Ez-AI has published a study examining how six AI assistants recommend AI service providers in New Zealand. It found that results varied widely between assistants.
The Auckland consultancy tested 22 buyer-style questions across ChatGPT, Claude, Perplexity, Gemini, Grok and Amazon Nova, with web search enabled. It also tested GPT and Claude without web search. In total, it logged 176 answers and recorded every cited source page.
The research found that a small group of online rankings and company websites shaped many recommendations. The three most-cited pages were all articles listing the best AI consultants in New Zealand, and each appeared in 12 to 14 answers.
Deloitte was the most frequently named firm, appearing in 41 of the 176 answers. Datacom appeared in 39, while PwC, Enterprise DNA and AI New Zealand were each named in 36. Harkness AI and Automate AI followed on 25 each, and Theta appeared in 23.
The pattern shifted by assistant. ChatGPT tended to name larger established firms, with Datacom and Theta each appearing in 10 of its 22 answers. Grok, Claude and Amazon Nova relied more on published lists and agency directories, and named more specialist firms. Gemini more often drew on Google results and government sources.
When web search was disabled, GPT and Claude mostly named larger firms, including Datacom, Deloitte, PwC and Accenture, along with a smaller number of long-established New Zealand technology businesses. This suggests that the source material available to each model, and the use of live web search, can materially affect which providers are surfaced to business users.
Government sources also featured strongly. Business.govt.nz was cited in 28 answers across 14 of the 22 questions, often before any consultant was suggested, especially when a prompt pointed to advice for small businesses.
Patchy answers
Several prompts produced limited or inconsistent recommendations. When asked which New Zealand agency could help a business get found on ChatGPT and Gemini, none of the assistants named any of the 39 firms tracked in the study, and no agency was cited by more than three of the six assistants.
On another prompt asking for an AI consultant for a law firm, Claude returned no firms and no sources. Asked about AI automation for an insurance broker, ChatGPT cited only government and regulator websites and named no provider.
The counts were based on a tracking list of 39 New Zealand and international firms identified in earlier runs. Businesses outside that list were not included in the firm totals, although pages published by those businesses could still appear in the source counts.
Ez-AI itself was part of the test set and appeared in 28 of the 176 answers. Almost all of those mentions came from Grok, Amazon Nova, Claude and Perplexity, with none from ChatGPT.
The findings add to broader concerns for professional services firms and digital agencies as generative AI tools become another route for buyers searching for suppliers. If recommendation patterns are shaped by a narrow set of listicles, directories and indexed pages, firms may need to pay closer attention to where they appear online rather than relying only on their own websites.
Skene Bennellick, Founder of Ez-AI, said businesses should treat AI-generated recommendations with caution. "If you're hiring, an AI shortlist is a starting point, not a verdict. Open the list the assistant is quoting and check who wrote it, because several of the most-cited lists are written by a consultancy that ranks itself first. And ask two or three assistants. In this test they overlapped far less than you would expect," Bennellick said.
He said visibility in third-party sources mattered more than self-description. "Getting onto the lists and directories they cite, having independent reviews they can see, being indexed by Google, Bing and Brave, and stating plainly what you do, where, for whom and at what price: those are what decide who gets named," Bennellick said.