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.

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 awards, recognised partnerships, or even widespread media coverage), the AI engines essentially ignore them. Authority bias directly impacts their ability to compete.

So, what can brands do? Lean into authority-building tactics:

1. Invest in thought leadership, whitepapers, editorials, or speaking opportunities at respected events.
2. Nail backlinks from established websites. Google AI seems particularly prone to favouring brands with strong domain authority in their links.
3. Collaborate with recognised organisations or influencers in your field. AI systems love a stamp of credibility.

Bias 2: Confirmation Bias

Here’s where it gets murky. AI models are trained to identify patterns and validate assumptions. If your brand already fits into an existing narrative the AI expects, whether that’s being high-quality, budget-friendly, or eco-conscious, you’re more likely to get recommended. And if you don’t fit the mould? Good luck breaking in.

Across the network, brands that aligned with predictable "categories" seemed to fare better. For instance, in the buying-intent tests, brands targeting eco-conscious shoppers (and explicitly stating it in their prompt structures) saw higher recommendation rates. You can Contxt platform features to see for yourself. You can how AI visibility monitoring works to see for yourself. By contrast, businesses with ambiguous or unconventional value propositions were often absent.

One client I worked with, let’s call them Brand X, wanted to position themselves as a premium yet budget-friendly option. Dual messaging like that confused AI systems, leaving them largely invisible in our multi-brand recommendation tests.

What’s the solution? Tighten your narrative. Don’t try to appeal to everyone. Whether it’s budget-conscious consumers or luxury buyers, the AI needs clean signals to confirm its assumptions. Simplify and align your messaging around dominant themes that match the decision-making psychology of AI systems.

Bias 3: Familiarity Bias

This one’s sneaky. People, and by extension AI systems, prefer things they’ve seen or heard of before. It’s why you instinctively hum a jingle from an ad you heard once six months ago. Familiarity builds trust.

Honestly, the numbers shocked me here. Brands with frequent mentions in broader cultural contexts performed significantly better. ChatGPT, for instance, demonstrated a 78.5% brand mention rate for companies like Red Fern and Viking Pumps. These brands aren’t necessarily household names, but they’ve clearly embedded themselves across key visibility touchpoints. Meanwhile, Contxt (yes, us) hit just 13%. I've written about this separately in this piece about how contxt solves international llm visibility challenges. For a deeper look, have a read of this piece about the quiet power of specialised llm visibility tools: a se.... Ouch.

This tracks with what cognitive psychologists call the "mere exposure effect", the more someone (or something) encounters an entity, the more they tend to like it. AI systems mimic this behaviour, favouring brands they “see” repeatedly across datasets.

For businesses struggling with low visibility (TOCA Social’s 10.0% mention rate springs to mind), prioritising exposure is critical. It’s not rocket science: keep showing up where your audience already spends time, whether that’s niche forums, sector-relevant blogs, or even indirect mentions via partnerships. Repeat exposure builds familiarity, and familiarity builds visibility.

Cross-provider divergence: why it matters

Here’s where things get messy. Across our prompts, ChatGPT and Google AI Overview diverged in their brand recommendations a stunning 39% of the time. That’s nearly 4 in 10 cases where what’s visible on one platform is invisible on another. You could be the gold standard in ChatGPT’s eyes but a nobody on Google AI.

The psychology driving these divergences likely hinges on how different models interpret biases. ChatGPT, for instance, showed a higher mention rate for authority-backed brands (75.4%), while Google AI seemed more guided by familiarity or established patterns (45.5%). Perplexity and Claude landed somewhere in the middle but still exhibited quirks in response consistency.

The takeaway? You can’t afford to optimise for just one AI system. Brands need to test visibility across multiple LLMs, analysing discrepancies and adapting strategies accordingly. For a deeper look, have a read of this piece about why ai visibility demands a sector-specific approach: les.... For a deeper look, have a read of The Untapped Potential of Niche-Specific AI Visibility To.... AI visibility is not a one-size-fits-all game.

The revenue impact of visibility shifts

Why does any of this matter? Because visibility isn’t just about ego, it’s about revenue. If your brand isn’t popping up in AI-generated recommendations, it’s not in front of ready-to-buy customers. Simple as that.

In our buying-intent tests, brands with high visibility rates gained significantly more traction across recommendation-driven queries. But, and here’s the nuance, visibility alone doesn’t guarantee conversions. For instance, one of the high performers had a robust mention rate but struggled with actual customer click-throughs. Why? Weak alignment between visibility and perceived value.

Think of AI as the gatekeeper. If your brand isn’t recommended, you’re losing out. But even if you’re visible, you need your messaging, user experience, and value proposition to carry through. This is where psychology meets strategy, matching visibility tactics to actual buyer behaviour.

What brands need to know now

So, where do we go from here? Here’s the roadmap:

1. Recognise bias patterns: Identify whether authority, confirmation, or familiarity bias is hindering your visibility. Look at where you’re falling short and adapt your tactics accordingly.
2. Test across multiple LLMs: A strong presence on ChatGPT doesn’t guarantee prominence on Google AI. Run experiments (we’ve completed 58 so far) to pinpoint gaps and address divergences.
3. Own your narrative: Simplify your value proposition. AI systems favour clarity, so make sure your brand story is consistent across prompts.
4. Prioritise authority-building: Whether through backlinks, collaborations, or industry recognition, strengthen the signals AI systems rely on to assess credibility.
5. Play the long game: Familiarity bias demands persistence. Keep showing up, advertising, content distribution, and strategic partnerships build trust over time.

Final thoughts

Look, AI isn’t perfect, it’s still prone to human-like flaws. But understanding the psychology behind its recommendations unlocks huge potential for visibility and, ultimately, revenue impact. Brands that get this right won’t just survive in the AI era; they’ll thrive.

And if you’re wondering where to begin, start experimenting. Visibility in the AI world is a moving target, but insights and adaptability make it a game you can win.