CREA Just Opened the DDF Door to AI—Carefully

In August 2026, CREA notified members of a change to the DDF® terms and conditions that, for the first time, explicitly permits DDF participants to offer artificial intelligence, augmented reality, virtual staging, and similar tools on listing content. Since its launch in 2012, the DDF—the national, permission-based feed that powers listing display across members’ and partners’ websites—has been display-only by rule: no derivative use, no redistribution, active listings only, watermarks on. It had no written answer to the question every technology provider has been asking since 2023: can an AI touch this data at all?

The answer is now a conditional yes. The update introduces a defined concept CREA calls an augmentation: a visual, informational, or interactive enhancement shown using technology—virtual staging, paint-colour visualization, renovation renderings, AI- or AR-generated views—delivered as an alternate presentation a consumer can choose to see. The permission comes paired with safeguards stricter than most US MLSs have published, and it lands alongside CREA’s broader position on AI , which anchors REALTOR® use of AI to transparency, accuracy, and accountability—and to the existing advertising articles of the REALTOR® Code (Articles 13 and 15).

The timing detail worth knowing: the notice circulated to members carries no announcement date and no effective date, and the most recent publicly posted DDF Policy and Rules document still shows a January 2024 revision. The safeguards below are the design target; confirm timing with CREA and your board before shipping.

The updated terms hang together as one design pattern: the original listing is the record, and anything an AI produces from it is a labelled, temporary, revocable view. Eight requirements do most of the work.

Responsibility is assigned just as clearly. Participants must either build augmentation functionality themselves or use third-party technical providers—CREA and REALTOR.ca provide no augmentation tooling of their own—and participants carry full responsibility for those providers’ compliance, on top of existing board, association, MLS®, advertising, and co-branding rules. If a vendor’s virtual-staging model quietly trains on the photos it processes, that is the participant’s problem, not an abstraction.

One more reading note: the training prohibition is categorical. It does not distinguish fine-tuning from retrieval indexes built for other purposes, and it explicitly bars content from being used to improve any other product. Providers should assume any persistent secondary use of listing content is out of bounds until CREA says otherwise.

Everything in the update governs how AI may present listing content to a consumer on a participant’s website. Almost nothing in it governs how AI systems may work with listing data—which is where the real questions in agentic real estate live. The update is silent on:

• Agentic queries: whether an AI assistant may run bounded listing searches against DDF data on a realtor’s behalf, and under whose credentials.
• Custody: whether raw feed content may enter a model’s context window at all, or must stay inside an approved participant application.
• Derived work: CMAs, market statistics, and pricing analysis built from feed data—long restricted under the display-only baseline, and not revisited here.
• Identity and audit: how an AI acting for a member is identified, logged, rate-limited, and revoked.
• Write paths: showings, offers, and transaction actions that touch board systems—out of scope entirely.

None of this is a criticism of the drafting. Virtually staged photos and AI-enhanced media were the live consumer-facing issue—by May 2026, one study of roughly 72,000 REALTOR.ca listings estimated that about 37% of Canadian listing descriptions were likely AI-written—and CREA addressed the presentation problem with unusually clean mechanics. But technology providers should not read the update as permission for AI listing search, AI market analysis on feed data, or agent-to-agent workflows. Those remain governed by the existing permission-based DDF agreements and board rules—which is to say, mostly ungoverned by anything written for AI.

Our RAILS framework argues that safe agentic AI in real estate needs capability without custody: the approved participant application holds the credentials and executes queries server-side, while the AI receives schemas, bounded tools, opaque references, and rendered results with provenance. Measured against that standard, the DDF update covers the presentation layer—labelling, provenance, original-first display—and stops there.

That makes it a genuine step in the right direction. The no-training rule is custody thinking. The overlay-not-replacement rule is provenance thinking. The participant-responsibility rule is accountability thinking. What is missing is the access layer RAILS specifies: agent identity, permissioned query tools, server-side execution, audit trails, and revocation. Until CREA writes those rules, every Canadian AI product that goes beyond presentation is building on interpretation rather than policy—and the conservative interpretation is the only defensible one.

The rules change makes more sense against the corporate backdrop. REALTOR.ca became REALTOR.ca Canada Inc. in January 2025—a wholly owned, taxable, for-profit CREA subsidiary, approved by 77% of members voting—precisely so it could pursue product and revenue moves a not-for-profit could not. In January 2026 it hired a CEO from Kijiji with a track record of shipping AI features, and announced an RBC partnership that both sides say will grow into AI-enabled tools on RBC’s data and analytics.

Yet as of August 2026, REALTOR.ca has shipped no consumer AI feature—while its US counterpart Realtor.com launched an AI search assistant in June. Read together, the sequence looks deliberate: write the data rules first, ship the AI second. The DDF update is CREA laying the legal groundwork for AI on the national platform before anything user-facing exists—presentation rules now, with the harder access rules still to come. For Canadian realtors and proptech builders, that is the signal: the national platform is preparing for AI in public, one layer at a time.

If you are a realtor: expect virtual staging, AI photo enhancement, and similar features to start appearing on DDF-powered websites—legitimately. Hold your vendors to the safeguards: original-first, opt-in, labelled, no training, nothing routed off-platform. If a tool cannot show you its labelling and its data-use terms, it is not compliant, and under the updated terms the compliance failure lands on the participant side, not just the vendor’s.

If you build technology: the update gives you a clean specification for presentation features and a clear red line on data use. Build augmentations as stateless transforms—content in, rendered view out, nothing retained, nothing trained. Keep anything agentic—search, analysis, actions—behind an approved application boundary with the RAILS access pattern, and paper your provider agreements so a participant can demonstrate your compliance to their board on demand.

The organizations that will benefit when the access rules arrive are the ones already operating as if they existed.