A quick refresher on Generative Engine Optimization

For those who are new to the space, Generative Engine Optimization (GEO) is a new market that seeks to help brands analyze, understand, and improve how they appear in AI search platforms such as ChatGPT, Google Gemini, and Perplexity. GEO took shape alongside the rapid adoption of LLMs, as companies realized that many of their customers were shifting from browsing blue links in Google to completing their entire purchase journey within LLMs. Many brands discovered the hard way that they were missing from key AI search queries in their category, resulting in declining website traffic while competitors benefited from tapping into this new distribution channel.

As brands raced to figure out this shift in customer behavior, a wave of vendors entered the market promising to help companies increase their visibility in AI search. As the GEO market has developed, one thing has become abundantly clear – effective GEO strategy is not one-size-fits-all.

There is no universal formula whereby all companies can target a specific type of content on their website or a particular type of earned media, magically increasing visibility. Instead, strong GEO strategy requires a thoughtful, case-by-case approach to establish what areas a brand is lacking in, and what can be done to improve these areas and in turn grow visibility.

What's less understood is that AI visibility doesn't behave uniformly across industries. This misunderstanding is why many GEO efforts fall short.

Why GEO isn't one-size-fits-all

Unlike search engines, LLMs are probabilistic systems whose core function is to predict the most likely next token given a context. Everything they output is the result of learned statistical relationships between words, concepts, entities, and patterns observed during training and reinforcement.

At a fundamental level, an LLM does not “decide” which brand is best. It estimates the likelihood that one token follows another. When a model produces an answer that mentions a company, product, or concept, it is doing so because, mathematically, that mention has a high conditional probability in that context.

Those probabilities are not universal. They are shaped by what the model has learned about how the world works across different domains. Each industry encodes different truths into the data the model is exposed to. For example, in healthcare, authority may be statistically associated with clinical studies, institutional language, and regulatory entities. In consumer technology, authority may be associated with developer documentation, community adoption, or product comparisons. In finance, it may correlate with institutional credibility, consistency of terminology, or regulatory alignment.

Because of this, authority is not a single global signal. It is an emergent property that varies by domain, use case, and question type. The same brand can carry very different statistical weight depending on whether the model is answering a medical question, a B2B software comparison, or a consumer purchasing query. These differences are learned implicitly from data distributions, not explicitly encoded as rules. What this means is that they are also subject to change over time.

This is where AEO and GEO fundamentally diverge from traditional SEO. Google Search is a system explicitly designed to return a ranked set of results. It has a defined algorithm whose purpose is to order links and brands relative to one another. Even though the algorithm may also be complex and opaque, the objective is clear: rank documents.

LLMs do not have that objective. They do not rank brands. They do not optimize for links. They do not even operate on the concept of “results.” Instead, they generate text that is statistically coherent with the prompt and the model's internal representation of the domain. Brand mentions emerge as a byproduct of learned likelihoods, not as a scored ranking outcome.

As a result, the inputs that influence GEO positioning are different, and they vary by industry. Signals that meaningfully shift token likelihood in one domain may be irrelevant in another. Non-linked mentions, community discourse, instructional content, domain-specific language patterns, and even how problems are framed all influence how likely a brand is to appear in an answer. This is why platforms like YouTube, Reddit, forums, documentation hubs, and earned editorial content can disproportionately influence AI visibility without producing traditional clicks.

This also explains why GEO cannot be treated as a one-size-fits-all problem. There is no universal playbook because there is no universal authority function. Each industry has its own statistical structure, its own language patterns, and its own notion of credibility as learned by the model.

When Petra Labs was founded, we had a hypothesis that the websites ChatGPT cited varied meaningfully across industries. Our analysis consistently showed that they do. We learned quickly that the industry a company operates in –whether it be healthcare, technology, consumer goods, or otherwise– dictates the sources of information ChatGPT refers to in order to answer your question.

What we can do is analyze LLM responses and the websites they cite in their answers. By identifying patterns in the content LLMs cite, we can pinpoint a company's visibility gaps and design a targeted content strategy. That strategy spans owned content on your website, earned media placements, and social channels such as Reddit and YouTube, and is unique and custom to your industry.

A concrete example: how industry can drastically impact ChatGPT citations

This makes sense in theory, but to prove it, we put it to the test. Using our software, we analyzed ChatGPT's citations for queries related to recruiting software and compared it to payroll software. For each category, we analyzed a large set of ChatGPT responses across representative queries and aggregated the domains cited across those answers. Because both categories fall under HR/HCM and often have the same buyer, you'd expect ChatGPT to rely on similar sources when answering questions about each.

Here are a few stats that prove otherwise:

• Earned media is 60% more influential in payroll than in recruiting
• There is only 33% overlap in the top 15 most cited domains across the two categories
• Payroll citations are 45% more concentrated among the top sources, while recruiting relies on a broader long tail

Below is a summary of the top 15 most cited domains. Red shading indicates unique domains that do not overlap with the other category.

Notably, the overlapping domains (e.g. Forbes, Reddit) tend to be horizontal platforms, while category-specific authorities diverge almost entirely.

The takeaway: Recruiting and payroll sit next to each other in HR tech, but LLMs treat them as fundamentally different information problems. This demonstrates that your GEO strategy must be custom-fit to the intricacies of your industry.

How to run a quick GEO test on your brand

Reading this, you might be thinking: Okay, this all makes sense… but what should I do about it?

Well, there are a few things. First, you should decide what AI prompts you care most about and want to appear in. A good place to start is your Google keywords. For example, if you're a recruiting software startup, you likely track keywords such as:

• “Applicant tracking system”
• “Hiring software”
• “Talent Acquisition software”
• “ATS tools”

Take these keywords and turn them into natural language prompts, as shown below:

• What are the best applicant tracking systems for startups?
• What hiring software should I choose for my small business?
• What are the best talent acquisition software platforms?
• Which ATS tool should I use to help find new candidates?

Ask these questions in ChatGPT with small variations in phrasing, tracking both brand presence and the citation patterns the model surfaces. This certainly won't encompass all of the queries your customers are asking, but it's a good litmus test for whether or not you're appearing in AI search results at all.

Another quick experiment is sentiment. Sure, ChatGPT might be mentioning your brand, but in what light? Is it in line with your brand messaging? Is the sentiment positive or negative overall? To test for this, put yourself in the shoes of a customer debating whether or not to test out your recruiting software:

• Is this ATS tool cost effective?
• How does ATS platform A compare to ATS platform B?
• Is ATS platform A a good vendor for me to use if my company is growing quickly?

Come up with a comprehensive list of questions that mirror your customers' key purchasing criteria and test them in ChatGPT. Questions like these will give you a sense for ChatGPT's perception of your brand. You'll likely find that there are at least some things that are inconsistent with how you'd like your brand to appear. If that's the case, then GEO should play a central role in your marketing strategy.

Developing a comprehensive, long-term GEO strategy

The recruiting versus payroll example is just one illustration of a broader pattern: AI visibility is determined by industry-specific trust structures that can't be addressed with a generic, one-size-fits-all, playbook.

If you've made it this far, the next step is understanding how LLMs perceive your industry and brand and where your visibility gaps exist. Petra Labs takes an “N of 1” approach, working with each customer to understand the intricacies of your market and your brand's presence in AI search, then building a bespoke content plan to help execute on visibility improvement.

We're an end-to-end GEO provider, covering everything from visibility analytics to content strategy to actually writing the content that will help boost your brand's presence in LLMs.

If you want to hear more, get in contact with us here!