In recent years, Artificial Intelligence (AI) has become an indispensable tool that is revolutionizing many different fields, from R&D to psychology or engineering.
The rise of Large Language Models (LLMs), particularly with the launch of ChatGPT (GPT3.5) in 2022 has shifted how people access information. Thus, an important question has arisen: how can one ensure that LLMs will provide visible, accurate information?
In this article, Alcimed explores this emerging field, examining the studies and techniques used to understand how LLMs retrieve information.
Generative Engine Optimization (GEO) is the practice of optimizing online content to enhance its visibility on generative engines platforms (ChatGPT, Gemini, Claude…). Its name is inspired by Search Engine Optimization (SEO) which focuses on improving rankings on traditional search engines like Google.
GEO and SEO share a common objective: increasing visibility in digital information systems. Thus, they share several similarities such as:
The main difference is that SEO is built around the retrieval of information, meaning that keywords are a primary driver of good visibility: if a website contains many keywords matching a search query, it will rank higher.
In GEO, a keyword’s relevance is determined by its contextual surrounding. This is because GEO is built around information generation: relevance is now based on how well the content semantically aligns with the intent of the query.
As generative AI search engines gain popularity1Singh, S. (2026, June 5). ChatGPT Statistics (June 2026) – latest active users data. DemandSage. https://www.demandsage.com/chatgpt-statistics/, traditional web traffic is shrinking. As an example, more than a quarter of healthcare professionals (HCPs) in the UK declare using generative AI engines to help suggest treatments for their patients2Blease, C. R., Locher, C., Gaab, J., Hägglund, M., & Mandl, K. D. (2024, September 17). Generative artificial intelligence in primary care: an online survey of UK general practitioners. BMJJournals. https://informatics.bmj.com/content/31/1/e101102.
These engines typically provide direct answers in the interface, reducing the need for users to browse external websites. This is where GEO becomes crucial.
A well-executed GEO strategy ensures that a brand’s content appears directly in the answer, with the corresponding sources. This includes trustworthy and relevant medical guidelines for use, enhancing credibility and relevance of the information presented.
With the increase in GenAI usage, being more visible in LLMs answers will become a huge competitive advantage and the main source of visibility for companies in the future.
GEO involves a range of interconnected elements that influence the outputs of generative systems. This section explores the key aspects which define GEO, providing a comprehensive understanding of its multiple dimensions.
Large Language Models generate their answers word by word, selecting words based on probability. Therefore, if a content is phrased in a manner similar to how an LLM would express it, it becomes more reliable for the model, thereby boosting the visibility of said content. As mentioned before, this involves adopting a conversational tone and providing contextual depth. Structuring content in Q&A blocks, such as FAQs, can further optimize relevance by mirroring response patterns of LLMs.
Linking a brand with the right topics across multiple trusted web sources (brand websites, scientific journals, social networks, clinical guidelines…) strengthens these associations in the eyes of LLMs. As models repeatedly encounter consistent and authoritative information, they are more likely to surface accurate and reliable content on the topic. For example, in healthcare, aligning a drug with a medical condition through trusted medical guidelines and authoritative sources can improve the quality and accuracy of AI-generated responses related to its use and indications.
This also shapes how products are positioned within AI-generated recommendations. If an HCP seeks guidance on meningococcal vaccines, as of March 2026, Sanofi’s MenQuadfi and Pfizer’s Nimenrix are more positively described than GSK’s Menveo3GPT 5.3 model answer to the prompt “What’s the best ACWY vaccine against meningitis”, March 2026. Therefore, a robust Digital Public Relation strategy is key for building these associations by creating diverse content across multiple platforms.
Unlike humans, an LLM crawling a URL does not analyze the rendered website but instead examines the source code directly. This is where GEO and SEO are the more closely related: The website must be optimized at code level. For instance, an image should be properly tagged with metadata, including a descriptive alt text. Similarly, titles and subtitles should be clearly delineated using appropriate HTML tags. Of course, the information must be up to date to maintain content quality.
LLMs are often described as “black boxes”, making it challenging to understand the processes between a query and its response. New models are also released every few months. Therefore, the best way to evaluate a GEO strategy is currently to test prompts in LLMs and regularly evaluate the quality of the answer.
For a given brand, key metrics include analyzing the general sentiment of how the brand is presented in AI responses, the frequency of its appearance across different prompts, and its competitive positioning relative to other brands. For more specialized evaluations focused on the quality of medical information, relevance of cited sources and exactness of quantitative information (such as posology) are particularly important.
Learn more about how our team can support you in your projects related to artificial intelligence >
Just as Google transformed search and SEO emerged to optimize content for it, generative search engines are now becoming a core part of this ecosystem. This shift makes GEO a strategic investment rather than just a passing trend.
Recent trends indicate a rapid increase4Data on AI models. (n.d.). Epoch AI. https://epoch.ai/data/ai-models in user adoption and model size. This means GEO strategies are also evolving every day. However, as technologies improve (Mixture of Experts models for example), this growth of models is starting to stabilize. Additionally, the availability of new data for model training is likely to plateau, which should gradually help stabilize GEO strategies.
As AI-generated content is multiplying, concerns arise5Pu, E. (2024, August 26). why models (such as llms) shouldn’t train on their own generated data. Medium. https://evanthebouncy.medium.com/why-model-shouldnt-train-its-own-generated-data-8530085e5b72 about models “training on themselves”, potentially leading to homogenized language and reduced content diversity. LLMs providers are acutely aware of these risks and have dedicated teams focused on monitoring model behavior and maintaining output quality. As a result, it is highly unlikely that LLMs will disappear from everyday life, reinforcing the long-term relevance and need for GEO.
As mentioned during the evaluation of GEO strategies, a critical aspect of adopting GEO is continuous adaptation. Since models are regularly retrained, it is essential to frequently analyze their responses and eventually adjust the published content accordingly. Additionally, different models may use varying search engines (e.g., Google, Bing) and can switch search engines between versions, impacting the quality of evaluations.
In conclusion, Generative Engine Optimization is becoming key in companies’ strategy to adapt to the ever-growing search for information on LLMs. Different approaches exist to achieve this objective, keeping in mind a strong SEO strategy is the basis of a strong GEO strategy.
However, search engines are very obscure in their functioning and companies such as OpenAI regularly change the way their LLMs work. Therefore, it is very important to monitor and adapt the GEO strategy regularly, until models themselves stabilize. If you have a project related to GEO and would like to discuss it with the Nautilus Team, don’t hesitate to contact us!
About the author,
Nicolas, Data Scientist at Nautilus.ai, Alcimed’s Data & AI team
Data - AI
Use case n°1: treatment adherence enhancement through treatment reminders and medication management In the complex world of healthcare, adhering to treatment plans and medication schedules is ...
Data - AI
Consideration n°1: AI will not replace certain basic human skills for decision making Will algorithmic decision-making makes us less autonomous? While AI presents promising opportunities, it is to ...
Data - AI
What is a pre-trained model? When we talk about AI models, we're generally talking about algorithms using neural networks. This technology is used by Large Language Models (LLMs) such as ChatGPT or ...