What happens when you ask ChatGPT, Google AI Mode, and Microsoft Copilot the same question about KiwiSaver, power companies, or accounting software? A new field study from The Optimisers set out to find out, using real people and real accounts rather than lab conditions.
Over a 14-day window in July 2026, 76 New Zealanders asked all three AI engines the same three commercial questions, then repeated the exercise in ChatGPT's Temporary Chat mode. The result: 888 submitted answers that reveal just how inconsistent AI recommendations can be.
The headline finding: the "winning" brand changed depending on which AI engine you asked. KiwiSaver had three different leaders across the three engines. Power companies split just as sharply. Only accounting software converged on a single answer, with Xero named first in every single response. Even switching to a fresh, memory-free chat changed the recommended provider for 69% of participants at least once.
Featuring commentary from Rand Fishkin (Co-Founder of SparkToro and Moz), the report argues that AI visibility isn't a fixed fact but a distribution, shaped by which engine you use, who's asking, and what condition their account is in. For brands and marketers, that means measuring both mentions and lead positions, planning by engine, and treating any single clean benchmark as a starting point rather than the full picture.
Author: The Optimisers, 18th August 2026