data, privacy law applies. In most provinces that means PIPEDA for commercial activity, or provincial laws such as BC PIPA. Canada's privacy commissioners have said that collecting or using personal information to train AI needs a legal basis, often consent, and that personal information should only be used for the purposes it was collected for, or consistent ones.

Source: Office of the Privacy Commissioner of Canada, principles for generative AI

If your business is subject to Quebec's private-sector privacy law, Law 25 requires a privacy impact assessment before personal information is sent outside the province. Most AI buyers are outside Quebec, so plan for it.

Source: LégisQuébec, Act respecting the protection of personal information in the private sector

The law is also moving. The federal AI and Data Act died with Bill C-27 in January 2025. A new privacy bill, C-36, was introduced in June 2026; it would enact a new private-sector privacy law and amend PIPEDA, and was still before Parliament as of October 2026.

Source: DLA Piper, Bill C-27 dies on the order paper

Source: LEGISinfo, Bill C-36

The practical answer for most businesses: remove or blur personal information wherever you can, get fresh written consent where you cannot, and leave health data and data about children out entirely unless a lawyer has designed the process.

Source: First Nations Information Governance Centre, OCAP

Step 4: Package and document it

Buyers evaluate a sample before they commit. A sellable dataset is organized, de-duplicated, consistently formatted, and comes with a data card: a short document covering where the data came from, how it was collected, what consent covers it, licence terms, and known limits. Industry groups such as the Data & Trust Alliance have published provenance standards that describe what buyers look for.

Source: CMSWire, data provenance standards

Step 5: Find the buyer

Here is the part the hype skips. You will rarely sell straight to a frontier lab with a cold email. OpenAI runs a public data partnerships program, launched in 2023, with an interest form for organizations offering either an open-source archive or a private dataset. Most other labs publish no intake process at all, and most deals are never announced.

Most licensed data reaches labs through intermediaries that already have buying relationships:

• Curated licensing platforms that vet datasets and sell them to AI developers, often in a specialty such as health, video, or text
• Specialist brokers and advisers who package an owner's data, find the right buyer, and negotiate the deal
• General data marketplaces where datasets are listed for buyers to find
• Direct programs run by a lab, where one exists

Each route has trade-offs in fees, reach, and how much work stays with you. The right one depends on your data type, its volume, and how much help you need getting it ready.

In Canada, Cohere is the major homegrown model builder, and the federal government has committed up to C$240 million toward its compute expansion. Canadian data with a natural edge includes parallel French and English content and Canadian legal, regulatory, and technical material.

Source: Government of Canada, investment in Cohere

Step 6: Negotiate terms, with counsel

Common terms to understand before you sign:

• Pricing model: per minute, per image, per title, a flat annual fee, or a revenue share
• Term: how long the licence lasts, and whether models trained during it can be used afterwards
• Exclusivity: pays more, but blocks other buyers
• Use restrictions: for example, no digital replicas of people who appear in your footage
• Warranties: what you promise about rights and consent, and what happens if a claim arises

Intermediaries take a share of the deal. Ask for every fee in writing before you agree to anything, and compare what you would net through each route, not just the headline price.

Red flags

• Anyone asking for an upfront fee with a promise of a guaranteed buyer
• "Six figures for your data" pitches without a look at your actual data
• Contracts that make you warrant rights you are not sure you hold
• Buyers who want personal information left in
• Vague payment terms. Ask any intermediary how and when you get paid, in writing

Signal & Form helps Canadian businesses through this process, from the first inventory to routing a dataset to the right buyer, paid through a success fee. See our AI data licensing service for how it works.