Introduction to GEO
Objective and methodology
Objective: The objective of this essay is to explain how Generative Engines (GEs) are changing the way people look for information and how this affects visibility for brands. It first looks at classic search engines and Search Engine Optimization (SEO), then describes how GEs (for example ChatGPT or Perplexity) create answers that reduce clicks. Based on this change in finding information, the main objective of this essay is to define Generative Engine Optimization (GEO) and identify specific methods that help companies show up inside GE answers. A second goal is to evaluate whether common SEO practices (keywords stuffing, unique words) still matter for GEO, which ones have real positive results based on statistics, and which ones are not effective or even harmful. The essay has a value because it brings findings from multiple studies into one spot and makes them into simple, structured set of GEO methods that readers can use.
Methodology: The methodology is a review of recent academic and industry sources for SEO and focused especially on GEs and GEO (from 2024–2025). It goes through each study one by one and explains the methods used and findings. The essay does not do its own research or experiments. Instead, it takes main claims from these studies and puts them into one place. It also mentions how existing studies had to create new ways to measure visibility in GE to explain why some methods work better than others. The key limitation is usage of only secondary sources in a fast changing space, findings show what the recent studies say rather than long-term rules that will be viable in the future,
Table of contents
Objective and methodology
Introduction
1 Background: Search engines, SEO, and the rise of Generative engines
1.1 Traditional search engines and SEO
1.1.1 Explanation of crawling, indexing, ranking
1.1.2 Basic SEO methods
1.2 Rising popularity of Generative engines
2 What are Generative Engines?
2.1 Examples of Generative Engines
2.2 Consequences for actual visibility
3 What Is Generative Engine Optimization (GEO)?
3.1 GEO and its origins
3.2 Goals of GEO
4 Generative Engine Optimization methods
4.1 Review of GEO: Generative Engine Optimization
4.2 Review of What Generative Search Engines Like and How to Optimize Web Content Cooperatively 4.3 Review of Generative Engine Optimization: How to Dominate AI Search
4.4 Review of E-GEO: A Testbed for Generative Engine Optimization in E-Commerce
4.5 Review of Role-Augmented Intent-Driven Generative Search Engine Optimization AND Beyond Keywords: Driving Generative Search Engine Optimization with Content-Centric Agents
Conclusion
References
Introduction
The recent rise of generative engines (GEs) is changing how people find information on the internet. Tools like ChatGPT, Perplexity, and Google’s AI Overviews answer questions in sentences that are easy to understand and don’t require any manual searching through websites. Users now get one clear reply instead of a long list of blue links. They read short summaries and often stop there. They click fewer links and visit fewer sites. For many years, classic search engines were the only way to access web. They crawled pages, stored information about them, and ranked them based on multiple metrics. Search Engine Optimization (SEO) became the new norm of how to be seen online. Site owners either hired SEO specialists or tried learning on their own how to find good keywords, write titles, get backlinks, and structure their pages. Higher ranking in search results meant more visits and more sales. (Aggarwal et al., 2024; Yalçın & Köse, 2010)
Generative engines are starting to break this old pattern. They still use web pages as sources, but they answer in their own words. GEs may mention a brand by name or entirely skip it. They may show a link at the bottom of a long answer. A page can help train the answer but still get no credit. This change raises a question for brands. How can a site stay visible when people no longer click and go through result lists? (Aggarwal et al., 2024)
Recent studies introduce Generative Engine Optimization (GEO) to face this problem. GEO looks at how to get a brand into the answer itself. With a simple goal. brand should appear in the answer often, in a clear way, with correct credit if it’s used. Studies I have reviewed show that small changes to content can drastically change which sources GEs cite and use. For example, writers can add short quotes, clear stats, and direct comparison with competition. They can use page structure that is easy for models to read. (Aggarwal et al., 2024)
This essay explains these ideas in a structured way. First, it gives a short overview of classic search engines and SEO. It covers crawling, indexing, ranking, and common SEO tactics. Next, it describes generative engines and gives real examples. It looks at how GEs affect visits, clicks, and brands. The essay then defines GEO and its main goals. It shows GEO methods such as rewriting of old content, technical tweaks, and earned media. It also shows how studies had to create new ways to measure brand visibility inside GE answers. The essay also looks at old SEO methods and if they work for GEO.
The essay uses only existing academic and industry studies. It does not do new tests or makes its own data. This is a clear limitation of the essay, since generative engines change very fast. Still all the studies used to determine GEO methods are either from 2025 or 2024. Bringing these sources together has value. The essay gives readers one place to learn basics of SEO what GEs are and methods to optimize your website to be more seen in them. It also sparks a thought to think about GEO as a new kind of way to get visibility.
Background: Search engines, SEO, and the rise of Generative engines
Traditional search engines and SEO
Search engines are programs that collect, save, and sort web pages based on a number of metrics. They use small programs called crawlers, spiders, or bots to scan the web. These programs open and follow links, read code, and save page addresses, words, and links. The data then goes into a large index, which is the search engine’s own database. (Yalçın & Köse, 2010)
Search Engine Optimization (SEO) is the process of helping sites appear higher in these results. The main goal of SEO is to move a site toward the first page for chosen keywords. Higher rank brings more visitors, which is very important for companies that primarily use the web to reach their customers. And in the modern day and age almost every company is. (Yalçın & Köse, 2010)
Explanation of crawling, indexing, ranking
Search engines use a pipeline of crawling, indexing, and ranking to answer user queries.
Crawling is the first step. Crawlers move from link to link and collect information about every URL they find. They save the address, visible text, code and links of each page. (Yalçın & Köse, 2010)
Indexing is the second step. The Search engines save the data in its index. The index is a huge database that lets the engine find pages fast when a user sends a query. (Yalçın & Köse, 2010)
Ranking is the last step. When a user types a query, the engine looks in the index for pages that match the words. It then ranks these pages by how relevant and useful they seem. Relevance depends on many factors, such as keyword use, links, and page quality. The engine shows the sorted list to the user, with the “best” pages at the top. (Yalçın & Köse, 2010)
Basic SEO methods
SEO tries to exploit the way search engines rank pages to make the page they are trying to rank more visible. The way search engines work is always changing so the methods to optimize are as well, but below are the proven basic methods of SEO that have worked for a long time. (Yalçın & Köse, 2010) divide SEO into internal (on-site) and external (off-site) work.
Internal SEO looks at how the site is built and written. Important methods include:
• clear, simple HTML pages that load fast
• good structure without heavy use of frames or large Flash parts
• keyword use in titles, headings, meta tags, alt text, and optimized URLs
• reasonable keyword density in the text, not over-use that looks like spam
• unique, short page titles and clear meta descriptions
• clean navigation and a site map that links to all pages (Yalçın & Köse, 2010)
External SEO look at data outside the website. Important methods include:
• backlinks from other websites, which raise page rank 10
• links from sites with similar content, which are better than out website
• adding the website to guides and directories
• use of social media platforms to gain more links and visits (Yalçın & Köse, 2010)
Together, these methods try to help crawlers find the site, help the index save it in the right way, and help the ranking step see it as a strong result for chosen keywords.
Rising popularity of Generative engines
Use of generative engines has grown very fast in the last few years. ChatGPT alone has hundreds of millions of weekly active users in 2025. (Singh, 2025) Other tools such as Gemini, Claude, and Perplexity also reach large user bases. GEs are now common in everyday life. Around half of consumers say they actively use AI-powered search engines. About half of Google searches already show Google AI Overview which is an LLM written summary, and the amount of Google searches with Google AI Overview is rapidly rising. (Silliman, 2025)
Consumers also report that they prefer responses from GEs because of their more easy to understand answers. A bigger share of ChatGPT use is for finding inforamtion, not only writing or coding. Reports show that GEs now influence real product research and purchase choices, across phones, laptops, banking, and more. (Chatterji, 2025; Chen et al., 2025a)
Overall, GEs have moved from small tools to challenging long standing giants for online information and shopping.
What are Generative Engines?
Generative engines (GEs) are search engines combined with generative models that use large language models (LLMs) to search for information and generate an answer using multiple sources. They find documents, websites and information online read them, put important information together and write a full answer in natural easy to understand language. They then add citations that lead back to their sources (Aggarwal et al., 2024; Wu et al., 2025)
This is different from classic search engines. Search engines evaluate your query and then display a long list of blue links ranked by multiple internal metrics. Users then must click and read pages one by one and find the information they were looking for. Generative engines skip all of that. They instead give you a single, compiled answer that feels like a summary of what you were looking for. The links are still used but are hidden in the text as citations (Aggarwal et al., 2024)
Generative engines can also be interactive. Many allow follow-up questions. They can understand tasks you give them, find new sources, and refine their answer in a short dialog. (Wu et al., 2025)
Examples of Generative Engines
This is a non-comprehensive list of GEs, just the most important ones and the ones that were used in the studies that I examined later:
• Google AI Overview shows a short answer at the top of the results page. It uses an LLM with retrieval and cites web pages inside the summary. Google AI Overview might be the biggest threat to traditional SEO because of the way its integrated, users don’t have to visit any external site to use it because it’s used automatically when they search anything on Google (Wu et al., 2025)
• ChatGPT can search the web, read pages, and answer with inline links. Has infinite number of other use cases and functions. It acts as a general-purpose generative engine across many tasks (Wu et al., 2025)
• Perplexity AI presents one main answer plus a list of cited sources. It combines chat with search and is used in several GEO studies as a real test engine (Aggarwal et al., 2024; Chen et al., 2025b)
• Other AI search tools such as Gemini and Claude are also used and tested as GEs in GEO studies (Wu et al., 2025)
All these systems have differences in how they work, use sources and in the way their interface looks. But they share the same basic idea, they use number of sources found online, then generate one understandable answer with citations. It’s also important to note that the way you optimize for each Generative engine is different. (Chen et al., 2025b)
Consequences for actual visibility
Because generative engines answer in full text, they change what “visibility” means. In classic search engines, the key question is: “What rank does my page get?” Users scan the first few links and click one or two. Higher rank means more clicks and more traffic. In generative engines, users often read only the answer on the screen. They may never open the cited sites which changes the way you measure visibility you gain from GEs. I mention some of the new ways you measure visibility below.. (Aggarwal et al., 2024)
What Is Generative Engine Optimization (GEO)?
Generative Engine Optimization (GEO) is a way to make websites show up more in Generative Engines (GEs). It focuses on how often a website is used as a source and and how strong it looks to users. GEO does not entirely replace SEO but build on top of it. SEO cares about position on a results page of a Search engine GEO cares more about how much of an answer is about the subject they are trying to rank and how often the subject is used as a source. This essay goes deeper into what GEO methods are effective and how they differ from SEO methods. (Aggarwal et al., 2024)
GEO and its origins
GEO started when early studies showed that classic SEO tracked metrics no longer matched user experience in AI search. A page could rank high in Google but still be barely cited in a GE answer. First studies tested simple stylistic changes to existing pages, such as adding quotes, statistics, clearer wording, and better structure. They found that these edits increased new impression metrics, while keyword stuffing and similar SEO tricks often failed or even hurt performance. (Aggarwal et al., 2024)
Later studies stopped guessing and began to learn what engines like from data. AutoGEO turns many GE decisions into a set of preference rules and then rewrites already existing pages to follow those rules. (Wu et al., 2025)
Across these studies, GEO is shifting from simple tips to a more formal and data-driven field. The GEO space is still in early stages and is evolving fast.
Goals of GEO
GEO has three main goals that repeat across the studies I used.
The first goal is visibility. Content should show up in more answers, with more cited words, and earlier than competition. New metrics, such as Position-Adjusted Word Count and related scores, try to show just that (Aggarwal et al., 2024; Coelho, 2025)
The second goal is quality and trust. Higher visibility should not make answers worse because that can ruin the trust of your brand. Good GEO keeps facts correct and language clear and avoids tricks that make the users confused. Recent studies measures both visibility and generative engine utility (GEU), GEU is checked by other LLMs that tracks whether the quality and consistency is met. (Coelho, 2025)
The third goal is brand impact. GEO looks at and tries to change how a brand shows up across its own pages and across earned media, such as reviews and articles that GEs often like (Coelho, 2025)
Generative Engine Optimization methods
In this section I will review multiple studies that analyzed ways to be more visible in GEs
and how to perform GEO.
Review of GEO: Generative Engine Optimization
This study (Aggarwal et al., 2024) is one of the first studies on Generative Engine Optimization. The core method they used is to optimize already existing content of a website to make it more visible in GEs. Because the output of a GEs is very different to a normal Search Engine the authors argue that the classic rank metrics no longer work. Because of that they created two new impression metrics to measure their outputs:
“1. Position-Adjusted Word Count, which combines word count and position count. To analyze the effect of individual components, we also report scores on the two sub-metrics separately. 2. 2. Subjective Impression, which is a subjective metric encompassing seven different aspects: 1) relevance of the cited sentence to the user query, 2) influence of the citation, assessing the extent to which the generated response relies on the citation, 3) uniqueness of the material presented by a citation, 4) subjective position, gauging the prominence of the positioning of source from the user’s viewpoint, 5) subjective count, measuring the amount of content presented from the citation as perceived by the user, 6) likelihood of the user clicking the citation, and 7) diversity of the material presented. These submetrics assess diverse aspects that content creators can target to improve one or more areas effectively. Each sub-metric is evaluated using GPT-3.5“ (Aggarwal et al., 2024)
The study also creates GEO-Bench, a benchmark for GEO research that is used to compare how effective certain GEO methods are:
“GEO-bench, a benchmark consisting of 10K queries from multiple sources, repurposed for generative engines, along with synthetically generated queries. The benchmark includes queries from nine different sources, each further categorized based on their target domain, difficulty, query intent, and other dimensions.“ (Aggarwal et al., 2024)
This benchmark is also often used in the studies that I will mention later in the essay.
The study proposes 9 GEO methods. Each method takes the original content and rewrites it
in a specific way. The changes are done by using an LLM, prompted to use specific changes:
“1: Authoritative: Modifies text style of the source content to be more persuasive and authoritative, 2. Statistics Addition: Modifies content to include quantitative statistics instead of qualitative discussion, wherever possible, 3. Keyword Stuffing: Modifies content to include more keywords from the query, as expected in classical SEO optimization. 4. Cite Sources & 5. Quotation Addition: Adds relevant citations and quotations from credible sources respectively, 6.) 6. Easy-to-Understand: Simplifies the language of website, while 7. Fluency Optimization improves the fluency of website text. 8. Unique Words & 9. Technical Terms: involves adding unique and technical terms respectively wherever possible“ (Aggarwal et al., 2024)
They then compare impressions before and after applying each GEO method. The results show clear gains from several methods. Quotation Addition and Statistics Addition give the largest boost in visibility. They raise Position-Adjusted Word Count by about 40% percent on average. Subjective Impression also rises by 20-30%. In comparison standard SEO techniques like Keyword Stuffing can even hurt performance. This shows that methods used in classic SEO do not help in GEO and may even hurt it.(Aggarwal et al., 2024)
The bellow table from the study shows reals examples of how the LLMs added or deleted parts of the text using different methods.
Another part of this study looks at results including low-ranking websites in search results compared to top-ranking websites in search results. Their study finds that GEO helps lower ranked sites much more than top ranked sites. For Rank-5 sites, method Cite Sources can even double visibility. Rank-1 sites see small drops when optimized. This means that GEO may give small sites a way to fight with large well optimized websites. (Aggarwal et al., 2024)
Review of What Generative Search Engines Like and How to Optimize Web Content Cooperatively
This second study I will review (Wu et al., 2025) builds on the first GEO study but has a different goal. Instead of guessing what generative engines like, it tries to learn it from data. The authors call their own framework AutoGEO and it has two main functions
“AutoGEO first learns preference rules by leveraging large language models to automatically analyze the preference usage of retrieved content from generative engines. It employs LLMs to explain the preferences on document pairs with visibility differences, extract these explanations into concise insights, merge insights into candidate rules, and filter insights into preference rules. Through this pipeline, AutoGEO transforms tens of thousands of generative engine preference observations into an actionable set of rules that effectively capture how generative engines prioritize content. AutoGEO then applies the preference rules to construct GEO models, which are used to rewrite target documents and thereby enhance content visibility.“ (Wu et al., 2025)
The GEO model constructed by AutoGEO is called AutoGEOAPI which is a strong LLM like GPT-4 that rewrites content based on set rules by AutoGEO. Visibility is measured with the same GEO metric as (Aggarwal et al., 2024) Position-Adjusted Word Count. But unlike (Aggarwal et al., 2024) this study also measures Generative Engine Utility (GEU) that tracks whether the quality and consistency is met. They do this by using DeepResearchGym framework to determine quality by relevance, faithfulness and quality. (Wu et al., 2025)
This study uses three datasets:
The first, GEO-Bench (Aggarwal et al., 2024), is a largescale GEO benchmark containing diverse user queries across multiple domains. In addition, we contribute two new datasets: Researchy-GEO, an open-domain benchmark featuring high-quality research queries from Researchy Questions (Rosset et al., 2024), and E-commerce, commercial queries filtered from LMSYS-Chat-1M (Zheng et al., 2023). We build generative engines on these datasets and frontier LLMs which include Gemini, Claude, and GPT. (Wu et al., 2025)
Across all of these datasets AutoGEOAPI has the best results in GEO metrics. On most of the metrics it increases visibility by about 30–50% even over the GEO methods used in (Aggarwal et al., 2024) It’s also important that these gains do not affect the answer quality which can be seen on the GEU metrics that stay similar to the baseline.(Wu et al., 2025)
This study shows a progress in GEO from the first one. Instead of using hand-made prompts, it learns what engines like based on comparison and turns those preferences into rules. It shows that prompts based on rules can beat simple prompts and do it without hurting the final quality of the response. The code to AutoGEO is publicly available here: https://github.com/cxcscmu/AutoGEO
Review of Generative Engine Optimization: How to Dominate AI Search
This study chose a different approach compared to the ones I reviewed before. It looks at GEO at the level of brands, not single pages and it starts off by comparing GEs with Google search results. For this they use one thousand “top 10” style prompts like “Rank the best smartphones from 1 to 10” and “Which laptops are considered the best in 2024“. For each query they collect Google’s top links, and the GEs cited links. They mark each domain as Brand (official sites), Earned (reviews, media), or Social. The main result is strong. AI search engines use a lot of Earned media. Brand sites and social platforms like Reddit and YouTube appear way less often. Google on the other hand Is a lot more mixed set of Brand, Earned, and Social links. These statistics holds across sectors like electronics, cars and software. (Chen et al., 2025b)
They then repeat similar tests for local services, languages, and paraphrases. Other experiments compare Claude, GPT, Gemini, and Perplexity on the same queries. Which shows that GEs differ from each other. Claude and ChatGPT mostly cite Earned domains. Perplexity and Gemini use more of Brand and Social links, but still favor Earned. Domain similarity between engines is low, meaning each engine has many different sources. The study finds that language matters a lot, some engines reuse English sites, others switch to local language sites instead. Paraphrasing matters less, changing query a bit changes links a bit, but not the main brands. They also find clear big brand bias. When users ask for “top 20 cola brands” without naming a company, GEs often like giants. Big brands like Coca-Cola and Pepsi show up a lot more in the answers. Small brands appear less often, even though GEs know about them.(Chen et al., 2025b)
From these results, the authors set out a GEO agenda. The main recommended GEO methods are following:
1. Focus on earned media first
The authors of this study state that this is the most important and overwhelming finding, because GEs show visible bias toward earned media compared to Brand and Social media types. While Google chooses its media types more evenly. Based on this study it’s important to change strategy from creating content for you own website more into collaborating with industry leaders and partners to show expertise. And invest in features and reviews in big publications within the industry. (Chen et al., 2025b)
2. Make content easy to scan and justify for GEs
AI needs clear reasons to recommend a product. Pages should offer structured data, comparison tables, pros and cons, and clear claims. Schema markup and clean technical SEO will make sites easy to read for GEs.(Chen et al., 2025b)
3. Plan by engine.
Because engines cite different sources, GEO strategy must be engine specific. For Claude and ChatGPT, the priority should be top review and publisher sites in certain field. For Perplexity focus on YouTube sources and major retail sites is more important. While Gemini likes to cite Brand media types more.(Chen et al., 2025b)
4. Localize in Earned media too, not just text
Simple translation of a website is not enough for non-English markets. To reach certain GEs like Chat GPT, coverage in local Earned media is very important. English language authority in Earned media still helps on engines that reuse English sources like Claude.(Chen et al., 2025b)
5. Cover the whole customer journey
AI shows content not only at discovery but also after purchase. For example, create strong FAQ pages, guides and repair help to get more visibility. If these pages are not present GEs will recommend competitors when users, ask for help.(Chen et al., 2025b)
6. Strategy for smaller brands
The big brand bias is very visible and will be hard to overcome for small brands. Small brands must try get more Earned media authority in narrow niches. Based on the study they: “should try to over-invest in building tangible, verifiable authority. “(Chen et al., 2025a)Deep expert content and reviews can help.(Chen et al., 2025b)
Review of E-GEO: A Testbed for Generative Engine Optimization in E-Commerce
This study tried to optimize GEO for ecommerce. It goes over how they built their own dataset and how they performed their experiments. Because it’s fairly similar to other studies I already mention I will skip these parts and just mention the methods they used and results.(Bagga et al., 2025)
This study tried to optimize product descriptions to see if it will help pages rank better in GEs. The main metric is change in rank. For each query, one of the ten products is used. The authors record its rank before and after rewriting the product description. Rank improvement is measured as old position minus new position, so positive values mean the product moved up. They used LLMs to optimize their product description. First with their own handwritten initial prompts to change and improve linguistic styles of product descriptions. These Initial prompts showed fairly mild improvements with some of them even being negative. Then they asked LLMs to improve their prompts and optimize the product description, these prompts showed significant improvements. After applying LLM prompt optimization, all 15 prompts improved the position in GEs, some by over +1.6 average rank positions. (Bagga et al., 2025)
An important result is that optimized prompts share a similar pattern that includes:
• talking about ranking and outperforming other products
• aligning with user intent and likely queries
• mentioning unique selling points and competitive advantages
• using reviews and ratings as external proof
• persuasive and confident tone
• clear, scannable structure with headings and bullet points(Bagga et al., 2025)
Image below shows heatmaps of which features appear in each prompt, before and after optimization. Initial prompts show mixed and low amount of features.
The best way they found to optimize product description was to tell the LLM to rewrite it to be more competitive and to compare with competition. This strategy was seen very beneficial in the previous study I already mentioned. This is the exact prompt that yielded the best result: (Bagga et al., 2025)
Review of Role-Augmented Intent-Driven Generative Search Engine Optimization AND Beyond Keywords: Driving Generative Search Engine Optimization with Content-Centric Agents
I merged these two studies into one section because they are fairly similar in what methods they used and what results they achieved. They both combine the techniques used in (Aggarwal et al., 2024) and improve on them. They both use their own techniques of improving already existing content but either try to give them intent with the usage of 4W framework (who, what, where, and when) or MACO agent that improves the text, evaluates and chooses the best results and then repeats this process multiple times.(Chen et al., 2025a; Chen et al., 2025c)
The results both showed improved GE visibility showing that 9 baseline GEO methods
proposed by (Aggarwal et al., 2024) work.
Conclusion
Overall, the studies point to good methods for Generative Engine Optimization. Simple SEO tricks like keyword stuffing and vague style changes do not work and can even hurt visibility in GEs. GEO methods that work share the same idea. They make it easy for GEs to trust and reuse a source. High quality content gives clear facts and simple numbers about what it wants to share. It uses simple language, avoids unimportant words and doesn’t run away from the point. It has a clean structure with headings, lists, pros and cons and FAQ. Important points appear early, not hidden at the end of the page.
The studies also show that GEs care about where the information is from. Texts work better when they point to outside sources, reviews, and studies. Pages that explain why a product or answer is good beat pages that just claim it. In ecommerce, the best method is to compare with competition to share unique strengths and evidence from ratings. Newer studies goe beyond hand-made prompts. It learns what engines like from many examples and turns this into rules that it then uses to create better prompts. Optimized prompts show much better results than handwritten ones. LLM models then rewrite pages to follow these rules while keeping the same meaning. All resulting in a better GE readability and in turn website visibility.
Another important method that differs from common SEO strategies is the importance of Earned media. SEO also likes mentions in Earned media but not nearly as much. GEs show a strong bias towards Earned media such as reviews, news sites, and blogs from experts. Brand and social media shows up less often. So, GEO is absolutely not only about content on your website. It also needs a good quality technical SEO and a plan to earn coverage and links from trusted third-party sites. Across all studies, one result is clear. GEO works best when it improves GE answer quality and helps users at the same time. Content that is clearer, has more evidence, and is easier to scan gives websites more visibility in GE answers.
References
Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K., & Deshpande, A. (2024). GEO: Generative Engine Optimization [Indian Institute of Technology Delhi, Princeton University]. https://dl.acm.org/doi/pdf/10.1145/3637528.3671900
Bagga, P., Farias, V., Korkotashvili, T., Peng, T., & Wu, Y. (2025). E-GEO: A Testbed for Generative Engine Optimization in E-Commerce. https://arxiv.org/pdf/2511.20867
Coelho, J. (2025). DeepResearchGym: A Free, Transparent, and Reproducible Evaluation Sandbox for Deep Research [Carnegie Mellon University]. https://arxiv.org/pdf/2505.19253v2
Chatterji, A. (2025). How People Use ChatGPT [Duke University, Harvard University]. https://cdn.openai.com/pdf/a253471f-8260-40c6-a2cc-aa93fe9f142e/economic-research-chatgpt-usage- paper.pdf
Chen, M., Wang, X., Chen, K., & Koudas, N. (2025b). Generative Engine Optimization: How to Dominate AI Search [University of Toronto]. https://arxiv.org/pdf/2509.08919
Chen, Q., Chen, J., Huang, H., & Shao, Q. (2025c). Beyond Keywords: Driving Generative Search Engine Optimization with Content-Centric Agents [Zhejiang University]. https://arxiv.org/pdf/2509.05607
Chen, X., Wu, H., Bao, J., Chen, Z., Liao, Y., & Huang, H. (2025a). Role-Augmented Intent-Driven Generative Search Engine Optimization [University of Science and Technology of China]. https://arxiv.org/pdf/2508.11158
Silliman, E. (2025). New front door to the internet: Winning in the age of AI search. McKinsey & Company. Retrieved December 12, 2025, from https://www.mckinsey.com/capabilities/growth-marketing-and-sales/our- insights/new-front-door-to-the-internet-winning-in-the-age-of-ai-search
Singh, S. (2025). ChatGPT Users Stats (December 2025) – Growth & Usage Data. Demand Sage. Retrieved December 12, 2025, from https://www.demandsage.com/chatgpt-statistics
Wu, Y., Zhong, S., Kim, Y., & Xiong, C. (2025). What Generative Search Engines Like and How to Optimize Web Content Cooperatively [Carnegie Mellon University]. https://arxiv.org/pdf/2510.11438
Yalçın, N., & Köse, U. (2010). What is search engine optimization: SEO? [Gazi Universit]. https://www.researchgate.net/publication/235763091_What_is_search_engine_optimization_SEO/fulltext/026 e55670cf26271f589e16e/What-is-search-engine-optimization-SEO.pdf