Why 39% of AI Systems Disagree on Brand Recommendations (and What It Means for GEO)

Why 39% of AI Systems Disagree on Brand Recommendations (and What It Means for GEO) Here’s a scenario that’s way too familiar if you work in digital marketing. You optimise your content, polish your brand messaging, and invest hours testing keywords to hit all the right notes for search engines. Then you ask AI systems like ChatGPT and Google AI Overview simple recommendation prompts about your industry , and they give wildly different answers. Honestly, it’s enough to make you question if any of that hard graft made a difference. My theory? It’s not that your efforts were wasted. It’s that AI systems recommendations operate in a fragmented ecosystem where data sources, algorithms, and context collide, often with bizarre results. And the 39% divergence rate between ChatGPT and Google AI recommendations from our data proves it. Brands visible on one platform are frequently invisible on another , for the same query. So what on earth is going on here? And more importantly, if AI can’t even agree on which brands to recommend, how do we fix it? Buckle up, because we’re diving into this problem headfirst, with real examples, hard numbers, and actionable answers. A split-screen comparison of ChatGPT and Google AI Overview responses to a query The 39% Divergence Problem: Why AI Systems Don’t See Brands the Same Way Let’s start with the hard stats. Across our dataset of 2,641 prompts run across 30 businesses, 39% of the time ChatGPT and Google AI Overview gave conflicting recommendations. That’s not a small discrepancy. You’ve got brands that shine on ChatGPT with a mention rate of 74.4%, only to plummet to 44.1% when tested on Google’s AI. The gap widens further with providers like Perplexity, where mention rates drop to 40.6%, and Claude, which for all its niche uses, still gives consistently different answers from Google AI Overview. Why does this divergence happen? In my experience, it’s down to three main factors: Data Training Sources: Different AI systems are trained