Treadstone Associates
Case File · AI Tools & Workflow

An AI translation that changed the meaning

Anonymised, illustrative composite. A one-click AI translation turned “sold as-is, buyer to verify the heating system” into “heating included and in working order” — in the French version only.

Treadstone Associates · Updated 2026

At a glance

  • • Ottawa-area listing, English MLS remarks translated to French with an AI tool for a bilingual market.
  • • The English remarks read “sold as-is, buyer to verify heating system”; the AI French translation read as “chauffage inclus et fonctionnel” — heating included and working.
  • • A francophone buyer's agent asked a clarifying question that didn't match the English data sheet, surfacing the discrepancy before an offer was written.
  • • The furnace itself needed an estimated $5,400 replacement — a cost the French version implied was the seller's problem, not the buyer's.

The situation

A listing in a bilingual Ottawa-area market ran an AI translation tool over the English MLS remarks to produce a French version for francophone buyer traffic — a routine step for the listing agent's office, done in seconds rather than sending it out for professional translation on every listing.

The problem

The English remarks were specific and protective: “Sold as-is, no warranties; buyer to verify heating system.” The AI-translated French version came back reading, in substance, as though heating was included and in working order — the “as-is, buyer to verify” qualifier collapsed into a plain statement of fact about the system's condition. Nobody proofread the French output against the English original before it went live on the listing.

The numbers

A francophone buyer's agent, working from the French remarks, asked the listing agent a clarifying question about the heating system's age — a question that only made sense if the buyer believed the system's working condition was already represented, not something they still needed to independently verify. That mismatch against the English MLS data sheet, which listed the furnace's age but made no condition claim, is what flagged the discrepancy before any offer was drafted. A subsequent inspection, run before an offer rather than after, priced a full furnace replacement at roughly $5,400 — a real cost the French remarks had implicitly signalled wasn't the buyer's to worry about.

The rule that decided it

Competition Act s.52 and s.74.01(1)(a) make no exception for which language the misleading version was published in — a representation to the public that is false or misleading in a material respect is caught regardless of translation, and a brokerage running listings in two languages exposes itself in either one it gets wrong — the same deceptive-marketing-practice exposure a doctored listing photo carries. RECO Bulletin 5.1's “factually correct, accurate, and verifiable” standard likewise applies to whichever version a given buyer actually reads. This case sits one step past the AI-image-edit case above: there, the exposure was purely a marketing-accuracy question caught before any offer. Here, the mistranslation had already shaped a buyer's negotiating assumptions before the discrepancy surfaced — a live APS-interpretation risk, not only an advertising one, because a buyer who'd made an offer relying on the French version could reasonably argue the seller represented the heating system as sound.

The outcome

The listing agent pulled the French remarks the same day, replaced them with a direct, professionally checked translation matching the English "as-is, buyer to verify" language exactly, and added an explicit special condition to any future offer template on this listing confirming the property's as-is status in both languages, not just the one the buyer happened to read first. The buyer's agent proceeded with a revised offer that priced the furnace risk in rather than assumed it away, and the deal closed without a dispute over what had been represented. Both versions now also state the financing and inspection conditions identically, word for word, rather than trusting a translation tool to keep two legal documents in sync.

What it would have cost otherwise

Had the buyer's agent not asked the clarifying question, and instead written an offer on the French version's plain reading, the realistic failure mode was a firm deal followed by a post-closing furnace failure and a buyer pointing directly at the seller's own listing remarks as a representation of working condition — in their own language, published by the brokerage. That is a materially harder case for the seller to defend than a straightforward as-is sale ever should have been, and the $5,400 furnace cost would have been the smallest number in a dispute over whether the listing itself created a warranty the seller never intended to give.

The tell

The tell was a buyer's agent's question that didn't fit the English data sheet — a mismatch between what one side of a bilingual listing implied and what the other side actually stated. Any listing running in two languages is only as accurate as its least-checked version; a straight back-translation of the AI output into English, compared side by side against the original remarks before publishing, would have caught the dropped qualifier in under a minute.

Takeaways

  • • Competition Act ss.52 and 74.01(1)(a) apply to whichever language version a buyer actually reads — translating a listing doesn't translate away the accuracy duty.
  • • An AI translation can quietly drop a qualifier ("as-is, buyer to verify") and assert the opposite claim as plain fact — check the translated version against the original before publishing, not after a buyer relies on it.
  • • A mistranslated condition claim is riskier than a mistranslated marketing line — it can shape a buyer's negotiating position and become an argument about what the seller represented, not just an advertising-accuracy problem.
  • • Bake the as-is language into the special conditions of the offer itself, in both languages, rather than relying on the listing remarks alone to carry it.

See where AI pays off first in your business.

A 30-minute call is enough to tell you whether AI pays for itself here.