Why 39% of AI Systems Disagree on Brand Recommendations (and How to Fix It)

Why 39% of AI Systems Disagree on Brand Recommendations (and How to Fix It)

Here’s the uncomfortable truth: if you’re relying on AI to help your brand show up in recommendations, there’s a good chance it’s not working as well as you think. You might be visible in one system but invisible in another. And the kicker? Most businesses don’t even realise this is happening.

My theory? It’s because we’ve been so focused on creating content that we forgot to ask a critical question: how do AI systems actually decide which brands to recommend? It’s a messy, imperfect process, and it’s leaving many companies on the sidelines when it comes to AI brand recommendations.

Here’s what I keep seeing: when we run identical prompts across different AI systems, like ChatGPT and Google’s AI Overview, the results vary wildly. Across our dataset of 2,641 prompts, 39% of the time, these systems give conflicting recommendations. In practice, that means a brand visible in ChatGPT might be nowhere to be found on Google’s AI. For businesses trying to manage their digital presence, this is a massive problem.

And it’s not just theory. We’ve got the data to prove it, and strategies to fix it. Let’s break this down.

The 39% Problem: Why AI Systems Don’t Agree

You’d think AI systems would all draw from the same pool of information, right? They’ve got the internet, stacks of training data, and similar mechanisms for processing user queries. But in reality, their priorities, and outputs, are anything but consistent.

In our tests, ChatGPT mentioned brands 74.4% of the time, while Google AI Overview was significantly lower at 44.1%. Perplexity came in at 40.6%, while Claude actually outperformed at 77.1%, although it was tested on a smaller sample size. These disparities aren’t just statistical quirks; they’re meaningful. They mean your brand could show up in one conversation but be completely absent in another. Related reading: From 0% to Cited: What Brands Scoring 80%+ on AI Actually.... Related reading: Why B2B Brands Are Almost Entirely Invisible to AI (And W.... For example, in one experiment with buying-intent prompts, we found a single brand that was strongly recommended by ChatGPT but didn’t even register on Google AI.

Why the difference? It comes down to how these systems are designed. Every large language model (LLM) has its own training data, algorithms, and biases. Some are great at pulling from publicly available sources; others prioritise specific partnership data or aim to minimise perceived “promotional bias”. Related reading: this piece about the two visibility problems every business now has (and w.... Related reading: this piece about how to optimize your website for generative ai systems in.... But here's the thing: does any of that actually matter if AI can't find you?

Brands Falling Between the Cracks

Let’s talk about visibility distribution. Across the 30 brands we track, half enjoy high visibility (70%+ mention rates across prompts). Viking Pumps, IPU, and Red Fern, for instance, all hit 100%. But then you’ve got brands like TOCA Social (10%) and Iconsys 2 (6.7%), barely registering. My own company, Contxt, isn’t exactly smashing it either at just 13%. It’s clear some brands have cracked the code, and others, like us, still have work to do.

The biggest shock? Even within the same industry, disparities are rampant. Viking Pumps dominates across the board, but another manufacturing brand in our dataset is struggling to reach 30% mention rates. And the scariest part is that most companies are completely unaware of how they’re performing. They don’t run experiments. They don’t check cross-platform consistency. They’re flying blind in a system that’s anything but neutral.

Why This Matters More Than You Think

Let’s bring it down to brass tacks. AI isn’t just a fancy tool anymore; it’s how people make buying decisions. If your brand isn’t being recommended when someone asks ChatGPT or Google AI about your industry, you’re losing sales. Full stop.

According to our buying-intent experiments, brand mention rates vary dramatically based on how users phrase their questions. A brand that looks strong in one context can vanish in another. For example, Viking Pumps might pop up on a query like “best industrial pump manufacturers”, but a more niche prompt like “sustainable pumping solutions” might not trigger the same recommendation. You can visibility tracking tools to see for yourself. You can the Contxt scanning process to see for yourself. It’s not just about being visible, it’s about being visible everywhere that counts.

And with 39% of prompts showing divergent results between AI systems, businesses can’t afford to assume they’re covered just because they rank well on one platform. AI brand recommendations are fragmented, and this fragmentation is costing brands real money.

The GEO Principles: How to Take Back Control

If there’s one thing I’ve learned through all our testing, it’s this: visibility isn’t accidental. The brands that perform well in AI recommendations are the ones that optimise intelligently. That’s where the GEO principles come in: gather, experiment, optimise.

Step 1: Gather the Right Data

You can’t fix what you can’t measure. The first step is understanding where you stand. Use platforms like Contxt to run cross-provider tests and identify how often your brand is mentioned (or not) across systems like ChatGPT, Google AI Overview, and others. I've written about this separately in this piece about why most brands fail at llm visibility (and how contxt is.... Trust me, you need this baseline data.

In our network, we’ve completed 82 tracker runs and 58 experiments. That’s how we’ve spotted patterns like the 39% divergence rate and the massive gaps in brand visibility. It’s also how we know which types of prompts work best for specific industries. Without this kind of analysis, you’re guessing in the dark.

Step 2: Experiment Relentlessly

Once you’ve mapped your visibility, it’s time to test. And I don’t mean running a couple of prompts and calling it a day. I’m talking about systematically experimenting with different query types, languages, and even sentiment approaches to see how your brand performs.

Here’s an example. In one of our buying-intent tests, a brand that was totally absent for “best [industry] providers” became a top recommendation when we phrased the query as “top-rated [industry] suppliers in 2023”. Same intent, slightly different wording, completely different results.

You’ve also got to test against competitors. In one competitor-displacement experiment, we found that a particular brand was more likely to appear alongside its rival when the prompt included specific attributes like “cost-effective” or “award-winning”. Without these experiments, no one would’ve known how to tweak their messaging to own the space.

Step 3: Optimise for AI Systems (Not Just Search Engines)

Here’s where most brands get it wrong. They’re so focused on traditional SEO, keywords, backlinks, technical SEO, they forget that AI systems don’t work the same way as Google Search. These models rely on semantic understanding, contextual relevance, and, increasingly, verified data sources. If you’re not addressing those factors, you’re invisible.

How do you optimise for AI? Start by ensuring your content answers specific questions directly. These systems love clarity. Structure your website in a way that’s easy for models to navigate. And for the love of everything holy, prioritise schema markup and structured data. If you’re not feeding these systems the information they need, they’ll just fill the gaps with other brands.

Oh, and don’t forget partnerships. Some AI systems, like Google’s, prioritise data from established databases or partners. If you’re not in those ecosystems, you’re missing out on a chunk of potential visibility.

The Road Ahead for AI Visibility

Here’s my prediction: the gap between AI systems is only going to get wider unless businesses take action. With new models rolling out constantly and updates happening behind the scenes, it’s a moving target. But that doesn’t mean you’re powerless.

In fact, I’d argue that brands who invest in understanding and optimising for AI now are going to dominate their markets in a few years’ time. This isn’t going away. It’s the future of brand discovery, and the winners will be the ones who adapt early.

If you’re starting to feel overwhelmed by all this, you’re not alone. When we first started diving into AI visibility, it felt like a bit of a minefield. But the penny eventually dropped: innovation is about asking the why.

Why does one brand perform well on ChatGPT but fail on Google AI? Why do specific prompts trigger different responses? Why do some brands disappear entirely in low-visibility queries? It’s in answering these questions that the real progress happens.

Your Next Move

So, what should you do now? If I were in your shoes, I’d start by getting a clear picture of where your brand stands today. Run some tests. See how often you’re being recommended across ChatGPT, Google AI, and other systems. Then dig deeper. Experiment with different prompt types, track competitors, and look for gaps you can exploit.

And if you’re already feeling ahead of the curve, double down on optimisation. Make sure your content is structured in ways that AI can easily understand, and don’t stop experimenting. Visibility is a moving target in this space, and staying complacent is the fastest way to fall behind.

We’ve made it our mission at Contxt to help brands crack this code. If you want to see how your business stacks up, or just chat about AI visibility, drop me a line. Because at the end of the day, the brands who win will be the ones who refuse to let 39% divergence define them.

The future is bright, but you’ll only get there if the systems driving decision-making can actually see you. Make yourself visible. The rest will follow.