The Invisible Hand of Prompts: How AI Systems Shape GEO and What Brands Can Do About It

The Invisible Hand of Prompts: How AI Systems Shape GEO and What Brands Can Do About It

AI is already redefining how brands are discovered, considered, and chosen. But let’s be honest, most businesses don’t fully understand the mechanics behind it. They assume that AI just "knows" what’s out there. That it understands their brand and offers fair representation. Spoiler alert: it doesn’t.

The truth? AI systems are only as good as the prompts steering them. AI prompt engineering, the art and science of crafting effective prompts, is fast becoming a critical skill for anyone who cares about generative engine optimisation (GEO). If your prompts are vague, poorly structured, or just not aligned with how the AI interprets inputs, you’re invisible. And in the age of AI-driven recommendations, invisibility is commercial suicide.

So, let’s unpack what’s really going on. How do prompts shape GEO outcomes? Where do biases creep in? And more importantly, how can brands turn prompt engineering into their secret weapon?

How Prompts Hold the Power

Here’s the thing: AI doesn’t have "common sense". It doesn’t instinctively know which brands, products, or services should rise to the top. It works entirely on instructions, and those instructions often come in the form of user-generated prompts. This means every recommendation, every insight, every decision AI makes is framed by how it interprets the language of the prompt.

Our data at Contxt backs this up. Across our dataset of 2881 prompts spanning 31 businesses, we saw that the way a prompt is written dramatically changes visibility. It’s not just a small margin either, 39% of the time, ChatGPT and Google AI give completely different brand recommendations for the same query. Same inputs, same brands, entirely different outputs. Related reading: our article on the brand invisibility problem: why your business is invi.... That’s a massive divergence.

Why? Because the models process language differently and prioritise different factors when generating responses. It’s not rocket science, but it has big implications for your visibility strategy. Related reading: this piece about why 39% of ai systems disagree on brand recommendations (.... Related reading: our article on the hidden cost of ethical blind spots in llm visibility .... If your prompts aren’t tailored to the quirks of each AI system, you’re leaving your brand’s fate in the hands of a black box.

The Role of Brand Mention Rates

Let’s get specific. Across the 1016 unique prompt variations we tested, the average brand mention rate was 62.9%. Not bad on the surface, right? But dig deeper and you’ll see some shocking inconsistencies:

• ChatGPT delivered a 75.4% mention rate.
• Google AI Overview? Just 45.5%.
• Claude performed slightly better at 78.5%, while Perplexity hovered at 50.0%.

This shows a clear pattern: your brand might thrive on one platform but be invisible on another. ChatGPT’s higher mention rate often makes it the darling of GEO strategies, but if your audience is using Google AI instead, that 30% drop-off could be costing you ideal leads. Now, imagine if you’re not even hitting the 62.9% average. I've written about this separately in Why 38% of AI Systems Disagree on Brand Recommendations (.... That’s not just bad news, it’s catastrophic.

One example? TOCA Social, a brand in entertainment/sports tech, scored a 10.0% visibility rate across the prompts we tested. That’s a blip on the radar compared to Viking Pumps, which achieved an impressive 100%. It’s worlds apart, and it all comes back to how effectively the brand aligns with AI's interpretation of prompts.

Bias in the Machine

AI isn’t neutral. That’s a myth people like to believe, but the reality is messier. Biases creep in at every stage, from the training data systems are built on to the weighting of certain keywords in prompts. And as much as we'd like to think we can control it, there are always factors out of our hands.

For example, AI systems tend to favour well-known brands and established players. This skews results against smaller companies or newer entrants who don’t yet have the "credibility" these systems look for. You can AI visibility tracking features to see for yourself. You can how Contxt works to see for yourself. It's a self-reinforcing cycle, big brands get more visibility because they’re already big, while smaller ones struggle to break through, no matter the quality of their offering.

But, and this is a big "but", the way prompts are crafted can mitigate some of this bias. In our experiments, we found that certain phrasing choices could nudge the AI to broaden its scope. For one manufacturing client, we tested a "competitor-displacement" prompt that explicitly asked the AI to recommend alternatives to the market leader. It worked. The client's mention rate jumped by 62% in that specific test, just by reframing the question.

Context Matters More Than You Think

An overlooked factor in all this is the context of the query itself. Our buying-intent tests revealed staggering variations in how AI responded. The same brand could be recommended in one context (e.g., "best B2B software for SMEs") but entirely absent in another (e.g., "top-rated enterprise solutions").

That’s where strategic prompt engineering can make or break you. Too many brands think one-size-fits-all when it comes to crafting their AI visibility strategy. You might also want to check out Why AI Visibility Demands a Sector-Specific Approach: Les.... But the truth is, if you’re not tailoring your prompts to reflect how customers actually search, you’re missing countless opportunities to influence their choices.

Stepping Into Strategic Prompt Engineering

Alright, here’s the big question: what can you actually do about all this? How do you make sure your brand isn’t just playing catch-up but actually driving the conversation within AI?

In my experience, it comes down to three key strategies:

1. Test, Measure, Refine

Let’s face it, this one’s not optional. If you’re not running structured experiments to see how different prompts affect your brand’s visibility, you’re flying blind. Across our platform, we’ve conducted 58 experiments for clients, covering everything from multi-language queries to citation tracking.

The result? A treasure trove of insight. Buying-intent tests revealed gaps in visibility. Prompt-sensitivity tests showed which phrasing delivered the best results. Even a single test can uncover actionable opportunities to refine your strategy. Start by running variations of prompts your target audience is likely to use. Then analyse which ones consistently put your brand front and centre.

2. Tailor for AI System Biases

As we’ve seen, not all AI systems are created equal. ChatGPT is leading the pack in brand mention rates, but Google’s AI Overview lags behind. So, if your audience uses multiple systems, don’t just optimise for one. Tailor your strategy to the quirks of each platform.

For example, Google’s AI might prioritise web authority and search ranking factors, so strengthening your SEO could indirectly boost visibility there. On the flip side, ChatGPT might favour conversational queries and practical recommendations, which requires a different kind of optimisation.

3. Craft Intent-Driven Prompts

Generic queries won’t cut it. You need prompts that reflect real-world buying intent. One way to do this? Mimic the language of your audience. If customers often ask, "What’s the best X for Y?", design prompts that mirror that structure. If they’re more likely to compare features, build prompts around head-to-head comparisons.

The more specific and aligned your prompts are with actual search behaviour, the higher your chances of being included in the AI’s responses. And remember, context isn’t just about phrasing, it’s about timing, location, and query type. Take these factors into account when engineering your prompts.

Looking Beyond the Prompt

This next point might sound counterintuitive, but here it is: prompt engineering isn’t everything. Yes, it’s critical, but it’s only one piece of the AI visibility puzzle. If you’re not also strengthening your brand’s underlying digital presence, no amount of clever prompts will save you.

Think of it this way. Prompts are how you open the door, but your brand’s broader credibility determines whether you’re invited inside. That means investing in things like:

• SEO and content marketing to build domain authority.
• Robust product descriptions and schema markup to help AI systems understand your offerings.
• Consistency across platforms to reinforce your brand’s identity.

At the end of the day, GEO is about creating a comprehensive strategy that combines prompt optimisation with broader brand-building tactics. Ignore one at your peril.

My Take: Why This Matters More Than Ever

I’ll be honest, when I first started digging into prompt engineering, I wasn’t convinced it was a big deal. But after seeing the data and working with clients across multiple industries, I’ve changed my mind. This stuff isn’t just important, it’s foundational.

The AI landscape is evolving fast. Models update, new systems emerge, and user behaviour shifts. If you’re not actively monitoring how your brand is represented (or not represented) in these spaces, you’re ceding control to algorithms you don’t fully understand. That’s a risk no business can afford to take.

So where do you go from here? Start small. Run a few tests. Play around with prompt variations. Measure your brand’s visibility across different AI systems and track how it changes over time. From there, refine, iterate, and scale. It’s not a quick fix, but the long-term upside is too big to ignore.

And if you don’t have the time or expertise to tackle this on your own? That’s where we come in. At Contxt, we specialise in helping brands navigate the murky waters of AI visibility and prompt engineering. Ready to get started? Let’s talk.