The Psychology of AI Recommendations: How Cognitive Bias Shapes LLM Visibility and Revenue Impact

The Psychology of AI Recommendations: How Cognitive Bias Shapes LLM Visibility and Revenue Impact I’ll get straight to it: AI recommendations aren’t as neutral or objective as people like to think. They’re algorithmically driven, yes. But the decisions large language models (LLMs) like ChatGPT or Google AI Overview make are heavily influenced by cognitive biases baked into their training data. My theory? Understanding these biases is going to be critical for brands trying to optimise their visibility in an AI-dominated world. Because if you don't know why AI agents prefer one brand over another, you're flying blind. And at the end of the day, blind spots cost revenue. Here’s the deal: across the 2881 prompts we ran, visibility outcomes varied wildly. Some brands were consistently recommended no matter the query, while others, some with clearly high-quality offerings, barely got a look in. Why? It’s not all down to algorithms. Human-like biases creeping into AI models play a major role. Abstract illustration of a brain interacting with AI nodes Bias 1: Authority Bias Ever been swayed by a "Doctor recommends" label or assumed a well-known publication must be trustworthy? That’s authority bias in action, and AI models aren’t immune. In our tests, brands with established reputations or affiliations to major institutions consistently earned higher mention rates. Take Viking Pumps and IPU, they’ve got a 100% visibility rate across our network. One explanation could be their perceived authority within their niche industries. When you're a recognised name, there's a certain weight of credibility that algorithms latch onto. Related reading: this piece about the cross-provider visibility gap: why ai systems disagre... . But here’s the kicker: it’s not always logical. Consider Iconsys 2. They’re a solid contender in manufacturing and industrial automation, yet their visibility sits at a painful 6.7%. Our experiments suggest that without strong authority signals (think industry