CREA’s AI Guidance: Transparency, Accuracy, and Accountability, in Code

On June 1, 2026 the Canadian Real Estate Association published CREA Supports Responsible AI Use in Real Estate and a standing Artificial Intelligence page . The document is short. It says the use of AI must be guided by transparency, accuracy, and accountability; that consumers and REALTORS® should clearly understand when AI is materially influencing information, communications, or representations; that the adoption of AI does not alleviate the professional responsibilities of REALTORS®; and that REALTORS® remain fully accountable for the information, advice, and services they provide. It points to two existing articles of the REALTOR® Code, commits CREA to working with boards, associations, industry partners, and REALTOR.ca, and promises resources and best practices to come.

CEO Janice Myers framed it as balance: innovation must be balanced with responsibility, and CREA’s job is to help REALTORS® adopt AI confidently in ways that match the standards Canadians expect. Chair Garry Bhaura added the point that will matter most in disputes: the human expertise, professional judgment, and accountability REALTORS® provide remain essential. Neither quote is anti-technology. Both are anti-abdication.

Why principles instead of rules? Partly jurisdiction: CREA writes the Code and runs REALTOR.ca, but licensing and discipline belong to the provinces. Partly timing: the tools change monthly and a rule written for chatbots in 2024 would already miss agentic systems that act. Principles that attach to existing Code articles age better, and they leave room for the specific rules CREA controls, which arrived two months later in the DDF®.

Most coverage of the CREA statement treated it as advice to agents. We think it is better read as a requirements document, because a REALTOR® cannot personally deliver transparency, accuracy, or accountability for output generated by software they do not control. The software has to make each principle possible.

The word doing the most work is materially. CREA did not say every spell-check must be disclosed. It said the consumer should understand when AI is materially influencing what they are told. That is a judgment, and judgments need inputs. A system that cannot tell the difference between correcting a typo and inventing a comparable sale cannot support the REALTOR® in making it. The technical answer is to classify influence at generation time: drafted, summarized, calculated, recommended, or merely formatted. The disclosure decision then has something to run on.

CREA did not write an AI article into the REALTOR® Code. It pointed at two that exist. Article 13 governs advertising content and accuracy; Article 15 governs advertising claims. The message to members is that an AI-written listing description, neighbourhood summary, or market claim is advertising like any other, and the REALTOR® who publishes it answers for it under the same standard as a description they typed themselves.

This mirrors NAR’s approach south of the border, where Article 2 of the Code of Ethics (no exaggeration, concealment, or misrepresentation of pertinent facts) is cited as the rule for AI-enhanced content, and it mirrors the regulators. Canada’s Competition Act prohibits false or misleading representations to promote a product or business interest; provincial real estate rules, from BCFSA’s Rules sections 40 and 41 to Ontario’s TRESA advertising requirements, prohibit false or misleading advertising. None of them contain an AI carve-out. CREA is telling members what the regulators would tell them anyway: the tool changes the workflow, not the duty.

The practical consequence for listing content is a verification loop. A description that says renovated kitchen, steps to transit, or lowest price on the street is a claim, and the REALTOR® must be able to substantiate it. That means the AI drafting it should be pulling facts from the listing record, the board’s data, or the REALTOR®’s own notes, and flagging anything it could not source. Fluent prose with invented facts is the single most common way an AI puts a REALTOR® offside Article 13, and it is entirely preventable at the system level.

Canada has no AI-specific law in force. The federal Artificial Intelligence and Data Act died with the 2025 prorogation, and the June 2026 national AI strategy that TRREB welcomed is a policy program, not legislation. That absence is often misread as a vacuum. It is the opposite: because there is no AI statute, every existing rule applies with full force, and the provincial real estate regulators have been filling in the specifics.

Two features of this stack matter for anyone building or buying real estate AI in Canada. First, the strictest written guidance is provincial, and BCFSA’s is the most detailed in the country, so a national product should be built to the BC standard. Second, the data rules and the conduct rules come from different bodies, CREA and the boards for listing data, regulators for conduct, privacy commissioners for personal information, which means one workflow can be governed by four documents at once. A system needs jurisdiction awareness, not a single compliance switch.

Here is the translation we use internally. It is deliberately concrete, because the gap between a principle and a shipped control is where most compliance failures live.

The frontier model providers have converged on the same requirements from the other direction. Anthropic’s usage policy treats housing eligibility decisions and legal interpretation as high-risk uses that require a qualified professional’s review and a disclosure to the consumer that AI is involved, and prohibits presenting AI output as human-generated. OpenAI’s October 2025 update prohibits tailored legal advice without a licensed professional and automating high-stakes housing decisions without human review. Their agent products refuse to bypass CAPTCHAs or handle credentials, and hold the user responsible for sites’ automated-access terms. A Canadian real estate AI that ignores CREA’s three principles is therefore also fighting its own model provider, which is a fight it loses quietly, one refused task at a time. Compliance has become the interface through which the capable models will actually work.

Our LAIRE framework (Layering Agentic Intelligence on Real Estate, formerly RAILS) was built on capability without custody: the AI may be granted narrow, identity-bound, auditable capabilities, while listing, client, and transaction data stay in the systems that own them. Version 2.0, published today, adds a regulatory-alignment section. The CREA mapping is direct.

CREA says it is developing resources and best practices, and its collaborators include the REALTOR.ca team, which became a for-profit subsidiary in January 2025 and has signalled AI ambitions through its RBC partnership. The sequence so far, principles in June, data terms in August, is consistent with a platform preparing to ship AI features on its own terms. We expect the next documents to be more specific: guidance on disclosure wording, on what may be entered into which tools, and, eventually, the agentic access rules the DDF® update still lacks, covering how an AI acting for a member is identified, permitted, logged, and revoked.

Boards will not wait. TRREB’s statement welcoming the national strategy tied AI directly to MLS® data integrity and to PropTx, and PropTx’s image rules already bind most Ontario listings. The likeliest path is the one we described in our DDF® analysis : presentation rules first, access rules second, with the organizations that already operate as if the access rules existed best placed when they arrive.

If you are a REALTOR®: adopt the three principles as a checklist for every tool. Can it tell you, and your client, when AI materially shaped an output? Does it show you where a fact came from? Does it make you approve the exact version before it goes out, and keep a record that you did? A tool that fails any of the three is asking you to carry accountability it has not given you the means to discharge.

If you manage a brokerage: write the policy now, using BCFSA’s managing-broker considerations as the template even outside BC: approved tools, prohibited inputs, consent for client data, labelling rules for media, review before release, vendor due diligence, and supervision. Ask vendors for the telemetry that lets you supervise. CREA’s accountability principle lands on the brokerage as much as the licensee.

If you build technology: implement disclosure events, provenance, exact-version approval, and jurisdiction profiles as product features. Build to the BCFSA standard nationally, treat CREA’s DDF® terms as the template for every protected data class, and read the model providers’ usage policies as requirements. That is how we built Homies, and it is why a statement like CREA’s reads to us as a description of the product rather than a constraint on it. We hope the rest of the industry builds the same way.