Last week we launched our new services. To give you a bit of an insight, we'll delve into each service with a detailed blog. Up for today: Generative Engine Optimization (GEO). We'll explore what GEO is, why GEO is important, how you can implement GEO and what Initium does for GEO. So, buckle up and let's dive right into it.
GEO stands for Generative Engine Optimization and revolves around one simple question: How can we make a brand or person pop up in AI answers? This is a relatively new field, and builds upon the practice of Search Engine Optimization (SEO). While GEO and SEO build upon the same pillars, GEO requires a slightly different setup. This has to do with how AI engines, such as Gemini, ChatGPT, Claude and the like, handle retrieving and displaying information. Roughly speaking, there are 2 types of information that live within an AI model: internal and external.
Internal information is everything that is cooked into a model. You can't point to this information, but it is a crucial part of why LLMs can give answers.
A quick recap on LLMs: An LLM "guesses" the next word based upon a given input. If you input: "The apple does not fall far from the", the model will run a calculation to determine the next word, and say "tree". It manages to do this via millions, sometimes billions, of so-called weights. As the input flows through the LLM, each weight adjusts the output, and depending on how the model is trained, the output should be a fitting answer on the input.
Any given model starts with randomized weights, and LLMs in this case literally outputs gibberish. This is where training comes in. For LLMs, this means terabytes of standardized prompts with answers. During training, a prompt is fed into the model and the output is evaluated on whether or not it found the right answer. Even simpler, whether or not the "guessed" word is the correct one. Depending on the correctness, the weights are adjusted. To keep this article low-tech, we will not dive into gradient descent but in theory this process of feeding and adjusting eventually leads to a model which can quite accurately construct answers to questions, as seen with the conventional models.
Now, the data that is used in training a model is extremely important in determining the capabilities. A LLM trained on purely coding data, will be very good at coding, but no on general questions. Broader internet datasets usually sound much more "human" and can be used for simple Q&A. This distinction in training data is also why ChatGPT might answer differently compared to Gemini. Google natively integrates its massive Search index into its ecosystem, while OpenAI relies on dedicated web crawlers (like GPTBot), search API partnerships, and strategic data-licensing agreements to continuously build its models. Hence the difference in output of Gemini compared to ChatGPT. It is also important to note here, that training happens once. So all data within a model is frozen from the moment it goes live to the public. A good example of this is that Gemini 3.1, which launched in February 2026, does not know what Macbook Neo, launched in March, is, without searching the internet.
Now back to GEO, Why is training data important when looking at GEO? Simple, if your brand was included in the training data, the LLM now has "internal" knowledge of your brand. And this is a major advantage: if an AI model already understands your brand internally, search routers are far more likely to recognize your brand name as a distinct entity and surface you accurately during RAG retrieval.
We've already touched upon it in last paragraph, but external information refers to information that does not exist within an AI model. Contrary to internal information, you can point to this information. For LLMs, this is essentially all information it can retrieve using tools. Tools can be internet search, but also looking into local documents, depending on what type of LLM you are using. The most used version of external information within general purpose models is internet searching. You might already have seen this with the Google AI overview you get when using Google Search, or why seemingly every AI knows the news of today despite having a hard cutoff in training data.
The fancy term for obtaining external information is called RAG. RAG stands for Retrieval-Augmented Generation, and combines traditional information retrieval with specific AI tailored systems. We'll dive into RAG next.
We can write an entire article about RAG, but here are the quick must-knows. The main problem RAG solves is the combination of getting every relevant piece of information that lives outside of the LLM. Usually, the first step in RAG starts with obtaining information, in most cases via a search algorithm. These algorithms leverage vector databases to efficiently retrieve information. Vector databases store information as embeddings in a high-dimensional space. To make this less complicated, let us give a stylized example: instead of storing "initium_google_ contractv1.25_ extended_ adjustv2.pdf", you embed "Contract between Initium and Google" in the database. The main advantage here is that engines do not have to search for an exact keyword match, but rather can search for every number of combinations as long as the concept is somewhat clear. This system is extremely compatible with AIs, which in a similar way encode internal information in a high-dimensional space.
Before parsing the found information directly to the LLMs, RAG applies reranking to the data. This is a fancy way to screen and clean the obtained information for relevancy and factual density. Long texts that do not include relevant information are penalized, whilst consice information is favoured. Marketing fluff for example is usually penalized.
In practice, this means that when an AI is asked about the president of the United States, it will use RAG to query for the concept the president, rather than matching the first search result, yielding better answers. Therefore, using RAG significantly increases the relevancy, accuracy and quality of LLMs.
GEO is the art of making sure that the internal and external information shows your brand, website or personality. For internal information, this is a long game as being included in training datasets is fully dependent when companies are training the next versions of their models, more on this in "What is a crawler?". External information however, is something that is quickly becoming a focal point for marketing agencies. RAG pulls most information from search engines, and this is also why GEO does not replace SEO. In fact, getting picked up by RAG necessitates proper SEO. But, the fact that LLMs work with concepts rather than keywords, adds an extra dimension. So essentially, your brand needs to become a "concept" on the internet.
To establish your brand as a recognized concept, AI models rely heavily on something that is called Entity Linking. They cross-reference your site checking, amongst other things, sources like industry review sites, news outlets, and discussions on platforms like Reddit. If these outside sources confirm what you do, AIs "trust" your brand much more.
To make this concrete, let us give an example.
Let's say you are a bakery and you want to make sure you shown within ChatGPT or Gemini, you need to make sure that the concept of "getting bread" is directly linked to your bakery, preferably cross-referenced with outside sources. Because when people open up the Gemini app and ask: "where can I find good bread?" You want the RAG to directly pick up your bakery and feed that back into the answer, whilst also showing a good review on a recent reddit post.
GEO aims to do this for all conventional LLMs, as well as other evolving search engines.
You might also be wondering, if internal information is set in stone, why would I optimize for that? Well, every minute of every day, the large AI companieGEOs are running so-called "Crawlers" across every corner of the internet. These AI Bot crawlers have 1 specific job: obtain data that can be used in training. They function different to the traditional search engine crawler. A traditional crawler indexes an entire page for keyword matching, AIs need conceptual information, and as such the AI bot crawlers focus much more on "high signal" information. High signal information can come in many forms, but we'll outline a few:
Moving back to our bakery example, the bakery can list a question: "Why is this the best bread in the neighbourhood?" and immediately answers in the correct format: "Our sourdough is baked fresh daily using a 48-hour natural fermentation process and 100% locally milled organic flour," followed by a concise bulleted breakdown of key details (e.g., zero artificial additives, award-winning starter culture, traditional stone-deck ovens).
Because this format delivers immediate, dense facts rather than conversational fluff like "We put our heart and soul into every loaf," the AI crawler can instantly ingest, chunk, and attribute the data. This makes the bakery significantly more likely to be cited as the definitive answer when a user asks an AI, "Where can I find the best organic sourdough nearby?"
External information also relies on this concept, albeit to a lesser extent. Additionally, modern AI crawlers are multi-modal. They parse text, images, video, you name it, to build a better conceptual map of your website/brand.
With RAG relying on conceptual searching, poorly structured websites without concise questions and answers in any format, are penalized. Good structured websites however, are reranked favourably and can be used by the LLM as citation. These citations are direct links to your website and significantly improve findability within AI engines.
Simply put, LLMs, and AI-based searching will become the future. Already, the worlds most used search engine, Google, is doubling down on the inclusion of AI in every aspect of searching the web. Google Search gives an AI overview on top of the page and provides a dedicated "AI Mode". On top of this, we'd urge you to ask your (maybe less technical) friends how they obtain information. 40% of people use AI as their primary way to search things online, and this number keeps growing. Hence, not showing up in AI results essentially locks you out of a significant portion of the market. And to reiterate, no, traditional search will not be replaced by AI, and SEO remains a very valid discipline. But if you are a business in 2026, you must think how GEO impacts your visibility online.
The first major improvement most websites, including portfolios, can make is adding significant structure to all important pages. Start by adding questions and answers, using correct headings, implement structured tables with concise keywords relevant to your product or service and allow RAG to use your website in simple citations. To further edge out an advantage, it is vital that your brand is consistent across all outlets. RAG will punish inconsistencies, even across multiple social media, meaning that if you claim to be a bakery on your website yet on Linkedin you post about cars, you risk tanking your GEO performance. This is also exactly why GEO is measured across time, because your content strategy impacts the process. Tracking your performance periodically, via standardized prompt sets for example, allows you to see how efficient your brand is within the answer engines. Lastly, do not slack on SEO as much of GEO builds upon core SEO principles.
We break GEO down into 2 components: establishing presence and building structure.
Establish presence is done by approaching your website or portfolio from the answer engine perspective. That means looking into:
Initium Strategies works with you to audit your brand and formulate your GEO strategy:
After we've identified what is needed to establish GEO presence and the existing gaps, we can move on to building the structure.
Building structure starts with a proper back-end, building out citations and afterwards monitoring progress in the long-run. The following steps are important here:
Initium Strategies works with you to implement your GEO strategy:
At Initium Strategies, we implement GEO for a multitude of companies and government bodies. Whether you are a sole-proprietor starting out with your first website, an influencer in need of better AI presence, a municipality wanting to increase landmark visibility or an SME looking to improve findability, Initium Strategies got you covered with a end-to-end full service GEO offering.
You can contact us via your preferred platform, using our contact page.
Alternatively, you can contact us directly using LinkedIn:
So if you want to take your GEO strategy to the next level, or simply want to connect, reach out to us!