A new study in MIT Sloan Management Review, published October 13, argues that brands face two problems in AI search, not one: being seen, and being described correctly. Its lead case shows how a brand can perform well in classic SEO while AI assistants repeat wrong facts about it.
What happened
The article, "The New Rules of Brand Visibility in AI Search", is by Shahper Richter and Alex Richter of the University of Auckland Business School, Wietske Van Osch of HEC Montréal, and Alla Lvovich of the agency Atlas Digital. The authors interviewed 15 senior marketing and digital leaders across financial services, SaaS, retail, media, higher education and professional services, and studied one regulated investment provider in depth.
That provider, a New Zealand investment platform the authors call Stakewise, saw ChatGPT referrals grow to about 700 sessions a month by mid-2025, up from a handful a year earlier. Those visitors bounced less and viewed more pages than other referral traffic. Its keyword rankings and click-through rates were strong.
Yet the team could not see what AI systems were telling customers. When they audited the answers, they found errors. A widely repeated claim said its KiwiSaver product offered five funds; it offers seven. Fee structures were described wrongly, a protective safeguard was presented as a restriction, and features were mixed up with a separate product. According to the authors, the case traced the errors back to specific source URLs and corrected the underlying content.
The study draws three themes from its interviews: blind spots in AI visibility metrics, inertia in established SEO workflows, and underestimating how fast AI can amplify inaccurate brand information.
Why it matters
Most dashboards still measure clicks. AI answers happen before the click. As the authors write, "Conventional analytics showed the traffic arriving, but not the conversations happening before the click."
Separate research points the same way. In a study reported by PPC Land on October 5, Lily Ray, vice-president of SEO and AI search at Amsive, looked at 100 "best B2B software" queries. In Google AI answers that cited vendors' own "best of" pages, the brand that ranked itself first was left out of the recommendation 83% of the time in her September run, up from 69% in June. She also found AI Overviews citing 38% fewer of these vendor-written lists than in June. PPC Land notes these are her own measurements on a small sample, not independently audited.
Taken together, the message is plain. AI systems decide what to say about a brand from what they can find about it, and ranking yourself first on your own page is no guarantee of being recommended.
What this means for brand and marketing leaders
Audit the answers, not just the rankings. Ask ChatGPT, Gemini, Perplexity and Google's AI features the questions your buyers ask. Record what they say about your prices, products and claims. This matters most in regulated fields such as BFSI and healthcare, where a wrong fact is a compliance risk as well as a lost sale.
Fix the sources. When an answer is wrong, find the pages it draws on, your own and other sites', and correct them. Keep your facts the same everywhere they appear.
Earn mentions rather than claim them. Lily Ray's data suggests self-ranking lists do little for your own recommendations. Clear, accurate product pages and credible coverage from others are a safer bet.
Add new metrics. Track how often you are cited, how accurately, and next to which competitors, alongside traffic. This is the work of generative engine optimization, and it is the focus of ADFLU's autonomous SEO and GEO service.
"Finding out meant auditing the answers themselves." MIT Sloan Management Review
The bottom line
AI search has added a second test to visibility: accuracy. Brands that audit and correct what AI says about them will compete on facts. Those that do not may be losing customers to answers they never see.
