Generative Engine Optimization: A New Frontier for Search Engine Visibility
5 key takeaways about Generative Engine Optimization:
Generative engines are a new type of search engine powered by AI that synthesize answers by gathering and summarizing information from multiple sources.
Unlike traditional SEO, there are currently no established techniques for websites to optimize visibility in generative engine results.
Generative Engine Optimization (GEO) is a new framework proposed to help website owners increase their visibility in generative engine responses through content optimization.
GEO techniques like adding relevant statistics, quotes, and simplifying language have been shown to boost visibility in generated answers by over 40% through experiments.
GEO introduces novel visibility metrics tailored for judging source visibility in the rich, multi-source responses of generative engines.
The advent of large language models and conversational AI has given rise to a new class of search engines dubbed "generative engines." Unlike traditional search engines that simply return a list of websites, generative engines can synthesize answers by gathering and summarizing information from multiple sources. This emerging technology shows great promise in providing more accurate and personalized responses to users. However, it also poses new challenges, especially for website owners seeking to optimize their visibility. In this blog post, I'll provide an overview of generative engines, explain why traditional SEO fails, and summarize key strategies proposed in a new paper on "Generative Engine Optimization" (GEO).
What are Generative Engines?
Generative engines refer to search engines powered by large language models like GPT-3 that can generate free-form responses to queries, rather than just links. They typically work by first retrieving relevant documents using a standard search engine, then feeding these documents into a language model to summarize and synthesize an answer grounded in the sources.
For example, if you asked a generative engine "What are the best Mexican restaurants in Austin?", it may first pull up restaurant review sites and articles discussing Austin's food scene. The language model would then generate a response summarizing the most highly rated restaurants based on the retrieved documents, directly citing sources in the final answer.
Compared to traditional "ten blue links", this allows for more natural conversations, personalized and nuanced responses, and answers synthesized from diverse sources. Early examples of commercial generative engines include Anthropic's Claude, You.com, and Perplexity's AI.
The Challenge for Websites and Creators
While generative engines enhance the user experience, they could significantly disrupt content creators and websites that rely on search engine traffic. When users get direct answers from the engine, they no longer need to navigate to original sites. This removes a major discovery channel, threatening creator livelihoods built on site traffic.
Unlike traditional search engines where tactics like keyword optimization and link building influence rankings, generative engines are black boxes that synthesize responses behind the scenes. There's currently no established way for websites to optimize their visibility in these new systems. But given rapid user adoption of generative engines, solutions are urgently needed to empower website owners.
Introducing Generative Engine Optimization (GEO)
To address this problem, researchers have proposed Generative Engine Optimization (GEO) - a new paradigm to increase the visibility of websites in generative engine responses. The key idea is to optimize website content itself to boost the likelihood of being cited and summarized in the final generated answer.
Since generative engines are proprietary black boxes, GEO focuses on engine-agnostic strategies based on fundamental properties of language models. For example, including relevant statistics and quotes from authoritative sources can increase perceived credibility. Simplifying complex language can make a website more readable and engaging for the model.
The paper introduces and evaluates various GEO techniques like "Statistics Addition", "Quotation Addition", and "Easy-to-Understand". Beyond predefined strategies, GEO allows creators to customize visibility metrics to reflect business goals. The overall framework arms website owners with concrete levers to improve visibility in this new search landscape.
Novel Visibility Metrics Tailored for Generative Engines
A core contribution of GEO is proposing new visibility metrics suited for generative engines' rich, multi-source responses. Traditional SEO uses rankings on search results pages. But with generative engines, multiple sources are interwoven in a single block of text. Notions of ranking don't apply directly.
To address this, the paper introduces impression metrics that account for various factors affecting source visibility:
Word Count: Portion of response based on a source
Position-Adjusted Word Count: Reduces weight for lower ranked sources
Subjective Impression: Aggregates relevance, uniqueness, influence, and other subtle factors using a language model
These specially designed metrics provide a quantitative lens for website owners to track and improve performance in generative engines' responses. The paper also enables creators to define custom visibility metrics aligned with business goals.
Evaluating GEO Techniques on a New Benchmark
To facilitate effective GEO research, the paper authors construct a benchmark of 10K queries spanning diverse domains and difficulties, coupled with relevant web documents. They evaluate various methods on this benchmark using the proposed visibility metrics.
Their experiments reveal that techniques like adding relevant statistics, quotations, and citations can boost visibility by over 40%. Simply improving readability and fluency also leads to a 20% increase. These large gains highlight the viability of optimizing content specifically for generative engines.
Interestingly, traditional keyword stuffing proves ineffective, confirming that historically popular SEO strategies do not transfer directly. The paper also analyzes query categories where different GEO techniques are most impactful, allowing for domain-targeted optimization.
In summary, this rigorous benchmarking and set of proposed strategies represent an important first step in adapting to generative engines' new paradigm. As these AI systems continue improving and get deployed at scale, the GEO framework provides website owners concrete levers to control visibility.
Implications for Generative Search Landscape
As discussed, GEO has profound implications for both generative engine developers and content creators. For developers, it underscores the need for transparency and mitigating unfair bias, so that visibility is driven by relevance rather than opaque optimization tricks. Constructing diverse benchmarks like the paper's will also accelerate progress.
For website owners, GEO finally provides some control in navigating this seismic shift in search. Smaller creators also benefit, since GEO disproportionately boosts lower-ranked sites. This helps democratize access and counters consolidation of power by tech giants.
However, creators must thoughtfully assess optimization strategies and avoid "contriving" solely for engines. High-quality content that organically cites sources will fare best long-term as the technology matures.
Looking Ahead
Generative search marks an exciting new evolution of information discovery. As these AI systems grow more powerful and ubiquitous, GEO provides an initial compass for navigating the landscape responsibly and equitably. This field remains in its infancy, but the solutions and mindsets established today will set the trajectory far into the future. Researchers must continue pushing boundaries, while collaborating closely with creators and users. With wisdom and foresight, this technology can open up new vistas that empower humanity.