Who does ChatGPT name for an emergency plumber in Brampton?

The exact wording was “Recommend an emergency plumber in Brampton, Ontario. Name specific companies.” Three fresh sessions ran through gpt-5.4-mini on 2026-09-23, en-CA locale, browsing turned off, nothing else touched between them, and we asked no follow-up in any of the three.

This prompt asked directly for names, unlike some open-ended questions elsewhere in this panel, phrased more like “which practices would you recommend and why.” That directness may be part of why this run, unlike the Toronto dentist test, produced actual business names in every one of the three tries.

Run 1 named: Mr. Rooter Plumbing of Brampton, Drain King Plumbers, Dima Plumbing, Priority Plumbing & Drains, Everest Drain & Plumbing, and Mike The Plumber.

Run 2 named: Mr. Rooter Plumbing of Brampton, Drain King Plumbing and Drain Service, Action Restoration / Action Plumbing, Priority Plumbing & Drains, ASAP Plumbing & Drains, and Superior Plumbing & Heating.

Run 3 named: Drain King Plumbers, Mr. Rooter Plumbing of Brampton, H2O Plumbers, The Irish Plumber, Rooter-Man, Priority Plumbing & Drains, and City Home Comfort.

Two names showed up in every single run, worded identically each time: Mr. Rooter Plumbing of Brampton and Priority Plumbing & Drains. Those are 2 of the 14 distinct exact company-name strings in this dataset, and the only 2 the model returned, unchanged, three times out of three.

A plumbing brand referred to as Drain King also showed up in all three runs, but the model did not use a consistent exact name for it: “Drain King Plumbers” in runs 1 and 3, “Drain King Plumbing and Drain Service” in run 2. We are noting that as one likely brand named two different ways rather than a clean third repeat, because we cannot confirm from the model's output alone that both phrasings point to the same legal business. The remaining 11 distinct strings, 11 of the 19 total mentions, showed up in exactly one of the three runs and nowhere else.

This shows that, for this specific prompt, on this one date, with browsing off, two names were stable and the majority of the list was not. It does not tell you why those two names were stable, whether it reflects review volume, brand recognition, franchise scale, or something else in the model's training data, and we did not test for any of that.

It also does not tell you what a browsing-enabled ChatGPT session would return, since that could pull from live search results and name completely different companies. Nor does it tell you what another assistant, or a different phrasing of the same request, would produce. Three runs on one date is a small sample; it shows a pattern within this sample, not a durable ranking.

While the company names varied, the practical guidance around them did not: every run told the user to shut off the main water valve immediately for a major leak or burst pipe, to confirm whether a company offers true 24/7 dispatch specifically to Brampton, to ask about after-hours or dispatch fees up front, and to call more than one company when it is a genuine emergency.

That consistency is worth noting on its own. Across three runs where the specific business names shifted by roughly 75 percent, the underlying consumer advice the model gave stayed essentially the same each time.