AI can model pricing scenarios, draft talking points and summarise comparable sales in seconds. None of that is the same thing as sitting across a kitchen table from a nervous seller weighing two offers, which is where a negotiation actually happens.
Key takeaways
None of what follows is an argument against using AI in a negotiation-adjacent workflow. Modelling how a seller might respond to a low offer with a short closing, drafting a first pass at counter-offer language, or summarising the last six comparable sales in a neighbourhood are all things AI does quickly and reasonably well. All of that happens before the table, as preparation. The question this article is about is what happens once you are actually in the room, or on the call, with the other side.
The REALTOR® Code’s Article 3 frames the primary duty to a client as protecting and promoting their interests, and Article 3.9 is explicit that dealing fairly with the other side “does not in any way reduce a REALTOR®’s obligation to fulfill his or her fiduciary duties to a Client and follow the Client’s lawful instructions.” A fiduciary duty is owed by one person to another — it is a relationship of trust and judgment, exercised in real time as circumstances change. An AI tool can suggest a counter-offer number; it cannot weigh, in the moment, that your client just went quiet when the other agent mentioned a firm closing date, and adjust the next sentence accordingly. That read, and the judgment call that follows it, is the part of the job a transcript of the conversation cannot fully capture after the fact, let alone a tool that was never in the room.
The federal, provincial and territorial privacy commissioners’ own generative-AI principles set a standard worth applying directly here: “consider whether the use of a generative AI system is necessary and proportionate… the tool should be more than simply potentially useful. This consideration should be evidence-based and establish that the tool is both necessary and likely to be effective in achieving the specified purpose.” Applied to a live negotiation, that test is unforgiving: an AI tool is not necessary for reading a client’s hesitation, and it is not effective at building the trust that makes a client willing to take your advice on a counter-offer under time pressure. Those are the two things that actually decide what happens at the table, and neither is a text-generation problem.
A constructed illustration of where this bites: a scripted counter-offer, generated in advance from the comparable-sales data, assumes the seller will react to numbers. In the room, the seller’s real hesitation turns out to be about the closing date clashing with their child’s school year, not the price at all — something no amount of market data would have surfaced, and something only a live conversation revealed. The prepared script is still useful as a starting point; it is not a substitute for noticing the actual objection and responding to it.
Canada’s own voluntary code for advanced generative AI systems includes a manager-level obligation to “ensure that systems that could be mistaken for humans are clearly and prominently identified as AI systems,” which is a voluntary commitment, not a binding law, but it reflects a real judgment: the distinction between a person and a system is something a consumer is entitled to know, not a detail that can be blurred for convenience. Extend that logic to a negotiation, and the reason an AI summary of “likely seller motivations” cannot replace being in the room is the same reason the code treats human-versus-AI as worth disclosing in the first place — it changes what the other side is actually relying on.
None of this is a case against AI in a negotiation-adjacent workflow, and it is not a case for slower service either. It is a case for being precise about which half of the job a tool touches. Preparation is a drafting and research problem, and AI is a genuine speed gain there. The moment at the table is a judgment and trust problem, and no amount of better prompting turns a language model into the thing a client is actually paying for when they hire a REALTOR® to represent them in a negotiation.
CREA’s AI guidance states its underlying premise directly: AI use “must be guided by transparency, accuracy and accountability… the adoption of AI does not alleviate the professional responsibilities of REALTORS®,” who “must remain fully accountable for the information, advice and services they provide to clients.” A negotiation outcome is advice and service in its most consequential form. Whatever AI contributed to the preparation, the accountability for how the conversation actually went sits with the REALTOR® who was in it.
Related: the short answer to whether AI replaces the agent’s role entirely, how liability is actually allocated when an AI tool contributes to an error.
Preparation, not the live exchange — comparable-sales research, a first draft of counter-offer language, or a summary of a competing offer’s terms. Anything that requires reading the other side’s reaction in real time and adjusting is the part that stays human.
Most will not, provided the outcome reflects genuine judgment rather than a script read without adaptation. What clients notice is whether their agent responded to what actually happened in the room — which is exactly the part AI cannot do for you.
Using AI to prepare is not itself the problem — treating AI-generated talking points as a substitute for judgment during the actual negotiation is. The standard is what a prudent REALTOR® would have done in the room, and preparation tools do not change that standard.
That is a related but separate question about what an AI assistant may say on a call before a person is involved, covered elsewhere in this hub — the rules there are about consent and disclosure on the call itself, not about negotiation judgment.
The preparation itself is not the concern — REALTORS® have always used scripts, checklists and coaching. The concern is treating the prepared material as sufficient on its own, without adapting it to what actually happens at the table. A script a human wrote has the same limitation an AI-drafted one does if it is read without adjustment.
A short call is enough to map the workflow without overselling what any tool can actually do at the table.
