Because Canada’s Copyright Act has no provision addressing AI-generated output, the contract you clicked through when you signed up is often the only place your actual rights are spelled out. This guide is the order to read an AI tool’s terms in, and what each clause is actually deciding.
Key takeaways
STEP 01 OF 11
A Terms of Service sets the rules for using the product; a Privacy Policy governs what happens to your personal information; an End User Licence Agreement governs your right to use the underlying software itself. AI vendors frequently split these across separate documents, and the clause you need — often about output ownership or training use — can sit in any of the three depending on the vendor. See terms of service vs privacy policy vs EULA for the general distinction — it addresses these document types generally, not AI specifically, and the same structural split still applies.
Locate all three documents before starting, even if only one is linked prominently at sign-up. A vendor’s most consequential AI-specific commitments are sometimes in the document you would not have thought to open.
STEP 02 OF 11
This is the highest-priority clause specifically because Canadian copyright law does not clearly answer who owns AI-generated output on its own — the Act has no computer-generated-works provision, and ownership under s.13(1) runs on there being an identifiable author. In that gap, the vendor’s own terms are often the most concrete allocation of rights that exists. Read exactly what is granted: full ownership, a licence to use the output, or something narrower tied to your specific subscription tier.
Note whether the grant is conditional — some vendors reserve broader rights over output produced on free tiers than on paid ones, which is the subject of Step 9.
STEP 03 OF 11
Check whether the vendor reserves the right to use what you type in, or what the tool generates for you, to train or improve its own models. Look specifically for an opt-out, and read what it actually covers — some opt-outs apply only going forward, not to data already submitted, and some apply only to certain plan tiers.
If your business puts client or proprietary information into the tool, this clause determines whether that information could end up shaping outputs the vendor gives to other customers later. Treat a silent or vague clause on this point as a reason for caution, not reassurance.
STEP 04 OF 11
Check where the vendor actually processes your data, and whether it uses subprocessors that might be located elsewhere again. Canadian law does not prohibit processing outside Canada — the OPC’s own guidance confirms “PIPEDA does not prohibit organizations in Canada from transferring personal information to an organization in another jurisdiction for processing” — see Guidelines for processing personal data across borders — but your business, not the vendor, remains accountable under PIPEDA Schedule 1, clause 4.1.3: “An organization is responsible for personal information in its possession or custody, including information that has been transferred to a third party for processing” — see PIPEDA’s Schedule 1.
Because that accountability does not transfer with the data, check that the terms actually commit the vendor to protecting it at a comparable level — a vague statement that data “may be processed globally” with no further commitment leaves your business holding an obligation the contract does nothing to help you meet. See what to document when a business uses AI for how to record this once you know the answer.
STEP 05 OF 11
Almost every AI vendor’s terms disclaim responsibility for the accuracy of output in some form. This is not unusual or unfair on its own — it is a direct, contractual confirmation of the same point covered elsewhere on this hub: the tool’s output has to be checked before you rely on it, because nothing in the contract you agreed to promises otherwise. See how to sanity-check an answer from AI for what checking actually looks like in practice.
Do not expect to negotiate this clause away in a standard consumer or small-business agreement. Its presence is a reason to build your own verification habits, not a defect to argue about.
STEP 06 OF 11
Check what happens to your history, your saved outputs, and any custom configuration if you cancel — whether you can export it, and for how long it remains available after cancellation. A tool with no clear export path can leave a business dependent on staying subscribed simply to retain access to its own prior work.
This matters more for AI tools than for ordinary software specifically because the outputs — drafts, analyses, generated assets — are often the actual work product your business needs to keep, not just settings or preferences.
STEP 07 OF 11
Most AI terms include a list of prohibited uses — certain content categories, certain applications, sometimes specific industries. Confirm your intended use is not on this list, and note that violating it can void other commitments in the same agreement, including the output-ownership grant in Step 2.
Read this clause even if your planned use seems obviously permitted — some acceptable-use policies are broader than expected, particularly around regulated or sensitive content categories.
STEP 08 OF 11
If the terms describe a transfer of rights — from you to the vendor, or from the vendor to you — over inputs or outputs, note that Canada’s Copyright Act sets a specific formal requirement: “no assignment or grant is valid unless it is in writing signed by the owner of the right in respect of which the assignment or grant is made, or by the owner’s duly authorized agent” (s.13(4)) — see the Copyright Act, s.13.
A clickthrough acceptance of written online terms generally satisfies this requirement, but the distinction between an assignment (a transfer of ownership) and a licence (permission to use, ownership unchanged) is real and often blurred in plain-language summaries of a vendor’s terms. Read the actual word used in the clause, not a marketing description of it.
STEP 09 OF 11
Vendors commonly reserve broader rights over your data and output on free or low-cost tiers, and offer narrower training-use rights and stronger contractual commitments on paid or enterprise plans. If your business is deciding which tier to use for anything beyond casual, non-sensitive use, read both sets of terms side by side rather than assuming the paid tier only changes price and usage limits.
This comparison is often where the real difference between “fine for personal drafting” and “acceptable for client-facing business use” actually sits.
STEP 10 OF 11
Vendors update AI product terms more frequently than most software agreements, often reflecting fast-moving product changes. Record the date you reviewed the terms and, where the vendor notifies of updates, actually re-read the changed sections rather than accepting an update without review.
This is the same documentation discipline recommended elsewhere on this hub: a dated record of what you reviewed and when is worth far more than a general assumption that the terms have not meaningfully changed since you last checked.
STEP 11 OF 11
Where practical, save a copy or a dated screenshot of the actual terms you agreed to, alongside the note of when you reviewed them. Vendors do not always keep a public changelog of what moved between versions, and a vague memory of “the terms seemed fine when we signed up” is not something you can check against a later dispute.
This matters most for the clauses this guide flags as highest-priority — output ownership and training use — because those are exactly the terms most likely to shift as a vendor’s product and business model evolve.
Reading only the Terms of Service and skipping the Privacy Policy or EULA. A vendor’s most consequential commitment on data use or software licensing can sit in whichever of the three documents you did not open, per Step 1.
Assuming a free tier carries the same rights as the paid tier. Training-use rights and output ownership frequently differ by plan, sometimes substantially — compare both before committing meaningful business use to a free tool, per Step 9.
Treating an accuracy disclaimer as unusual or negotiable. Nearly every AI vendor includes one. Build your own verification process around it rather than treating its presence as a red flag specific to one vendor.
Confusing a licence with an assignment. A licence to use output and an assignment of ownership in it are different legal grants with different consequences. Read the specific word the clause uses, per Step 8, rather than assuming from context which one applies.
Reading in this order front-loads the clauses most likely to actually change what your business can do with the tool, rather than starting from whichever section happens to appear first in the document.
It is the most concrete answer available, since Canadian copyright law does not clearly resolve AI-output ownership on its own. It is not necessarily the final word in every scenario — a genuinely novel dispute could still test how the clause interacts with the underlying law — but for ordinary business use it is the governing document to rely on.
For a large enterprise contract, often yes, particularly around data use and liability. For a standard consumer or small-business signup, the terms are typically non-negotiable — which makes reading them before committing more important, not less.
It can. Free-tier training-use rights, per Step 9, often apply to whatever you input regardless of how you intend to use the output internally — if the input includes client or proprietary information, the training-use clause matters even for content that never leaves your business.
A fuller review looks at the vendor itself, not only the document it publishes.
