From Query Refinement to Query Fan-Out: Search in times of generative AI and AI Agents
The introduction of generative AI, LLMs and AI Agents, represents a significant evolution in search technology. It moves beyond traditional, static search results by employing a sophisticated new query fan-out technique, which fundamentally changes how user queries are processed and how information is retrieved and presented. This shift demands a fresh perspective on search engine optimization (SEO), Large Language Model Optimization (LLMO), Generative Engine Optimization (GEO) … or however you call it.
In this article I would like to cover the impact of the technological foundation of todays query fan out. I also cover the earlier innovation in search query processing like query refinement and query augmentation, which laid the technological foundation for today’s query fanout process.
The query fan-out technique is a fundamental capability in modern AI search systems, playing a critical role in both Retrieval-Augmented Generation (RAG) and the process of grounding information.
This comprehensive article is based on a fundamental research on all query fan out related patents and research paper in the SEO Research Suite and the great article Query Fan-Out: A Data-Driven Approach to AI Search Visibility by Andrea Volpini.
Contents
- 1 What is Query Fan-Out?
- 2 How does Query Fan Out work?
- 3 What is Query Refinement?
- 4 Methodology of Query Refinement
- 5 What is query augmentation?
- 6 What is the role of query fan out for Googles AIOverviews and AI Mode?
- 7 The Role of Synthetic Queries and Query Variants
- 8 Personalization in Query Fan-Out
- 9 How do LLMs influence query expansion?
- 10 How are synthetic queries generated?
- 11 What is the impact of the query fan out for keyword research?
What is Query Fan-Out?
Query fan-out is an information retrieval technique that expands a single user query into multiple sub-queries. Instead of treating each search as a standalone request, AI Mode breaks down the user’s initial question into various subtopics and issues a multitude of queries simultaneously on their behalf. This process allows search engines to delve much deeper into the web than traditional search methods, helping users discover hyper-relevant and comprehensive content.
How does Query Fan Out work?
When a user submits a query, Google’s systems analyze it using advanced natural language processing. This analysis aims to establish user intent, determine the query’s complexity, and identify the type of response needed.
Complex queries, such as “how to optimize for query fan out,” are more likely to activate an extensive fan-out process compared to simple factual queries like “capital of berlin.”
The system “fans out” the original query by exploring various facets and subtopics simultaneously. This is based on semantic understanding, user behavior patterns, and the logical information architecture around the topic. This parallel retrieval of information from different sources—including the live web, knowledge graphs, and specialized data like shopping graphs—significantly expands the information pool available for answer synthesis.
The query fan-out technique plays a vital role in modern AI search systems, particularly in the context of Retrieval-Augmented Generation (RAG) and the process of grounding.
- Expanding User Queries: The core function of query fan-out is to transform a single, potentially complex or ambiguous, user query into a multitude of more specific sub-queries. This process happens on the system’s backend, allowing the search engine to break down the original question into its various facets and underlying user intent. For example, a question about “Bluetooth headphones with a comfortable over-ear design and long-lasting battery” might trigger sub-queries exploring product listings, expert reviews, user experiences, and technical specifications, as well as implicit factors like charging speed or portability.
- Enhancing Retrieval for RAG: Within a Retrieval-Augmented Generation (RAG) framework, query fan-out significantly enhances the “retrieval” component. Instead of a single search, the numerous sub-queries are executed simultaneously across various data sources, including the live web, knowledge graphs, and specialized databases. This parallel execution expands the pool of diverse information available. The retriever component gathers a much richer and more comprehensive set of relevant documents or passages, which are then passed to the language model for synthesis. This ensures the language model has ample contextual information to formulate a detailed and accurate answer.
- Facilitating Grounding: Grounding is the process by which an AI system connects its generated responses to verifiable, real-world information. Query fan-out is crucial for this by identifying and retrieving semantically rich, citation-worthy chunks of content. By casting a wide net with multiple sub-queries, the system can pinpoint specific, relevant segments of information from various sources that directly support different aspects of the original query. This allows the AI to synthesize a comprehensive response that is firmly rooted in factual data, reducing the likelihood of generating inaccurate or unverified information. It essentially provides the evidence base needed for the AI’s answer.
The system then evaluates the content using quality signals, combines information from multiple sources and fan-out queries, and synthesizes a coherent, comprehensive response that addresses the original query while incorporating relevant supporting details.
What is Query Refinement?
Query refinement focuses on enhancing the relevance and accuracy of search results by adjusting or suggesting queries, aiming to provide users with more precise information. This process often involves leveraging historical user data and contextual information.
Methodology of Query Refinement
Here’s a breakdown of the typical steps and factors involved in query refinement:
1. User Query Submission and Initial Processing:
-
- The process begins when a user inputs an initial search query.
- The search engine processes this query to identify relevant documents and generate an initial set of search results.
2. Identifying Related Queries and Generating Candidates:
The system analyzes user session data to identify related search queries that follow the initial query, often refining or building upon its context.
- Various methods are used to generate candidate refined queries or modifications:
- Historical Data Analysis: This involves analyzing query refinements and “super-strings” (broader queries) found in user search histories.
- N-gram Substitution: Modifications of a previous query can be created by substituting n-grams (contiguous sequences of terms) from the previous query with those from the current query.
- Entity Association: The system selects a document from the initial search results and identifies one or more entities associated with it. These entities are then combined with terms from the original query (maintaining term order) to form new refined query candidates.
- Sibling Query Identification: The system identifies “sibling queries” which are related to a child query by sharing common parent queries, based on criteria like fan-out (number of child queries a parent has) and common-query size thresholds.
- Structural Similarity from Documents: New queries can be generated by analyzing “coding fragments” within structured documents to create query templates, which are then applied to other similar documents on the same website.
- Query Triggers from Documents: When a user is viewing a document, the system can identify “query triggers” (terms defined as potential queries) within the document. It then calculates rank scores for these triggers based on their frequency, popularity, and display context, and generates search query suggestions from them.
- Personalized Query Completions: Based on a user’s previous search activities, the system identifies likely queries that are expected to co-occur with a reference parameter in user activity sessions. These are ranked and provided as personalized query completions.
- Passage-Based Suggestions: Instead of just the original query, suggestions can be generated based on selected passages from search results, leveraging knowledge graph entries. These are often presented in a non-intrusive pop-up when users hover over text.
3. Evaluation, Scoring, and Filtering:
- Each candidate refinement or modification is evaluated and scored based on various criteria:
- Relevance: How well it aligns with the user’s current search context, the similarity between n-grams, and geographic location.
- Strength of Relationship: How frequently a “child query” follows a “parent query”.
- Popularity: Frequency of appearance in queries, submission rates, and co-occurrence likelihood of terms.
- Diversity: To ensure that the selected suggestions represent different aspects of user intent and are not too redundant.
- Quality Metrics: Including click-through rate (CTR), dwell time on search results, and inverse document frequency (IDF) of terms.
- Contextual Factors: Such as the user’s search history, current activity, profile data, query length, and device used.
- Clustering: Refinements can be grouped based on inferred user intent using click-through information and session co-occurrence, or by partitioning “visit probability vectors” from random walks on a graph of user search behavior. Clusters can also be formed by identifying refinements closest to a centroid of a term vector.
- Candidates are filtered if they do not meet predefined relationship thresholds or quality thresholds.
4. Ranking, Selection, and Presentation:
- The highest-scoring candidates are selected as the most appropriate refined search queries.
- These refined queries are then presented to the user, either alongside the original search results, as clickable suggestions, or through interactive elements like pop-up overlays.
5. Feedback Loop for Continuous Improvement:
- User interactions with the presented suggestions (e.g., clicks, acceptance, or rejection) are logged and analyzed.
- This data feeds back into the system to continuously refine the scoring mechanisms, update models, and improve the accuracy and relevance of future query suggestions.
What is query augmentation?
Query augmentation is a method that generates and utilizes additional queries to enhance search engine performance, particularly by addressing issues like poorly phrased initial queries or irrelevant results.
Methodology of Query Augmentation
Here’s the step-by-step process for query augmentation:
1. Query Log Analysis:
- The system retrieves a first query from a query log, which records user search queries.
- A quality signal (such as click-through rates and dwell time) is associated with this first query to evaluate its performance. High click-through rates and long click metrics indicate success, while high click-through reversions indicate dissatisfaction. Explicit user feedback (surveys, ratings) can also be used.
2. Performance Threshold Comparison:
- The performance of the first query is assessed against a predefined performance threshold.
- If the query’s performance exceeds this threshold, it is stored in an augmentation query data store. Query frequency (e.g., submitted at least 100 times in 24 hours) can also be a factor.
3. Synthetic Query Generation (for Augmentation):
- In addition to user-generated queries, synthetic queries (artificially generated queries that simulate real user searches) are crucial for augmentation.
- They are generated through:
- Mining Structured Data: Extracting information from structured document data, such as business listings or templated content.
- Document Titles and Anchor Texts: Identifying titles and anchor texts from documents as potential queries.
- Structured Rules: Employing structured rule sets to guide the extraction of meaningful information (e.g., combining business names, locations, and keywords).
- Frequency and Diversity: Generating synthetic queries from commonly searched terms and phrases found in query logs, ensuring they align with high-frequency terms while also diversifying query generation to cover various topics.
- Validation: Testing synthetic queries against historical data to gauge their potential for engagement (e.g., projected CTR and user interaction rates).
4. Augmentation Query Subsystem Activation:
- When a user submits a new search query, it is forwarded to the augmentation query subsystem for evaluation.
- Key terms from the user query are parsed and compared against the stored augmentation queries.
5. Candidate Augmentation Query Selection:
- The system retrieves potential augmentation queries that match the terms or semantics of the user’s query from the augmentation query store.
- These candidates are then ranked based on their similarity to the user query and their historical performance.
6. Augmented Search Operation:
- An augmented search is performed using both the original user query and the selected augmentation queries simultaneously.
7. Return Results and Feedback Loop:
- The search results, which may combine outputs from both the user’s original query and the augmentation queries, are then provided to the user. Results can be differentiated through visual indicators.
- User interaction data (like clicks, engagement time) is continuously collected after results are presented. This data derives new quality signals, which in turn inform the system for future queries, iteratively enhancing its ability to generate relevant results over time.
What is the role of query fan out for Googles AIOverviews and AI Mode?
Query fan-out plays a central and fundamental role in Google’s AI Overviews and AI Mode, enabling them to move beyond traditional, “stateless” search to provide more comprehensive, context-aware, and personalized responses.
Here’s a detailed breakdown of its role:
- Deconstructing Complex User Queries In AI Mode, query fan-out is a technique that breaks down a user’s initial question into multiple subtopics and issues a multitude of queries simultaneously on the user’s behalf. This is particularly beneficial for complex queries that require synthesis from different sources or multi-criteria decision-making. For example, a query like “Could you suggest Bluetooth headphones with a comfortable over-ear design and long-lasting battery?” is deconstructed into facets such as design, technology, and performance, leading to sub-queries targeting product listings, expert reviews, user experiences, and technical specifications. This process allows Google Search to “dive deeper into the web than a traditional search,” uncovering hyper-relevant content.
- Dynamic and Iterative Query Generation Query fan-out is a dynamic process. For instance, in frameworks like the Deep Researcher with Test-Time Diffusion (TTD-DR) — a Google research paper related to LLM-powered deep research agents — search queries are generated dynamically throughout an iterative workflow. Each revision of a draft report and feedback from auto-raters (LLM-as-a-judge) contributes to formulating new queries to fill information gaps. This dynamic generation leads to a “fan-out” where multiple related concepts or areas of inquiry are explored in parallel, increasing the scope of gathered information and reducing the risk of missing critical insights.
- Enhanced Information Retrieval and Synthesis Unlike traditional search where a single query returns one set of results, AI Mode simultaneously retrieves information for all fan-out queries. This parallel processing expands the information pool, which is then evaluated using Google’s ranking and quality signals. The information from multiple sources and fan-out queries is then combined to create a coherent, comprehensive response that addresses the original query. This capability allows the AI to provide a list of recommended products with reasons, detailed specifications with reviews, summaries of features, and sourced links.
- Intent Understanding and Personalization When a user submits a query in AI Mode, Google’s systems use advanced natural language processing to analyze factors like user intent, complexity level, and the type of response needed to determine if fan-out is necessary. This ensures that simple factual queries (e.g., “capital of germany”) might not trigger extensive fan-out, while complex ones (e.g., “how to optimize for query fan out”) would. Moreover, the query fan-out isn’t generic; it’s deeply personalized based on factors such as the user’s search history and interests, location inferred from past activity, and even the device being used.
- Leveraging Synthetic Queries and Generative Models LLMs are crucial for this process, specifically for generative query expansion and synthetic query generation. Synthetic queries are artificially generated queries that simulate real user searches, created by models trained on query-document pairs. These are vital for expanding labeled training data, improving recall, and enabling generative retrieval to scale to large datasets by filling data gaps.
- Connections to Other Google Search Mechanisms Query fan-out is closely related to and enhanced by other core Google search concepts:
- Thematic Search: This system identifies common themes within top search results and can automatically refine the query by combining the original search with a chosen theme, enabling a guided, drill-down exploration without manual input. This is explicitly stated to describe the process of generating themes/passages for AI Overviews and partly AI Mode.
- Subquery Generation: The system can process compound queries by breaking them into multiple subqueries. This is particularly relevant for voice commands, streamlining interactions by allowing users to combine multiple actions or information needs into a single input. The patent on “Subquery generation from a query” explicitly states it describes the Query fan-out process in detail.
- Stateful Chat: In a stateful chat system, synthetic queries are generated based on the user’s evolving state to enhance search relevance, including “drill down” queries and rewritten versions of the original query. This allows for conversational continuity and focuses on task completion rather than just finding relevant documents.
- Query Refinements and Clustering: The concept of query fan-out aligns with identifying alternative phrasings, synonyms, or variations of a query. By analyzing intent scores for similar queries, the system can understand how closely related different queries are, and if an intent score is near a threshold, it can “fan out” to explore queries that capture similar variations or nuances in user intent, thereby enhancing user behavior understanding.
The Role of Synthetic Queries and Query Variants
A critical component of modern retrieval systems, including those leveraging query fan-out, is the use of synthetic queries. These are artificially generated queries that simulate real user search queries. They are created by large language models (LLMs) which learn to generate possible queries that a document might answer. This process helps to:
- Expand training data: By generating query-document pairs for datasets that lack explicit search logs.
- Improve recall: Teaching retrieval models to associate documents with a broader range of search intents.
- Bridge the coverage gap: Ensuring that documents without associated human queries still contribute to training, especially as corpus sizes scale to millions of documents.
Beyond synthetic queries, systems also generate query variants. These are alternative phrasings, synonyms, generalizations, or clarifications of an original query. This dynamic generation of variants, influenced by user attributes and contextual information, enhances the search system’s ability to identify relevant results even for novel or rarely used queries, by effectively broadening the spectrum of potential matches.
Personalization in Query Fan-Out
A key characteristic of AI Mode’s query fan-out is its deep personalization. The sub-queries generated are not generic; they are tailored to individual user contexts. This personalization is driven by:
- User Account Data: Search history, interests, and past interactions directly influence the types of sub-queries generated.
- Location: The system infers a user’s location to suggest geographically relevant results.
- Device: The type of device being used can also tailor suggestions, as query lengths and formulations often vary between mobile and desktop.
- User Intent: Analyzing user behavior patterns helps the system discern evolving intent, allowing it to adapt suggestions dynamically within a multi-turn session.
This means that a “single rank” for a keyword is becoming less meaningful, as search results are hyper-personalized and dynamic, changing based on the user, their location, and their history.
How do LLMs influence query expansion?
Large Language Models (LLMs) fundamentally influence query expansion by moving beyond traditional methods to generate diverse, context-aware, and semantically rich query variations, which significantly impacts search system performance and content strategy.
Here’s a detailed breakdown of how LLMs influence query expansion:
- Generative Query Expansion LLMs are extensively used for generative query expansion, a modern approach that moves beyond traditional Pseudo-Relevance Feedback (PRF) methods. Instead of merely adding terms, LLMs can generate high-quality, short keywords through a reasoning chain to ensure relevance and usefulness. This process helps mitigate sensitivity to noisy keywords and distributional shifts in queries, which can harm stronger rankers. For instance, LLMs can be prompted directly with a query to generate keywords, or they can extract keywords from documents retrieved by PRF in a hybrid approach.
- Dynamic and Iterative Query Generation (Query Fan-Out) LLMs are central to dynamic query generation processes, particularly in frameworks like the Deep Researcher with Test-Time Diffusion (TTD-DR) and Google’s AI Mode.
- In TTD-DR, LLMs dynamically generate search queries during an iterative workflow, where each revision of a draft and feedback from auto-raters (LLM-as-a-judge) contribute to the formulation of new queries. This leads to a “fan-out” of queries, exploring multiple related concepts or areas of inquiry in parallel, thus increasing the scope of information gathered.
- Google’s AI Mode utilizes a query fan-out technique powered by LLMs like Gemini, which breaks down a user’s question into subtopics and issues a multitude of queries simultaneously. This deconstructs complexity and is especially beneficial for multi-criteria decision-making or questions requiring synthesis from different sources. The LLM analyzes user intent, complexity, and the type of response needed to determine if fan-out is necessary.
- Synthetic Query Generation LLMs play a critical role in generating synthetic queries, which are artificially created queries that simulate real user search queries.
- A sequence-to-sequence model (e.g., T5 or GPT) can be trained on real query-document pairs to generate queries that would likely retrieve a given document. These synthetic queries are crucial for expanding labeled training data, improving recall, and enabling generative retrieval to scale to large datasets by filling data gaps.
- Synthetic queries can be generated from structured document data, document titles, or anchor texts. They are designed to align with common search needs, adhere to structured rules, and are validated against performance metrics.
-
- In advanced search systems, synthetic queries generated by LLMs provide significant advantages over simple query expansion by leveraging the user’s full state, maintaining conversational continuity, reinterpreting underlying intent, and focusing on task completion. They are formulated as complete natural language questions or statements.
- Query Variants and Subquery Generation LLMs power the generation of diverse query variants and subqueries:
- A trained generative model (neural network) can generate various types of query variants in real-time based on the original query’s tokens and additional features like user attributes or temporal context. These variants can include equivalent queries, follow-up queries, generalizations, or clarifications.
-
- LLMs can produce variants for novel or rarely used (“tail”) queries by leveraging contextual information and learned patterns, overcoming the limitations of traditional rule-based methods.
-
- For complex, compound queries, LLMs (as part of a subquery generator) can analyze the query and break it into multiple subqueries. This streamlines interactions, especially for voice commands, by allowing users to combine multiple actions or information needs into a single input.
- Intent Understanding and Contextual Awareness LLMs enhance query expansion by providing a deeper understanding of user intent and maintaining contextual awareness:
- Natural Language Understanding (NLU) models, often leveraging LLMs, process user queries to generate candidate intents and intent scores. If an intent score is close to a threshold, the system can “fan out” by exploring queries that capture similar variations or nuances in user intent, thereby enhancing the understanding of user behavior.
-
- LLMs help represent user queries and candidate intents in an embedding space for more precise match evaluations based on semantic similarity.
- In systems providing interactive query suggestions, generative models (transformer models) create contextually relevant suggestions based on selected passages from search results and knowledge graph entries, without necessarily relying on prior user query history.
- Feedback Loops and Continuous Learning LLMs are integrated into feedback loops that continuously refine query expansion:
- Reinforcement learning techniques can be used to optimize generative models, improving the quality of query variants over time based on satisfactory responses.
- User interaction data with search results and generated suggestions (e.g., click-through rates, abandonment) feeds back into the system to fine-tune LLMs and improve future query generation and evaluation processes.
In essence, LLMs have transformed query expansion from a simple term addition process to a sophisticated, dynamic, and intelligent mechanism that anticipates user needs, understands complex intents, generates varied and relevant queries, and continuously learns from user interactions to provide a more personalized and comprehensive search experience.
How are synthetic queries generated?
Synthetic queries are artificially generated queries that simulate real user search queries, created using various methods and models to enhance search engine performance and user experience.
Here’s a detailed breakdown of how synthetic queries are generated:
- • Utilizing Pretrained Language Models (LLMs):
- A pretrained query generator model, typically a sequence-to-sequence (seq2seq) model such as T5 or GPT, is trained on real query-document pairs. This training enables the model to learn how to generate queries that would likely retrieve a given document.
- Once trained, this model is then applied to new or unlabeled documents to generate potential search queries for them. For example, given a document about “The Eiffel Tower is one of the most famous landmarks in Paris,” the model might generate synthetic queries like “What is the most famous landmark in Paris?” or “Where is the Eiffel Tower located?”.
- LLMs can also generate synthetic training datasets by combining received prompts with documents from a corpus, creating numerous query-document pairs. These prompts can be task-specific and guided by a limited number of “few-shot” examples to generate more meaningful and contextually appropriate queries.
- Leveraging Structured Document Data and Patterns:
- Synthetic queries can be generated from structured document data, such as business listings, templated information, document titles, and anchor texts.
-
- This process often involves identifying coding fragments within structured documents and generating query templates based on these fragments. These templates are then applied to other documents on the same website to identify candidate synthetic queries that match the templates’ structure.
-
- The query generation subsystem analyzes data from a structured document corpus, focusing on the structural characteristics of query terms as they appear in the documents.
- Contextual and Dynamic Generation:
- In systems like Google’s AI Mode and generative companions, synthetic queries are created dynamically based on the user’s evolving state and context. This includes the current query, prior queries from the same session, previously viewed search results, user schedule, location, and device information.
-
- The generative model aims to create queries that are alternative suggestions, supplemental queries, rewritten versions of the original query, or “drill down” queries for more specificity.
- For ambiguous queries, the system may generate multiple synthetic queries to explore different possible interpretations.
- The queries are often formulated as complete natural language questions or statements rather than just added terms.
- Iterative Refinement and Feedback:
- In frameworks like Test-Time Diffusion Deep Researcher (TTD-DR), search queries are generated dynamically during an iterative workflow. New queries are formulated based on the original user query and context from previous search iterations.
-
- Feedback loops, often from auto-raters (LLMs acting as judges), help identify gaps or weaknesses in draft responses, leading to the generation of new, more focused search queries to fill those gaps. This ensures the research process is responsive and comprehensive.
- The performance of generated synthetic queries is measured against predefined thresholds. Only those exceeding the threshold are designated as effective synthetic queries.
-
- Round-trip filtering can be employed, where the system verifies if the original document is retrieved when the generated queries are reversed through the retrieval model, ensuring high quality. This process can further refine the training dataset.
- Criteria for Effective Synthetic Query Generation:
- Relevance to the structured data or context from which they are derived, accurately reflecting common search needs.
- Use of structured rule sets that guide the extraction and formatting of meaningful information.
- ◦ Alignment with high-frequency terms or phrases found in query logs.
- ◦ Diversity in query generation, using multiple formats, synonyms, and variations to address various topics.
- ◦ Validation against performance metrics to gauge their engagement potential, such as projected click-through rates and user interaction rates.
By employing these methods, synthetic queries play a crucial role in expanding labeled training data, improving recall, and bridging the gap between document indexing and retrieval tasks, particularly when scaling to large datasets.
What is the impact of the query fan out for keyword research?
... You would like to read more about this exciting topic? You can read the full article as a member of the SEO Resesarch Suite. Complete access to full exclusive blog articles, analysis of the patents, research paper, other SEO related documents and use of AI research tools are only for SEO Thought Leader (yearly), SEO Thought Leader (monthly), and SEO Thought Leader basic (yearly) members.
Your advantages:
+ Get access to the full exclusive paid articles in the blog.
+ Full analysis of hundreds of well researched active Microsoft and Google patents and research paper.
+ Save a lot of time and get insights in just a few minutes, without having to spend hours analyzing the documents.
+ Get quick exclusive insights about how search engines and Google could work with easy to understand summaries and analysis.
+ All patents classified by topic for targeted research.
+ New patent summaries and analysis every week. Weekly notification via E-Mail
+ Use all 4 AI Research Tools to gain insights in seoncds from all documents in the taining databases, the Google Leak Analyzer, Patent & Paper Analyzer, Semantic SEO Research Agent, LLMO / GEO Assistant
+ Gain fundamental insights for your SEO work and become a real thought leader.
Get access to the SEO Research Suite and become a SEO thought leader now!
Your advantages:
+ Get access to the full exclusive paid articles in the blog.
+ Full analysis of hundreds of well researched active Microsoft and Google patents and research paper.
+ Save a lot of time and get insights in just a few minutes, without having to spend hours analyzing the documents.
+ Get quick exclusive insights about how search engines and Google could work with easy to understand summaries and analysis.
+ All patents classified by topic for targeted research.
+ New patent summaries and analysis every week. Weekly notification via E-Mail
+ Use all 4 AI Research Tools to gain insights in seoncds from all documents in the taining databases, the Google Leak Analyzer, Patent & Paper Analyzer, Semantic SEO Research Agent, LLMO / GEO Assistant
+ Gain fundamental insights for your SEO work and become a real thought leader.
Get access to the SEO Research Suite and become a SEO thought leader now!
Already a member? Log in here
- What we can learn about Googles AI Search from the official Vertex & Cloud documentaion - 19. September 2025
- What we can learn from DOJ trial and API Leak for SEO? - 6. September 2025
- Top Generative Engine Optimization (GEO) Experts for LLMO - 3. September 2025
- From Query Refinement to Query Fan-Out: Search in times of generative AI and AI Agents - 28. July 2025
- What is MIPS (Maximum inner product search) and its impact on SEO? - 20. July 2025
- From User-First to Agent-First: Rethinking Digital Strategy in the Age of AI Agents - 18. July 2025
- The Evolution of Search: From Phrase Indexing to Generative Passage Retrieval and how to optimize LLM Readability and Chunk Relevance - 7. July 2025
- How to optimize for ChatGPT Shopping? - 1. July 2025
- LLM Readability & Chunk Relevance – The most influential factors to become citation worthy in AIOverviews, ChatGPT and AIMode - 30. June 2025
- Overview: Brand Monitoring Tools for LLMO / Generative Engine Optimization - 16. June 2025
