GEO (Generative Engine Optimization): researched methods

Generative Engines (GEs), like Google’s SGE or Perplexity.ai, represent a new paradigm in information retrieval. Unlike traditional search engines that primarily provide a list of relevant website links for a user query, GEs utilize large language models (LLMs) to synthesize information from multiple sources and generate a comprehensive, personalized, and often multi-modal response. These responses are grounded in the retrieved sources, ensuring attribution and allowing users to verify the information, effectively eliminating the need to navigate directly to websites. This shift significantly improves user utility by offering faster and more accurate information, but it also creates challenges for content creators.

Generative Engine Optimization (GEO) is a novel paradigm introduced to help content creators improve their content’s visibility in Generative Engine responses. It’s necessary because the rise of GEs disadvantages website and content creators. GEs directly provide answers, potentially reducing organic traffic to websites, which many small businesses and individuals rely on. Furthermore, the black-box and fast-moving nature of GEs makes it difficult for creators to understand how their content is ingested and displayed. GEO provides a flexible black-box optimization framework that enables content creators to tailor and calibrate their content’s presentation, text style, and overall content to increase visibility.

Traditional SEO measures visibility by a website’s average ranking on a search engine results page (SERP), which typically presents a linear list of websites. However, this metric is insufficient for Generative Engines because GEs provide rich, structured responses with embedded citations that can vary in length, position, and style. The concept of visibility in GEs is more nuanced and multi-faceted.

To address this, GEO proposes new impression metrics, including:

• Word Count (Impwc): The normalized word count of sentences that cite a particular source within the GE’s response. A higher word count suggests the source played a more significant role in the answer.
• Position-Adjusted Word Count (Imppwc): This metric builds on word count but assigns higher weight to sentences cited earlier in the response, using an exponentially decaying function based on citation position. This reflects the intuition that content appearing earlier is more likely to be read.
• Subjective Impression: This comprehensive metric incorporates subjective factors like the relevance of cited material to the query, the influence of the citation on the response, the uniqueness of the material, subjective position and count, the probability of a user clicking the citation, and the diversity of presented material. These are often measured using LLM-based evaluation methods like G-Eval.

The research highlights several highly effective GEO methods, which generally involve enriching content and improving its presentation:

• Quotation Addition: Incorporating credible quotes from reliable sources significantly boosts visibility (up to 28% on Subjective Impression and 40% on Position-Adjusted Word Count).
• Statistics Addition: Adding relevant quantitative statistics instead of purely qualitative discussions also leads to substantial improvements (up to 23.7% on Subjective Impression).
• Cite Sources: Explicitly including citations from reliable sources enhances credibility and visibility.
• Fluency Optimization: Improving the fluency of the website text can lead to a significant visibility boost (15-30%), suggesting GEs value good information presentation.
• Easy-to-Understand: Simplifying the language of the website also contributes to increased visibility.

These methods demonstrate that GEs value not only the content itself but also how it is presented and supported.

Traditional SEO methods, such as Keyword Stuffing (modifying content to include more keywords from the query), have been found to offer little to no improvement in generative engine responses. In some cases, keyword stuffing even performed worse than a baseline with no optimization. This underscores a key difference between traditional search engines (which might rely heavily on keyword matching) and generative engines (which employ language models for a more nuanced understanding of text and user queries). Website owners need to rethink their optimization strategies for this new paradigm.

Yes, the efficacy of different GEO methods varies significantly across domains and query types. For example:

• Authoritative writing styles are more effective for debate-style questions and historical domains.
• Cite Sources is particularly beneficial for factual questions, as citations provide verification.
• Statistics Addition is highly effective in domains like ‘Law & Government’ and for ‘Opinion’ type questions, where data-driven evidence adds value.
• Quotation Addition works well in ‘People & Society,’ ‘Explanation,’ and ‘History’ domains, where personal narratives or historical events are common.

This domain-specific variation implies that content creators should adopt targeted GEO strategies based on the specific subject matter and intent of their content to achieve the best visibility.

GEO methods tend to have a differential impact based on a website’s traditional Search Engine Results Pages (SERP) ranking. Notably, lower-ranked websites often benefit significantly more from GEO. For instance, the “Cite Sources” method resulted in a remarkable 115.1% increase in visibility for websites ranked fifth in SERP, while top-ranked websites sometimes saw a decrease.

This suggests that GEO has the potential to democratize the digital space. Traditional search engines often favor larger corporations with established domain presence and backlinks, which are hard for small creators to achieve. Generative Engines, by focusing on content quality and presentation as optimized by GEO, can level the playing field, allowing smaller entities to improve their visibility and compete more effectively.

Yes, combining multiple GEO strategies can lead to enhanced performance. For example, the combination of Fluency Optimization and Statistics Addition was found to outperform any single GEO strategy by more than 5.5%. Additionally, “Cite Sources” significantly boosted performance when used in conjunction with other methods, even though it was less effective when used alone. This indicates that content creators are likely to employ a mix of strategies in the real world, and these combinations can yield synergistic improvements in visibility. NotebookLM can be inaccurate; please double check its responses.