Researchers found ways to increase visibility in AI search rankings by testing various AI search optimization techniques.
Researchers tested nine optimization methods for AI search engines at Princeton University, Georgia Tech, IIT Delhi, and the Allen Institute for AI.
Researchers were able to outrank large corporate sites that typically dominate search results by successfully increasing the visibility of low-ranked sites.
Three techniques for improving visibility on AI search engines outperformed other approaches.
Specific techniques proved most effective for particular knowledge domains (such as Law, people, society, etc.), demonstrating how various subjects call for multiple approaches to increase visibility.
Researchers conducted extensive experiments to optimize websites for AI-driven search algorithms, uncovering precise strategies to amplify visibility. Their breakthrough approach, dubbed Generative Engine Optimization (GEO), yielded remarkable results, elevating the visibility of smaller, less prominent websites by a staggering 115%. This surge allowed these underdog sites to surpass larger corporate entities that conventionally monopolized the top echelons of search results.
Hailing from Princeton University, Georgia Tech, Allen Institute for AI, and IIT Delhi, the researchers meticulously examined nine optimization techniques spanning diverse knowledge domains such as Law, history, and science. Their rigorous analysis delineated the efficacy of each method, delineating those that significantly bolstered rankings, those that elicited negligible impact, and those detrimentally affecting visibility.
Specific optimization strategies exhibited domain-specific efficacy, while three techniques emerged universally potent across various website categories.
Emphasizing the democratizing potential of GEO, the researchers underscored its role in reshaping the digital landscape:
“This discovery underscores GEO’s potential as a transformative tool, democratizing the online sphere. Crucially, many of these enhanced rankings benefit smaller content creators and independent businesses, historically disadvantaged in competing against corporate giants dominating search engine standings.”
Tested on Perplexity.AI
The researchers conducted experiments on the Perplexity.ai search engine and an AI search engine modeled after Bing Chat, discovering congruent results between the two platforms.
In section 6 of their research paper, they noted:
“Our observations align with findings from our generative engine. Quotation Addition exhibited the most notable performance in Position-Adjusted Word Count, boasting a relative improvement of 22% over the baseline. Additionally, techniques like Cite Sources and Statistics Addition, which demonstrated strong performance in our generative engine, showcased substantial improvements of up to 9% and 37% across these metrics.”
Comparative Analysis: Evaluating Methods on Bing Chat Modeled AI Search and Perplexity.AI
The researchers conducted comprehensive tests, applying their techniques to a generative search engine fashioned after the workflow of Bing Chat. Additionally, their methods were examined on Perplexity.AI, another AI-driven search engine.
In their documentation, they elaborate:
“We introduce a generative engine, housing various backend generative models and a source retrieval-oriented search engine.
This Generative Engine (GE) operates by taking a user query and generating a personalized natural language response, leveraging PU, which encapsulates user-specific details like preferences and browsing history.
The fundamental components of Generative Engines encompass:
a.) A collection of purpose-specific generative models G = {G1, G2…Gn}, each tailored for tasks such as query reformulation or summarization, and
b.) An associated search engine SE that, when given a query q, furnishes a set of sources S = {s1, s2…sm}.
We present a workflow model… that, at the time of documentation, closely resembles the architecture of BingChat. This process dissects the input query into more straightforward queries, optimizing their compatibility with the search engine’s operations.”
Crafting a Diverse Search Query Benchmark
The researchers meticulously constructed a benchmark comprising 10,000 search queries culled from various sources, spanning multiple knowledge domains and exhibiting varying levels of complexity. Notably, specific queries demanded intricate reasoning for resolution.
Their research publication delineates:
“Introducing GEO-BENCH, an amalgamation of 10,000 queries drawn from diverse sources, adapted specifically for generative engines, including synthetically generated queries. This comprehensive benchmark integrates queries sourced from nine distinct origins, each meticulously categorized according to target domains, complexity levels, query intents, and other pertinent dimensions.”
Outlined below are the nine primary sources of search queries:
- MS Macro
- ORCAS-1
- Natural Questions
- AllSouls: Comprising essay questions from Oxford University’s All Souls College
- LIMA: Featuring challenging queries necessitating Generative Engines to synthesize information and execute reasoned responses
- Davinci-Debate
- Perplexity.ai Discover: Curated queries from Perplexity.ai’sPerplexity.ai’s trending Discover section
- ELI-5: Incorporating questions sourced from the ELI5 subreddit
- GPT-4 Generated Queries: Supplementary queries generated by prompting GPT-4, spanning various domains (e.g., science, history) and reflecting diverse query intents (e.g., navigational, transactional), along with varying difficulty levels and response scopes (e.g., open-ended, fact-based).
Unveiling the Impact of Nine Website Optimization Techniques
In their research, the scholars experimented with nine optimization methodologies tailored for diverse search categories such as Law and government, business, science, people & society, health, history, and other thematic areas.
Their findings revealed a nuanced responsiveness of niche topics to specific optimization strategies.
The nine strategies subjected to testing encompass the following:
- Authoritative: Adjusting writing style to convey more authoritative claims persuasively.
- Keyword Optimization: Augmenting content with additional keywords derived from search queries.
- Statistics Addition: Substituting interpretative information with statistical data within existing content.
- Cite Sources: Incorporating quotations from credible and reliable sources.
- Quotation Addition: Introducing citations and quotes sourced from high-quality references.
- Easy-to-understand: Simplifying content to enhance comprehension.
- Fluency Optimization: Enhancing the articulation and coherence of content.
- Unique Words: Introducing less common, rare, and distinctive words without altering content meaning.
- Technical Terms: Integrating unique and technical terminology where contextually relevant, maintaining content integrity.
Which strategies proved most effective?
The foremost three optimization tactics were:
- Cite Sources
- Quotation Addition
- Statistics Addition
These strategies showcased relative enhancements of 30-40% compared to the established baselines.
The researchers highlighted the efficacy of these approaches:
“By incorporating pertinent statistics (Statistics Addition), integrating trustworthy quotes (Quotation Addition), and including citations from reputable sources (Cite Sources) into website content, these methods necessitate minimal alterations to the actual content itself. However, they wield substantial influence, amplifying the website’s prominence within Generative Engine responses while augmenting content credibility and depth.”
Which Optimization Strategies Fell Short?
Surprisingly, the researchers found that employing persuasive and authoritative language within content didn’t consistently bolster rankings in AI-driven search engines, unlike the efficacy seen in other approaches.
Likewise, adding additional keywords from search queries into the content yielded disappointing results. Keyword optimization underperformed compared to the baseline by a margin of 10%.
Optimization Varied Across Knowledge Domains
A noteworthy discovery in the study indicates that the effectiveness of optimization techniques varied depending on the knowledge domain (legal, government, science, history, etc.).
For instance, they observed that content associated with the Historical domain received superior rankings when employing the “Authoritative” optimization, employing more persuasive language.
The “Citation” optimization, enhancing content with citations from authoritative sources, notably excelled for factual search queries.
The integration of statistics proved beneficial for Law and Government-related inquiries. Statistics also proved advantageous for “opinion” queries, wherein searchers sought the AI’s viewpoint. The researchers noted:
“These findings suggest that including data-driven evidence enhances website visibility, particularly in these specific contexts.”
Furthermore, the addition of quotations yielded positive outcomes for the People & Society, Explanation, and History domains. This led the researchers to interpret that the AI search engine potentially prioritizes “authenticity” and “depth” for queries in these domains.
Conclusively, the researchers advocated for domain-specific optimizations as the most effective strategy.
GEO Helped Low-Ranking Websites Get Higher Rankings
The good news from this study is that these techniques for optimizing for AI search engines will help websites that are typically ranked lower.
They came to the following conclusion:
Interestingly, websites with lower SERP rankings, which often have difficulty becoming visible, gain considerably more from GEO than websites with higher rankings.
For example, websites ranked fifth in SERP saw a significant 115.1% increase in visibility thanks to the Cite Sources method. In contrast, the top-ranked website saw an average decrease in visibility of 30.3%.
These small content producers have a chance to significantly increase their visibility in Generative Engine results by utilizing GEO techniques.
They can level the playing field and more successfully compete with more prominent companies in the digital sphere by using GEO to improve their content and reach a larger audience.
Revolutionary Approach to SEO
This study offers a fresh approach to SEO for AI-powered search engines. Those who predicted that AI Search would surpass SEO were premature. According to this research, SEO will eventually change into GEO to compete with the upcoming AI search engines.
If you still need help and clarification, look at our monthly SEO packages and get professional assistance.
