Does the Ontario AI disclosure rule apply to a US company with no Ontario office?

The trigger is the employment relationship the posting is advertising, not where your headquarters sits. Ontario's Employment Standards Act reaches work performed in Ontario, and a remote role filled by someone working from Toronto is Ontario work in the ordinary case. A US company that advertises a publicly available posting for a remote role open to Ontario residents is therefore in the frame, even though nothing about the job description mentions Canada. The practical failure mode is not a company that decided the rule did not apply; it is a company whose global postings template was written in one country and syndicated everywhere, so the Ontario-facing version silently omits a statement the jurisdiction requires.

What counts as artificial intelligence for the purpose of the disclosure?

The concept in play is a computer system that uses data and machine methods to generate an output — a prediction, a recommendation, a ranking or a decision — used in screening, assessing or selecting applicants. Read functionally, that captures far more of a modern hiring stack than the phrase 'artificial intelligence' suggests to a recruiter. Resume parsing that scores fit, a knockout question set with a learned threshold, a vendor's 'match score' in an applicant tracking system, a video interview platform that rates responses, and a sourcing tool that ranks candidates in a queue are all doing exactly what the definition describes. A keyword filter your team wrote by hand is a harder call, and the harder call is exactly why the decision belongs in a documented classification rather than in a recruiter's judgement on the day.

What does the statement actually have to say?

It has to disclose the use, in the posting itself, where an applicant reads it before applying. It is a transparency statement, not a technical annex: the requirement is met by plainly telling applicants that AI is used to screen, assess or select, not by publishing model documentation. What gets employers in trouble is not brevity but vagueness that shades into inaccuracy — a line saying AI 'may be used to assist our team' when a vendor score in fact removes candidates below a threshold before a human sees them is a statement whose truth depends on facts the applicant cannot check. Write the sentence against what your ATS configuration actually does, then re-read it after any vendor change.

Do we have to keep a copy of the posting?

Yes, and this is the obligation most likely to be missed because it lives after the hire rather than before it. Copies of a publicly advertised posting, and of any associated application form information, have to be retained for three years after the posting comes down. An employer that advertises through a job board and never keeps its own copy is relying on a third party's retention policy to satisfy a statutory one. Build the archive step into the requisition-close workflow, capture the posting as rendered rather than as drafted, and store the vendor configuration alongside it — because a question about the AI statement is really a question about what the system was doing on the date the posting ran.

How does this interact with NYC Local Law 144 or Illinois' hiring rules?

They stack, and they are not the same shape. Ontario's requirement is a disclosure attached to the posting. NYC's regime attaches an independent bias audit and candidate notice to the use of an automated employment decision tool. Illinois requires notice and consent for AI analysis of video interviews and, separately, limits the use of AI that produces a discriminatory effect in employment decisions. An employer that builds one global 'AI in hiring' notice and assumes it satisfies all three usually fails on mechanism rather than on content: the audit, the timing of a notice before an assessment, and the placement of a statement in the advertisement are different deliverables. Map them per jurisdiction against a single inventory of tools, so that one change to the stack updates every obligation at once.

Does the rule apply to internal postings and to recruiting agencies?

The requirements attach to publicly advertised job postings, which is the distinction to reason from: a role circulated only to existing employees through an internal system is a different thing from an advertisement made available to the public. Agencies complicate the picture rather than remove it, because a posting placed on your behalf is still an advertisement for your role, and the party controlling the screening technology is often the agency rather than the employer. That is a contract problem with a compliance deadline: your agency terms should say who drafts the AI statement, who verifies it against the tooling in use, and who retains the posting record. Left unwritten, both parties assume the other one did it.

Ontario AI Job Posting Disclosure 2026: What US Employers Hiring Canadians Must Say

The Obligation Attaches to the Advertisement, Not the Assessment

Nearly every AI hiring rule written so far attaches to the moment of evaluation. A candidate is about to be scored, so a notice goes out; a tool is in use, so an audit is published. Ontario picked an earlier point in the funnel. The duty lands on the publicly advertised posting, which means it is discharged by a marketing artefact produced by a recruiter, in a template, often weeks before anyone touches the applicant tracking system's configuration.

That placement is the whole compliance problem. The people who know whether a match score gates the pipeline are in talent operations or IT. The people who publish the advertisement are in recruiting, and they work from a saved template that was last edited when the pay-transparency language was added. Nothing in a normal hiring workflow routes a change in vendor configuration back to the wording of the job ad, so a statement that was true in January can be quietly false by June without anybody editing a word.

Who Is Caught Without Knowing It

Employment standards follow the work. A US company with a fully remote engineering team that lists a role as "Remote — North America" is advertising work that can be performed in Ontario, and if it hires someone in Ontario it has an Ontario employment relationship with everything that implies. The company's own mental model is that it hires Americans and occasionally a Canadian; the statute's model is that it advertised a job to Ontario residents and screened them with software.

The same exposure reaches any company using a global job board, because syndication is indifferent to jurisdiction. One posting object is created once and published to a dozen surfaces. If the compliance layer is a checkbox in one country's template, the other eleven surfaces carry a posting that is missing a required statement, and the defect is reproduced automatically with every new requisition.

Build the Tool Inventory Before You Write the Sentence

The statement is one line. Knowing whether it is accurate is a project. Most hiring stacks contain at least three places where a machine-generated output influences who advances, and at least one of them was purchased by someone who no longer works there. Before drafting, list them and write down what each one does to the candidate pool.

Resume parsing and fit scoring inside the ATS, including any vendor "match" percentage.
Knockout and screening questions whose thresholds were tuned from historical outcomes.
Sourcing and outreach tools that rank or recommend candidates before a requisition opens.
Video or written assessment platforms that produce a score, a rating or a ranking.
Scheduling and chat assistants that decline, defer or triage applicants automatically.
Anything a recruiting agency runs on your behalf, which you may never see in your own systems.

For each row, record whether the output is advisory or dispositive. That distinction does not change whether you disclose — the duty is about use in screening, assessment or selection — but it determines whether your sentence can honestly say the tool assists a human reviewer, and it is the first thing a regulator or a claimant's representative will ask you to substantiate.

The Records Duty Is the Part That Ages Badly

Posting-content rules come with a retention rule: keep copies of the publicly advertised posting and the associated application form information for three years after the posting is taken down. Three years is longer than most job boards keep anything, longer than the average tenure of a recruiting coordinator, and considerably longer than the memory of the vendor contract that was in force when the ad ran.

Archive the posting as rendered to the public, not the draft in your template library, and store it with a dated snapshot of the screening configuration it describes. The question that arrives later is never "what does your template say"; it is "what did this posting say, and what was the system doing that week". Only a paired record answers it.

A Workable Sequence

Decide whether any posting you publish can be filled by someone working in Ontario. For most remote roles the answer is yes.
Inventory every tool that touches the candidate pool, naming an owner for each.
Classify each as in scope or out of scope for the disclosure, and write down the reason and the date.
Draft one statement that matches the configuration, and put it in the posting body rather than a linked policy.
Add a trigger so that any change in screening tooling opens a review of the posting language.
Archive each rendered posting for three years, paired with the configuration snapshot.
Amend agency and job-board terms to allocate drafting, verification and retention explicitly.

Questions Hiring Teams Are Asking

We use an ATS but nobody here calls it AI. Do we still disclose?

Ask what the software does rather than what it is called. If the product ranks, scores, recommends or filters applicants using a model built from data, it is doing the thing the rule is about, and the vendor's marketing vocabulary is not the test. In practice the fastest way to settle this is to open the ATS configuration with the person who administers it and walk one requisition end to end: where does a candidate get a number, and what happens to candidates below it? Teams that run this exercise usually find at least one scoring feature that was switched on by default during implementation and has never been reviewed since.

Can we just add the line to every posting worldwide and stop worrying?

A single global statement is a reasonable operational choice and it is not a complete answer, for two reasons. First, the statement has to be true wherever it appears, so if the tooling differs by region an undifferentiated line is inaccurate somewhere. Second, other jurisdictions impose mechanisms a posting line does not satisfy — an independent bias audit, a notice delivered before an assessment, an alternative-process offer. Use the global line as the floor, then keep a short jurisdiction matrix for the extras. The matrix is the artefact that survives staff turnover; the institutional memory does not.

What is the risk if we get this wrong — is it a fine or a lawsuit?

Employment standards regimes are complaint-and-inspection driven rather than litigation driven, so the realistic first contact is an officer asking for records with a deadline, not a statement of claim. The exposure that hurts is cumulative and administrative: a posting practice applied across hundreds of requisitions is a systemic finding rather than a one-off, and an employer who cannot produce three years of postings has a second problem on top of the first. There is also a downstream discrimination angle that has nothing to do with the posting rule — undisclosed screening that produces an adverse pattern for a protected group is a human rights matter in its own forum.

Our recruiting agency writes and places the ads. Are we off the hook?

No, and the agency arrangement usually makes the record-keeping worse rather than better. The posting advertises your role; the screening technology may be the agency's; the copy of the posting sits in the agency's system. Each of those facts sits with a different party, which is precisely how a three-year retention duty quietly goes unmet. Put three lines in the agency agreement: the agency will include the disclosure statement, will notify you of any change in screening tooling within a stated period, and will deliver rendered copies of every posting to you for retention. The third line is the one to insist on, because it is the only one that survives the end of the relationship.

Does telling applicants we use AI hurt our conversion rate?

The available evidence on hiring-funnel transparency is mixed and mostly self-reported, so treat any confident number with suspicion. What is more defensible is the comparison you control: measure application starts and completions on your own postings before and after the change, on the same roles and the same boards, and keep the window long enough to be more than noise. Employers who have run that comparison generally report a small effect swamped by role, pay range and posting recency. If your funnel is genuinely sensitive to the sentence, that is useful information about what else the posting is or is not saying.

Start With One Requisition, Not One Policy

Take the last role you posted publicly and reconstruct it: every tool that touched the applicants, every threshold that removed someone, and the exact text the public saw. That single reconstruction tells you whether your template is accurate and whether you could produce the record if asked.

Almost every team that runs it finds the same two things — a scoring feature nobody owns, and no archived copy of the advertisement. Fix those before rewriting the policy, because the policy is not what a regulator asks to see.