Introduction to GEO: how to produce and structure content to capture AI’s attention

GEO (Generative Engine Optimization) – also referred to as GAIO, AI SEO, AIO or LLMO – is the emerging practice of making your content visible to AI services, particularly LLMs (Large Language Models). As more customer journeys incorporate AI, showing up in AI responses – in the right way – becomes an opportunity and even a necessity. But it isn’t always straightforward, often requiring not just text optimization but in-depth integration into all your communications.

The rise of AI is undeniable. And, as it becomes more integrated in people’s lives, it also becomes an important touchpoint in customer buying journeys and their perceptions of brands.

Did you know that:

• 83% of respondents in one survey prefer searching on AI to using a classical search engine such as Google

• 60% said AI gives better, clearer answers than traditional search

• 55% of consumers said they use AI to research products at least weekly, with 50% using it to help make the final buying decision

These statistics may not be the same for the specialized environments Bayer operates in. But the principle is clear: AI services are an increasingly important driver of customer traffic, informer of buying decisions, and source of trust. Showing up in LLM responses to relevant prompts is fast turning from a competitive edge to a requirement.

Is GEO a replacement for SEO?

Not quite: you can do GEO and SEO (Search Engine Optimization) at the same time. Increased traffic to LLMs is often associated with a decline in Google search volumes, but in reality, both are still important tools for your awareness and reputation. You do not have to choose between them.

But there is one key thing to remember: GEO and SEO are fundamentally different. GEO is not just ‘the new SEO’ or ‘another part of SEO’. It requires a specific approach that’s less about optimizing individual assets and more about crafting a consistent presence across your entire content ecosystem.

The big picture: LLMs and how they select information

LLMs are often referred to as a ‘black box’, meaning we cannot understand every precise detail of their inner workings. However, there are things we can learn from their outputs and how they are created.

A model’s response is heavily influenced by its training data (a massive dataset of all kinds of human-written text, regularly updated with new information). The scale of this data makes it hard for a human to analyze it and draw patterns, but a general principle that has been observed is AI models learn to associate certain concepts with each other. The more that this association is reinforced across different pieces of training data, the stronger the effect on responses.

This means long-standing associations between your brand and a topic will make you more likely to show up when someone asks an LLM a question relating to that topic.

These associations are formed not just through quantity, but also perceived quality of source material, with different sources given different weightings. Earned media (e.g. publications or influencers writing about you) typically has more weight than owned media (e.g. your own website), for example.

This means many classical SEO practices, such as optimizing an individual content asset by filling it with keywords, will not work! GEO is less about individual occurrences and more about how you show up consistently, in high-quality and useful contexts, over time and across touchpoints.

Even if you feel like your brand does not have this now, it is not too late to start. By producing written and multimedia content in the right way, and appearing on different platforms, you can build the ‘big picture’ association you need, piece by piece.

Day-to-day: How do I actually make my content noticeable to AI?

Let’s say you are writing a webpage or article for a Bayer product brand. What GEO steps should you actually take? Exact execution will vary by brand and use case, of course, but there are some overall principles we can follow.

One of the original GEO experiments, the Princeton-IIT Delhi study, tested nine approaches and found that some methods can improve content visibility to LLMs by up to 40%.

The most effective measures are those that add content to your page. The study identified three things you can add for a positive effect on how often and how prominently a source is cited by AI:

• Citations from credible sources

• Direct quotations from credible sources, or interviews with them

• Concrete statistics in place of vague qualitative claims 

Measures that do not add content, but better present the content you already have, could also lead to meaningful gains. These include:

• Improving text fluency (how error-free, well-written and coherent your text is)

• Reducing reading complexity by simplifying language and adding clear structures, making text easier for a model to understand and cite

• Other measures, e.g. adding domain-specific language, increasing the number of unique words, and writing in an authoritative tone, had lesser but still positive effects

Based on this, we can conclude that AI likes information to be well-presented, associated with a clear purpose, relatively simple and to-the-point, arranged in a logical flow, and backed up with evidence and statistics where possible. These aren’t bad principles for communicating information to humans, either! This should give you some optimism that it’s possible to produce content that has ‘good GEO’ without losing its appeal and creativity.

Finally, the study confirmed that keyword stuffing does not work, so this is not a GEO practice we recommend!

Three recommendations to get started quickly

1. Make sure your content is actually crawlable by AI. Web text such as HTML, PDFs with extractable text, and structured data such as JSON or schema are all AI-crawlable. Content produced by JavaScript, or only generated after a page loads, may not be. Anything hidden behind a login or paywall is unlikely to be accessible to AI. 

2. Review your text for clarity and structure. Write factual, flowing text with relatively short paragraphs. Try to make each section or paragraph answer a clear, separate question. Implement a clean hierarchy of headlines and sub-headlines. Captions and alt texts for images, and transcripts for videos, also help.

3. Add extra content and formatting. Add source citations, interviews, quotations and statistics. Look for opportunities to reformat some paragraphs into bullet points, listicle sections, comparisons or tables.

Guardrails for practical GEO

Although AI is powerful and opens up new communication possibilities, it’s good to retain some perspective. Remember these four guidelines:

1. AI can make mistakes. You may have seen it hallucinate or misattribute information, leading to the risk of your brand being associated with something false or misleading. This actually makes GEO more important: having a clear, consistent and accurate presence helps AI pick up correct information.

2. One article on its own won’t change everything. A strategic GEO approach is about doing it consistently, for the long term, across the whole content landscape: web, social, thought leadership, and even platforms you don’t own. 

3. AI is always evolving. Our understanding of best practice is regularly updated, as are the tools we use. So check back here for further updates!

4. Your audience is still a human. AI isn’t the audience – it’s a tool to reach them. Therefore, although GEO is important, communicating to meet the needs of your audience, and deliver them a positive experience, is still top priority.

Got GEO success stories to share? We would love to hear them – get in touch at identitynet@bayer.com.