Uncategorized

The landscape of online search is undergoing its most radical transformation since the invention of the search engine itself. For decades, searching the web meant typing a query, hitting enter, and browsing through a list of blue links to find the answer. Even with the recent integration of generative AI summaries, search engines have primarily functioned as content aggregators and information synthesizers. That is all about to change. Google has announced a groundbreaking paradigm shift: the ability for users to build their own custom applications directly inside Google Search using agentic coding. Rather than simply retrieving existing web pages, Google Search can now write code, design user interfaces, and build fully functioning, interactive mini-applications on the fly, tailored precisely to a user’s unique requirements. Announced by Liz Reid, the head of Google Search, at the Google I/O conference, this update leverages advanced artificial intelligence to turn Google Search from an information retrieval tool into an active software generation engine. According to Reid, “Search can build the ideal response, in the right format for your question – completely on the fly. So you can get custom generative UI, including visual tools and simulations, tailored precisely to your needs.” What is Agentic Coding in Google Search? To understand the magnitude of this announcement, it is essential to define what “agentic coding” actually means. Traditional generative AI can write code snippets when prompted. If you ask a standard AI chatbot to write a basic HTML calculator, it will generate the raw code for you to copy, paste, and run in your own development environment. Agentic coding in Google Search goes several steps further. It refers to an autonomous AI system (an “agent”) that does not just write the code but executes, tests, renders, and embeds the application directly into your search results page in real-time. The AI determines what kind of application, layout, data integration, and interactive components are needed to solve your query. It then writes the software under the hood, instantly rendering a custom Generative User Interface (UI) that you can interact with immediately. This means searchers no longer have to navigate away from the search engine to use specialized tools, calculators, spreadsheets, or simulators. Google Search becomes the software development platform, generating tailored mini-apps on demand. Three Revolutionary Use Cases for Custom Mini-Apps in Search During the announcement at Google I/O, Google showcased several real-world scenarios where agentic coding completely redefines how users interact with information. These examples demonstrate the range of the technology, from deep educational modeling to multi-step planning and real-time data integration. 1. Real-Time Educational Simulations and Generative UI Understanding highly complex, abstract scientific or mechanical concepts has always been a challenge when relying solely on static text and basic diagrams. With generative UI driven by agentic coding, Google Search can design bespoke, interactive visual simulations on the fly. For example, if you want to understand the intricate physics of astrophysics or visualize exactly how a mechanical watch movement operates, Search does not just serve up articles or instructional videos. Instead, it builds custom layouts in real-time. It assembles interactive dynamic 3D elements, tables, graphs, and live physics simulations. You can slide controls to change variables, watch the visual models adapt in real-time, and explore the concept through a hands-on, custom-built application made specifically for your query. 2. Ongoing Task Widgets and Stateful Trackers Most search queries are transactional or informational, completed in a single session. However, complex life events—such as planning a wedding or coordinating a major home relocation—require weeks or months of continuous tracking, modification, and organization. To solve this, Google Search can now build custom dashboards and trackers. These function as stateful mini-apps that you can return to repeatedly. If you tell Google you are planning a wedding with a specific budget, guest count, and location preference, the search engine will code a custom dashboard complete with budget trackers, checklist widgets, and vendor comparison tables. As you make progress, you can return to Google Search, pull up your custom widget, update your data, and continue managing your project directly from the search engine interface. 3. Custom Fitness Trackers with Live API Integrations Another powerful application of agentic coding in Search is the creation of hyper-personalized tools that pull from real-time external data sources. Google demonstrated this by showing how a searcher can ask for a highly customized fitness tracker. When you ask Google Search to build a fitness plan and tracker tailored to your lifestyle, the AI goes to work coding a custom app. It does not just provide a generic static table. It builds an interactive tracker that pulls in live, real-time data sources. This includes local weather forecasts to suggest optimal outdoor running times, live maps to plot routes, and local business reviews to recommend nearby gyms or healthy dining options. The result is a fully functional, dynamic fitness companion embedded directly within your search experience, helping you stay on track week after week. Under the Hood: How Google Generates Software on the Fly The technology behind this capability relies on a combination of massive language models, real-time code execution environments, and dynamic rendering frameworks. To learn more about the announcement and the underlying technology, you can explore the official Google I/O announcement blog post. Historically, search engines worked by indexing pre-existing documents. When a query was entered, the system matched the keywords to the indexed documents and ranked them. With the advent of Retrieval-Augmented Generation (RAG) and Large Language Models, search engines began generating textual summaries of those documents. Agentic coding represents the next tier of this evolution. When a user submits a query that requires more than a simple text answer, Google’s AI agents interpret the intent as a software requirement. The system then: Drafts the Application Architecture: The AI decides what components are needed (e.g., input fields, data visualization charts, interactive maps, or buttons). Writes and Compiles Code: The agent writes the necessary frontend and backend code to make the widget functional. Assembles Generative UI: The layout is

The Dawn of Agentic Search The search landscape is undergoing its most profound transformation since the invention of the modern search engine. For decades, searching the web has been a transactional, query-and-response activity. Users input a keyword phrase, browse a list of links, and manually piece together the information they need. However, Google is fundamentally altering this dynamic by introducing autonomous capabilities directly into its core platform. Google has unveiled a suite of new search agents, including specialized information agents and advanced agentic experiences designed to automate complex, multi-step tasks. Instead of merely serving as an index of the web, Google Search is transitioning into a proactive coordinator that works on behalf of the user in the background. “We’re entering the era of Search agents, where you can easily create, customize and manage multiple AI agents for your many tasks, right in Search,” said Liz Reid, the head of Google Search. This shift marks a major milestone in Google’s evolution, moving from passive information retrieval to active task execution. What Are Google’s New Information Agents? At the center of this update is the “information agent,” a persistent AI assistant that lives within Google Search. Unlike traditional search queries that end the moment you close your browser tab, an information agent is designed to run continuously. It acts as an ongoing monitor, keeping track of your long-term goals, tasks, and interests. To achieve this, the information agent scans the entire web. It continuously monitors diverse sources, including blogs, news outlets, and social media platforms, while simultaneously tapping into Google’s freshest, real-time data streams. This includes live feeds for financial markets, sports scores, and e-commerce inventory. When the agent detects a relevant change or a new piece of information that aligns with your criteria, it synthesizes the findings and presents you with an actionable update. This persistent monitoring model transforms search from an active pull mechanism to an automated push mechanism, saving users hours of repetitive manual searching. How Information Agents Work in Practice To understand the practical value of information agents, consider the common, time-consuming tasks that typically require days or weeks of manual research. Google highlighted two primary use cases that demonstrate the power of this technology: Continuous Apartment Hunting: Finding a new home is historically tedious, requiring house hunters to constantly refresh multiple listing sites. With an information agent, you can input a detailed list of exact requirements—such as budget, neighborhood boundaries, square footage, pet policies, and proximity to transit. The agent will then continuously scan real estate sites across the web, instantly notifying you the moment a listing matches your exact parameters. Tracking Exclusive Product Drops: If you are a collector or a fan of a particular athlete or designer, keeping up with limited-edition releases can be incredibly difficult. You can instruct your information agent to monitor the web for specific announcements, such as an athlete revealing a new sneaker collaboration. The agent will alert you the instant the drop goes live, providing a direct path to purchase before the product sells out. By delegating these repetitive research tasks to an autonomous agent, users can easily manage multiple ongoing projects simultaneously without cluttering their cognitive load or their browser tabs. Availability and Rollout Plan These highly anticipated information agents will begin rolling out in the summer. Initially, access to these advanced agentic capabilities will be exclusive to subscribers of Google’s premium tiers, specifically those with Google AI Pro and Ultra subscriptions. This phased rollout allows Google to refine the technology with a subset of power users before potentially expanding it to a broader global audience. Expanding Agentic Experiences with Automated Booking Beyond information gathering, Google is pushing the boundaries of transactional search by expanding its agentic booking capabilities. This feature allows Google Search to act as an intermediary that can execute real-world transactions and bookings on your behalf, focusing heavily on local experiences, hospitality, and service industries. Rather than requiring users to visit multiple booking platforms, compare schedules, and fill out redundant forms, Google’s agentic booking handles the logistics end-to-end. If you are planning an event or looking for a highly specific local service, you can describe your exact criteria directly to Google Search. The system will parse the web to find matches, cross-reference real-time availability, aggregate pricing, and present direct booking options. Solving Complex Local Search Queries To illustrate how this works, Google demonstrated a complex local search scenario. Imagine you want to book a private venue for a group gathering on a specific evening. Your requirements are highly detailed: you need a private karaoke room, a specific timeframe, and a venue that serves a particular type of cuisine. In a traditional search environment, this would require cross-referencing multiple restaurant review sites, checking menus, calling venues to confirm private room availability, and navigating separate reservation portals. With Google’s agentic booking, the AI agent handles the entire process. It identifies the venues that meet your exact specifications, verifies real-time opening slots, checks pricing, and compiles direct links so you can complete the booking instantly. Industries Primed for Agentic Booking Google plans to roll out these transactional booking experiences this summer in the United States. The initial rollout will support several key service sectors, including: Home and Repair Services: Finding and scheduling local plumbers, electricians, or HVAC technicians with immediate availability. Beauty and Wellness: Booking appointments at salons, spas, or specialized wellness clinics that match specific service criteria. Pet Care: Securing slots for pet grooming, boarding, or veterinary services. Local Entertainment and Leisure: Reserving venues, booking recreational activities, and securing specialized dining experiences. The Global Expansion of Personal Intelligence In tandem with these search-specific agents, Google is significantly scaling the geographic and linguistic reach of its Personal Intelligence features. Previously restricted to limited testing markets, Personal Intelligence in AI Mode is expanding to approximately 200 countries and territories, supporting 98 languages. Personal Intelligence allows Google’s AI models to safely connect and synthesize information from your own digital life. By securely integrating with your Gmail, Google Photos, Google Workspace

Google has officially ushered in a new era of search technology. At its highly anticipated Google I/O event, the tech giant announced the launch of its latest and most efficient AI model to date: Gemini 3.5 Flash. This powerful new model is already active, transforming how users interact with information by powering Google Search’s AI Mode globally. The roll-out marks a significant milestone in Google’s ongoing efforts to integrate generative artificial intelligence directly into its core products. According to Liz Reid, Google’s Head of Search, Gemini 3.5 Flash is the company’s “newest Flash model delivering sustained frontier performance for agents and coding.” By integrating this model directly into Google Search, the company aims to deliver faster, more accurate, and highly sophisticated answers to complex user queries across the globe. But the integration of Gemini 3.5 Flash extends far beyond the standard search bar. From consumer-facing applications to developer platforms and enterprise-level ecosystems, Google is deploying this model to fundamentally change how we interact with technology. For digital marketers, SEO specialists, and webmasters, this update represents a critical shift in how search engines process information and deliver organic traffic. What is Gemini 3.5 Flash? Historically, artificial intelligence models have forced developers and platforms to make a difficult trade-off: choose a massive, highly capable model that suffers from high latency and high operational costs, or opt for a smaller, faster model that lacks deep reasoning capabilities. Google’s “Flash” series was originally designed to bridge this gap, focusing heavily on speed and efficiency. With the release of Gemini 3.5 Flash, Google has successfully challenged this compromise. This model is engineered to deliver the deep, agentic reasoning capabilities of a flagship model while maintaining the lightning-fast response times that search engines demand. It is optimized for high-frequency, complex tasks, making it the perfect engine to power real-time conversational search experiences on a global scale. Not only is Gemini 3.5 Flash driving the new AI Mode in Google Search, but it is also available immediately in the standalone Gemini app. Crucially, Google has made this advanced model accessible to all users of the Gemini app, not just those subscribed to its paid tiers. This democratizes access to frontier-level AI and sets a new baseline for what free conversational assistants can achieve. Breaking Down the Technical Benchmarks To understand why Gemini 3.5 Flash is such a major leap forward, it helps to look at the hard data. Koray Kavukcuoglu, the Chief Technology Officer of Google DeepMind and Chief AI Architect, shared several impressive technical achievements that highlight the model’s capabilities. According to Kavukcuoglu, Gemini 3.5 Flash delivers intelligence that rivals large, flagship models across multiple dimensions, all while operating at the rapid speeds expected of the Flash series. In fact, it has established itself as Google’s strongest model yet for coding and agentic workflows, outperforming even the highly regarded Gemini 3.1 Pro on challenging industry benchmarks. Some of the key performance metrics highlighted by Google DeepMind include: Terminal-Bench 2.1 (76.2%): This benchmark measures an AI’s ability to interact with terminal interfaces, run command-line tools, and solve complex system-level coding challenges. Scoring 76.2% demonstrates exceptional proficiency in executing technical, multi-step actions. GDPval-AA (1656 Elo): This metric evaluates agentic capabilities and data processing workflows. An Elo rating of 1656 places Gemini 3.5 Flash in a highly elite class of models capable of executing long-horizon tasks and logical reasoning. MCP Atlas (83.6%): Assessing the model’s ability to utilize tools, integrate external data sources, and operate within the Model Context Protocol, Gemini 3.5 Flash scored an impressive 83.6%, proving its readiness for complex, real-world agentic software applications. CharXiv Reasoning (84.2%): A premier benchmark for multimodal understanding, this test evaluates how well a model can interpret, analyze, and reason over visual charts, diagrams, and scientific papers. A score of 84.2% demonstrates that Gemini 3.5 Flash is highly visually literate. Beyond reasoning and comprehension, the defining feature of Gemini 3.5 Flash is its raw speed. When measuring output tokens per second, the model is four times faster than other frontier models on the market. In the independent Artificial Analysis index, Gemini 3.5 Flash landed squarely in the top-right quadrant—the sweet spot that indicates a model delivers frontier-level intelligence at exceptional speed. This benchmark proves that developers and users no longer have to trade quality for low latency. Broad Integration Across the Google Ecosystem While the update to Google Search is generating the most buzz in the digital marketing space, Google is simultaneously rolling out Gemini 3.5 Flash across its entire developer and enterprise ecosystem. This widespread deployment ensures that the model will become the backbone of various modern workflows. For Developers and Engineers Google has integrated Gemini 3.5 Flash into its primary development environments. It is now live in Google Antigravity, as well as the Gemini API within Google AI Studio and Android Studio. This allows software engineers to build faster, more intelligent applications, leverage highly responsive auto-completion, and construct complex agentic workflows directly within their coding environments. For Enterprise and Business Operations In the enterprise space, Gemini 3.5 Flash is now powering the Enterprise Agent Platform and Gemini Enterprise. Businesses can leverage the model’s high-speed reasoning to automate customer support, analyze massive data sheets in real time, and build custom internal agents that can execute tasks across disparate business systems without lagging or crashing. Why the SEO and Digital Marketing Industry Must Pay Attention The integration of Gemini 3.5 Flash into Google Search is not just a technical milestone; it is a paradigm shift for search engine optimization (SEO) and digital publishing. Because the model is already powering Google Search’s AI Mode globally, it is highly likely that it will soon become the primary engine behind AI Overviews (formerly known as the Search Generative Experience, or SGE). For years, search engines functioned as directories, matching keywords to index pages and serving a list of blue links. With Gemini 3.5 Flash, Google is rapidly transitioning from a search engine into an “answer engine.” Here is why this transition matters to your

For more than a quarter of a century, the Google homepage has been defined by its iconic, minimalist design: a clean white screen with a simple, static rectangular box in the center. While the algorithms behind it have evolved dramatically, the physical act of typing a few keywords into that box has remained largely unchanged. That is, until now. Google has officially unveiled the most significant redesign to its search bar in 25 years: the Intelligent Search box. This update represents a major paradigm shift in how billions of users will interact with the internet. Driven by a desire to make cutting-edge artificial intelligence tools instantly accessible, the new design completely redefines the core Google interface, turning a simple keyword search bar into an interactive, multimodal AI prompt window. At the heart of this massive transition is Google’s latest and most efficient AI model, Gemini 3.5 Flash. This integration signals that AI is no longer just an experimental feature or a secondary tab in Google Search—it is now the core interface through which users will navigate the web. The Redesigned Search Box: A Dynamic, Expanding Canvas The new Intelligent Search box is designed to accommodate how human curiosity actually works. Traditional search boxes have always limited users to short, fragmented keywords because of their physical and functional constraints. The new interface breaks these boundaries entirely. As you begin typing a query, the Intelligent Search box dynamically expands, giving you more physical space to formulate long, detailed, and highly contextualized prompts. Instead of condensing your thoughts into a simple phrase like “best family SUVs,” you can type an entire paragraph detailing your budget, safety requirements, preferred brands, and fuel-efficiency needs. The box expands gracefully to fit your input, encouraging a more conversational approach to search. According to Liz Reid, Google’s Head of Search, this redesign introduces an AI-driven suggestion system that “goes beyond autocomplete.” Standard autocomplete relies heavily on historical search volume and trending queries to guess your next word. The Intelligent Search box, however, uses the contextual reasoning capabilities of Gemini to understand the underlying intent of your question. It actively helps you structure complex, multi-part queries on the fly, offering smart suggestions that anticipate the direction of your research. Embracing the Multimodal Era: Beyond Text Queries The web is no longer made of text alone, and the way we search shouldn’t be either. The Intelligent Search box fully embraces multimodal inputs, allowing you to search using a wide variety of formats right from the home screen. Users can now easily search with: Text: Traditional keyboard inputs, now optimized for long-form, conversational prompts. Images: Seamless integration of Google Lens directly inside the search box, allowing for instant reverse-image searches and object identification. Files: The ability to upload PDFs, spreadsheets, or text documents directly into the search bar to ask questions, summarize content, or extract key data points. Videos: Users can upload or record video clips to ask complex questions about dynamic events, such as troubleshooting a flickering appliance or identifying a specific technique in a sports clip. Chrome Tabs: A groundbreaking feature that lets you search and synthesize information across your currently active browser tabs, creating a unified workspace. By putting these advanced capabilities directly at the user’s fingertips, Google is lowering the barrier to entry for highly complex AI tasks, making them accessible to everyday web surfers. The Engine Under the Hood: Gemini 3.5 Flash To power millions of highly complex, multimodal queries every single second without causing lag, Google needed a model that was incredibly fast, efficient, and deeply intelligent. Enter Gemini 3.5 Flash. Gemini 3.5 Flash is engineered specifically for speed and high-frequency workloads. It features a massive context window, allowing it to process vast amounts of information—such as lengthy documents or high-resolution video files—in a fraction of a second. This makes it the perfect engine for a real-time search interface, where users expect near-instantaneous feedback. By deploying Gemini 3.5 Flash inside the core search box, Google achieves the perfect balance between speed and reasoning. The AI can rapidly parse complex, multi-layered prompts, reference the live web, and generate highly accurate summaries without forcing the user to wait. Seamless Integration: Moving from Search to AI Mode Alongside the hardware-level interface changes, Google has globally rolled out its AI Overviews seamless link approach to AI Mode. First tested with a limited user base back in January, this feature is now fully live on both desktop and mobile devices worldwide. This update bridges the gap between traditional search engine results pages (SERPs) and conversational AI. Previously, if a user wanted to ask a follow-up question to an AI Overview, the transition could feel disjointed. Now, when you ask a follow-up question within an AI Overview, the interface instantly and seamlessly transitions you into a dedicated, full-screen “AI Mode.” This transition feels less like loading a new webpage and more like continuing an ongoing conversation with an expert. It allows you to dig deeper into complex topics, refine your criteria, and explore nuanced perspectives without ever losing the context of your original query. Why the Intelligent Search Box Matters to SEOs and Content Creators For digital marketers, search engine optimization (SEO) professionals, and content publishers, this update represents one of the most significant shifts in the history of the web. The redesign of the search box directly influences how users seek information, which will inevitably alter the flow of organic traffic. 1. The Rise of “Zero-Click” Searches and AI Mode As the search box makes it easier to engage directly with AI, more users may find the answers they need entirely within Google’s own ecosystem. When the Intelligent Search box immediately steers users toward an interactive AI Mode or a comprehensive AI Overview, the need to click through to external websites may decrease for basic informational queries. Publishers will need to focus on producing deeply analytical, opinion-based, or highly specialized content that AI cannot easily replicate in a quick summary. 2. The Evolution of Search Queries Because the new

In the current digital landscape, the most frequent question marketing professionals face is no longer about search volume or keyword difficulty. Instead, it is a question of measurement: How do we track our brand’s presence in ChatGPT? How do we know if Perplexity is recommending us? Does our work on grounding for AI-native search modes actually move the needle? As of 2026, the industry has yet to find a definitive, “out-of-the-box” solution. Any platform or consultant promising a clean, real-time dashboard that tracks grounding presence, display visibility, and conversion actions across search engines, assistive AI, and autonomous agents simultaneously is likely overpromising. Most current solutions provide little more than a “best guess” snapshot based on traditional search data that doesn’t fully translate to the agential era. The common advice—to track a list of queries you *think* users might ask—is fundamentally flawed. These lists are often built for convenience, mapping to existing SEO efforts rather than the unpredictable, conversational nature of AI interactions. To solve the measurement problem, we must stop looking for a precise micro-metric and instead adopt a macro-framework. This is the “Funnel Query Pathway.” The Visibility Paradox: Why Precision is the Wrong Goal The desire for a single, precise number on a dashboard is a leftover instinct from the last twenty years of traditional search. In that era, the surface was finite, rankings were relatively stable, and the click was a measurable, observable event. However, AI-driven assistive and agential surfaces operate differently. They are opaque, highly personalized, and geographically fragmented. Rather than seeking a precise KPI that doesn’t exist, marketers should look toward the discipline of macroeconomics. Economists measure systems that are too complex and opaque for direct observation by looking at signals, trends, and systemic health. The Funnel Query Pathway is a methodology that applies this macro instinct to brand measurement. It isn’t just a measurement tool; it is an operational artifact that combines strategy, measurement, and analysis into one cohesive workflow. Why AI Visibility is a Macroeconomic Problem The structural reasons why AI visibility defies traditional measurement mirror the challenges of macroeconomics. In a micro-environment, like a local retail shop, you can count every item of inventory. In a macro-environment, like a national economy, a central bank cannot observe every single transaction; it must rely on indicators. AI ecosystems are macro-environments for three primary reasons: 1. Brand-User-Algorithm (BUA) Opacity The internal state of a Large Language Model (LLM) is not observable in the way a search index used to be. The user cannot see which alternative brands the algorithm rejected. The brand cannot see the full journey within the “walled garden” of the AI chat. Perhaps most importantly, even the algorithm’s creators often cannot fully introspect on exactly why a specific recommendation was made at a specific moment. This BUA opacity makes direct tracking impossible. 2. Extreme Personalization In the AI era, there is no “standard” result. Every user receives a tailored answer based on their personal context, previous interactions, and real-time intent. This is the equivalent of “heterogeneous agents” in economics—everyone acts differently, and the system responds to them as individuals, making a single “ranking” number meaningless. 3. The Explosion of Interaction Surfaces The “search” surface has exploded beyond the browser. We now interact with AI through Copilot in Microsoft Word, ChatGPT inside Slack, Perplexity in Comet, and Apple Intelligence baked into the OS. We see it in hardware like the dedicated Copilot key on Lenovo laptops or Samsung’s Galaxy AI. This “ambient research” means recommendations often happen unprompted, based on environmental context, making the traditional query-to-click model obsolete. The New Unit of Measurement: The Cohort To measure within this complex system, we must change our unit of measurement. Traditional SEO groups queries by category (e.g., “Phuket hotels”). However, categories group things, whereas cohorts group people. Intent is about people, not objects. A query like “Phuket hotels” is a destination, not an intent. The person searching for “5-star luxury resorts in Phuket” and the person searching for “cheap hostels in Phuket” share a destination but have nothing else in common. They have different budgets, different decision-making criteria, and different downstream behaviors. If you group them together, you average your performance across two entirely different audiences, leading to muddy data. AI algorithms, such as those powering Gemini’s recommendations or Google’s Performance Max, don’t ask what category a query is in. They ask: “What cohort does this user belong to, and what is their specific intent?” The Intersection of Cohort and Intent The Funnel Query Pathway defines a “node” as the intersection of a durable cohort and a situational intent. This is where behavioral coherence lives. Defining the Cohort A cohort is defined by a durable identity. For example, “luxury travelers,” “parents shopping for toddlers,” or “IT procurement managers” are cohorts. These identities persist across time. A luxury traveler is still a luxury traveler whether they are booking a flight in July or buying a watch in December. Defining the Intent Intent is the situational vector. It is the “what” and “why” of a specific moment. Buying a winter coat, booking a weekend getaway, or upgrading a server are intents. Each intent can span many cohorts, but the way they approach that intent will differ wildly. The “node” is the meeting point: “Luxury travelers (Cohort) booking a hotel in Bali (Intent).” When you identify this intersection, you find a group of people who will behave in a similar way given a specific stimulus. This behavioral coherence is what makes a node trackable even within an opaque AI system. Qualifying Queries for the Pathway A query only qualifies as a node in the Funnel Query Pathway if both the cohort and the intent are legible within the query itself. Consider these examples: “Hotels in Bali”: This query shows intent but hides the cohort. It could be a backpacker or a billionaire. It cannot function as a stable node. “Cheap hostels in Bali”: Here, the budget cohort emerges alongside the intent. This is a qualified node because the

Reasoning lift: What happens to brand visibility when AI thinks harder The landscape of search is undergoing its most radical transformation since the advent of the mobile web. For decades, SEO professionals focused on ranking factors, backlink profiles, and keyword density within the confines of a traditional search engine results page (SERP). However, the rise of Large Language Models (LLMs) and conversational AI has introduced a new variable: reasoning. When an AI model stops to “think” or reason through a complex prompt, the way it interacts with the web—and the brands it chooses to cite—shifts fundamentally. Recent data-driven insights into GPT-5.2 suggest that we are entering an era of the “Reasoning Lift.” This phenomenon describes the surge in citation rates, search depth, and brand persistence that occurs when an AI model utilizes high-reasoning capabilities versus minimal reasoning. For digital marketers and SEOs, understanding this shift is no longer optional; it is the key to maintaining visibility in a world where AI agents do the research on behalf of the consumer. The Evolution of AI Search: From Chatbots to Reasoning Engines To understand the “Reasoning Lift,” we must first distinguish between standard conversational AI and reasoning-heavy models. Most users are familiar with the basic chatbot experience: you ask a question, and the LLM provides an answer based on its training data or a quick web retrieval. This is “minimal reasoning.” High reasoning, however, involves a more sophisticated process. When a model encounters a complex, multi-layered query, it doesn’t just pull a single answer. It breaks the prompt down into sub-tasks, performs multiple internal searches (known as fan-out queries), evaluates conflicting information, and synthesizes a comprehensive response. This “Thinking Mode” mimics human analytical processes, and as the data shows, it fundamentally changes which parts of the internet the AI decides to trust. Methodology: Measuring the Impact of Reasoning on SEO The insights discussed in this analysis are derived from a comprehensive study using the Semrush AI Visibility Toolkit. The goal was to track how GPT-5.2’s behavior changes when toggling between minimal and high reasoning across various stages of the consumer purchase path. The study analyzed 100 distinct prompts, each run twice (once in each reasoning mode), totaling 200 unique responses. These prompts were mapped across 20 different buyer journeys in four critical verticals: B2B SaaS, Finance, Consumer Tech, and Health/Lifestyle. To ensure a holistic view of the funnel, the journeys were divided into five stages: Problem: The user identifies a need or pain point. Exploration: The user researches potential types of solutions. Comparison: The user evaluates specific brands or products against one another. Validation: The user seeks social proof, pricing verification, or compliance data. Selection: The user looks for “how-to” guides or final onboarding steps. By tracking citation rates (the percentage of responses citing external sources), average citation counts, and fan-out queries, the study revealed a stark divergence between how “fast” AI and “slow” AI treat brand visibility. The Core Findings: High Reasoning Cites and Searches More The most immediate takeaway from the data is that when an AI model thinks harder, it relies more heavily on the live web. This is a crucial finding for SEOs who feared that LLMs would eventually “close” the ecosystem and stop sending traffic to websites. When high reasoning is activated in GPT-5.2, the citation rate jumps from 50% to 68%—a massive 18 percentage point increase. Furthermore, the average number of sources cited per response nearly doubles, moving from 2.6 to 4.5. Perhaps most significantly, the “fan-out” queries—the internal searches the AI performs to fact-check or expand its knowledge—increase by a factor of 4.6x. A Different Web: The Domain Overlap Gap One of the most startling revelations is that high reasoning doesn’t just cite more of the same sites; it cites a different web entirely. The study found only a 25.6% domain overlap between minimal and high reasoning modes. Out of the 173 unique domains cited during high-reasoning tasks, 99 of them never appeared in the minimal reasoning responses. This suggests that high reasoning “unlocks” a deeper layer of the internet. While minimal reasoning might stick to high-authority, generalist sites that are frequently found in training data, high reasoning digs into niche documentation, regulatory filings, and specific technical guides to provide a more accurate answer. If your brand is only visible on “top 10” listicles but lacks deep, authoritative technical content, you may vanish when the AI enters reasoning mode. How Reasoning Scales Across the Buyer Journey The gap between minimal and high reasoning is not a flat line; it fluctuates based on the user’s intent and where they are in the sales funnel. The model’s behavior effectively resembles an “hourglass” shape across the different stages of the journey. Early Funnel: The Research Gap In the Problem and Exploration stages (Top-of-Funnel or TOFU), the differences are most pronounced. Under minimal reasoning, the AI often answers from its internal weights—effectively answering “from memory.” However, under high reasoning, the model treats these early questions as research tasks. At the Problem stage, high reasoning showed a +35 percentage point increase in citation rates compared to minimal reasoning. Middle Funnel: The Investigation Peak The Comparison stage is where the “Reasoning Lift” reaches its peak. This is the “mini-investigation” phase. In this stage, high reasoning fires an average of 24.1 sub-queries per response, compared to just 5.5 for minimal reasoning. This is because comparing brands requires the AI to verify specific features, pricing tiers, and compatibility requirements across multiple sources simultaneously. Late Funnel: Specificity Drives Search In the Validation and Selection stages, the gap narrows but remains significant. Interestingly, the Selection stage showed the highest variance in search behavior. Prompts that were highly “bounded” or structured (e.g., “Draft an RFP for an agency”) required fewer searches. However, open-ended “Selection” prompts (e.g., “Build me a $3,000 home gym shopping list”) triggered as many as 40 fan-out queries. The lesson for marketers? The more degrees of freedom a prompt has, the more the AI will search the web to fill in the

For years, the standard workflow for SEO reporting was predictable, if a bit tedious. A typical Monday or end-of-month session involved logging into Google Search Console (GSC), exporting multiple CSV files, and spending hours cleaning that data in Excel or Google Sheets. From there, you would manually port that information into Looker Studio (formerly Data Studio) or a slide deck to create something presentable for stakeholders. While these dashboards served a purpose, they were often rigid, slow to update, and limited by the constraints of the visualization software. The rise of AI coding agents is fundamentally shifting this paradigm. We are moving away from static dashboards and toward dynamic, code-driven reporting environments. Tools like Claude Code—Anthropic’s terminal-based interface—allow SEO professionals to bypass the manual labor of data cleaning and visualization. By leveraging AI to write and execute code locally on your machine, you can transform raw Google Search Console data into polished, high-level reports in a fraction of the time it used to take. This guide will walk you through the process of setting up Claude Code, connecting it to Google Search Console, and building a custom reporting framework that adapts to your specific SEO needs. What is Claude Code and How Does it Differ from Claude.ai? Before diving into the technical setup, it is important to understand what Claude Code actually is. Most people are familiar with Claude.ai, the browser-based chatbot where you type prompts and receive text or code snippets in return. While powerful, the browser interface has limitations: it cannot interact with your local files, it cannot run scripts on your machine, and it has a limited context window for massive datasets. Claude Code is different. It is a command-line interface (CLI) tool designed for developers and power users. It functions as an AI coding assistant that lives in your terminal. Because it operates locally, it can read your project folders, write files directly to your hard drive, execute terminal commands, and even manage complex software dependencies. For an SEO professional, this means Claude Code can act as an automated data scientist, processing thousands of rows of GSC data and generating visual reports without you ever needing to open a spreadsheet. Instead of merely generating a response, Claude Code creates a local reporting environment. It treats your SEO data as a software project, allowing for deeper analysis, better version control, and much more sophisticated visualizations than a standard web-based chatbot could provide. Understanding the Learning Curve It is worth noting that using Claude Code requires a higher level of technical comfort than using a standard AI chat interface. If you are not a developer, the initial setup can feel intimidating. You will be working in a terminal (Command Prompt or PowerShell on Windows, Terminal on Mac) and interacting with APIs. The “reports in minutes” promise is real, but it applies to the long-term workflow. The initial configuration—installing environments, setting up Google Cloud permissions, and establishing a framework—may take a few hours. However, this is a one-time investment. Once the foundation is laid, you will be able to generate complex, custom reports with simple natural language commands. For those working in an enterprise or agency setting, you can often bridge this technical gap by collaborating with an internal developer for the initial setup. Once the environment is configured, the SEO team can take over the day-to-day reporting tasks. Step 1: Setting Up Your Environment To begin building your custom SEO reports, you need to prepare your machine to run Claude Code. The tool runs on Node.js, which is a JavaScript runtime environment. Install Node.js Claude Code requires Node.js to function. If you are on a Mac or Windows machine, you can download the latest Long-Term Support (LTS) version from the official Node.js website. If you are using a Chromebook, you can use the Linux subsystem to install it. Once installed, verify that it is working by opening your terminal and typing the following commands: node -vnpm -v If you see version numbers returned for both, you have successfully installed Node.js and its package manager, npm. Install Claude Code With Node.js ready, you can now install Claude Code globally on your system. Run the following command in your terminal: npm install -g @anthropic-ai/claude-code After the installation finishes, you can launch the tool by simply typing: claude The tool will guide you through an authentication process to link the CLI to your Anthropic account. While there is a free tier for Claude, most SEOs doing heavy data lifting will prefer a paid plan or API-based access to ensure higher usage limits and faster processing. Step 2: Establishing a Reporting Framework Once Claude Code is running, you need a way to visualize the data it processes. While Claude can generate text-based summaries, the goal is to create a professional dashboard. One of the most effective ways to do this is by using an open-source tool like the Observable Framework. Observable Framework allows you to build data-rich apps and dashboards using simple code. When you combine Claude Code’s ability to write logic with Observable’s ability to render charts, you get a powerful, automated reporting engine. When you start your project, you can prompt Claude to help you set this up. For instance, you might say: “I need to build a marketing report using Google Search Console data. Please help me set up a local directory and initialize a reporting framework using Observable.” Claude will then create the necessary file structure. It is highly recommended to store these projects in a dedicated code directory (e.g., /Users/Name/Projects/SEO-Reports) rather than standard folders like Documents or Desktop. This prevents issues with cloud-syncing services like iCloud or OneDrive, which can sometimes interfere with development environments. Step 3: Connecting to Google Search Console API While you can manually export CSVs from GSC and ask Claude to read them, the real power comes from connecting directly to the Google Search Console API. This allows for real-time data retrieval and more complex historical comparisons. To do this, you must

The headlines surrounding the rise of artificial intelligence have taken a decidedly dystopian turn. If you have spent any time following tech news recently, you have likely encountered a steady stream of warnings from some of the most influential figures in the global economy. The narrative is clear: a massive shift is coming to the white-collar workforce, and it may happen much faster than many are prepared for. In April, Dan Schulman, the former CEO of PayPal and a prominent voice in fintech, issued a stark warning that AI could potentially drive U.S. unemployment to 20% or even 30% within the next two to five years. Similarly, Anthropic CEO Dario Amodei has suggested that as much as half of all entry-level white-collar jobs could be eliminated within half a decade. Even in the automotive world, Ford CEO Jim Farley has stated that AI has the potential to replace “literally half” of the white-collar workforce in the United States. For those of us in Search Engine Optimization (SEO), these projections feel personal. SEO is, by definition, a white-collar, knowledge-based profession. If the robots are coming for the analysts, the writers, and the strategists, does that mean our industry is on the brink of extinction? The answer is more nuanced than the “doom and gloom” headlines suggest. While the landscape is undeniably shifting, the reality is that SEOs have been living in a state of constant evolution for decades. We are a cohort of professionals used to wearing multiple hats: part technical architect, part content strategist, part data scientist, and part user experience researcher. While AI will certainly make “shallow” SEO obsolete, it is simultaneously creating a world where true SEO expertise is more valuable—and more necessary—than ever before. The old version of SEO stopped working years ago The “SEO is dead” trope is one of the longest-running jokes in the digital marketing world. For as long as there have been search engines, there have been pundits predicting their demise. As early as 2005, Jeremy Schoemaker published a viral article echoing Jason Calacanis’ sentiment that SEO was a dying art. In 2009, Robert Scoble declared that SEO was no longer important, prompting a now-famous rebuttal from Danny Sullivan. The reason SEO didn’t die in 2005 or 2009 is the same reason it won’t die in 2026: search is a fundamental human behavior. However, the *way* we search—and what we find—has changed fundamentally. To understand the future, we have to look at the visual history of the Search Engine Results Page (SERP). Consider a search for a high-volume head term like “flowers.” Back in 2007, a No. 1 organic ranking was the holy grail of digital marketing. In that era, the top organic result sat proudly at the top of the page, capturing the lion’s share of clicks and revenue. At the time, major brands like 1-800-Flowers could build an entire business model around maintaining that top spot. Fast forward to 2026. That same brand might still hold the No. 1 organic position, but the SERP itself has been transformed. Today, that organic listing is buried beneath a mountain of Google Ads, Shopping carousels, Local Map Packs, and AI-generated overviews. In many cases, a user has to scroll past three or four screens of “features” before they even see a traditional blue link. If your definition of SEO is simply “getting to the top of Google’s organic results” by tweaking title tags and stuffing keywords, then yes, that version of SEO has been dead for a long time. But if you define SEO as understanding the intent behind a query and meeting a user wherever they are looking for answers—whether that’s a traditional search engine, a social platform, or an AI LLM—then your role has never been more critical. Why true SEO experts are uniquely positioned to thrive There is a specific phenomenon occurring with generative AI that mirrors other creative industries. When AI video tools first launched, social media was flooded with “look what I can do” clips. Most of these were flashy but hollow. However, the videos that actually resonate and gain traction are those created by people who actually understand the craft of filmmaking. They understand pacing, lighting, sound design, and emotional resonance. They use AI as a high-powered tool to execute a professional vision. SEO is entering a similar phase. We are seeing a surge of people typing basic prompts into ChatGPT and assuming they now “know SEO.” What these individuals fail to realize is that SEO was never about just reverse-engineering an algorithm. It was about reverse-engineering human psychology. The experts who will thrive in the AI era are those who can move beyond the prompt. They are the ones who can have a “deep conversation” with an LLM—teaching it, correcting it, and providing it with the specific context it needs to produce something useful rather than something generic. In a world where everyone has access to the same AI tools, the differentiator becomes the quality of the strategy and the depth of the expertise guiding those tools. 1. Performing SEO basics with unprecedented efficiency One of the most immediate benefits of AI for the seasoned SEO is the elimination of “grunt work.” However, there is a massive gap between AI-generated “slop” and AI-assisted professional work. Generic AI copy is becoming increasingly easy to spot. It often lacks personality, relies on repetitive phrasing, and fails to tell an authentic story. However, AI is exceptionally good at tasks that require compression and formatting—such as metadata. A novice might prompt an AI to “write a title tag for this page.” An expert, however, knows that a title tag isn’t just about being “pretty.” It must account for pixel width (not just character count), brand positioning, search intent, and competitor gaps. Furthermore, an expert uses AI to generate distinct assets for different platforms: a title tag for Google, an Open Graph (OG) tag for Facebook, and a Twitter card for X. By using AI to handle the heavy lifting of

Understanding the New Era of Generative Engine Optimization The landscape of search engine optimization is undergoing its most significant transformation since the invention of the backlink. Large language models (LLMs) like ChatGPT, Gemini, and Claude are no longer just tools for generating text; they have become the primary interfaces through which users discover information, compare products, and make purchasing decisions. This shift has given rise to Generative Engine Optimization (GEO), a discipline focused on making content more “citeable” by AI models. Large language models excel at synthesizing enormous amounts of information into personalized responses to plain-language prompts. These responses draw on massive training datasets and are often enhanced with real-time internet searches using a process known as Retrieval-Augmented Generation (RAG). For brands and digital publishers, the fastest way to influence what LLMs say is to influence the content they retrieve through those searches. If an AI model cannot find your content or finds it difficult to parse, your brand effectively disappears from the conversational search results. At Evertune Research, the team used the Evertune AI marketing platform to track hundreds of brands across 250 categories and every major LLM. This massive undertaking provided clear insight into which content AI models cite most often, particularly when users ask for brand or product recommendations. The results of the study, which analyzed 25,000 unique URLs, reveal a definitive preference in the AI ecosystem: AI search loves listicles. The Data Behind the Citation Revolution To understand the mechanics of AI citations, Evertune reviewed the 6,000 most-cited URLs per model across ChatGPT, Copilot, Gemini, Google AI Mode, Google AI Overview, and Perplexity for March and April. While the total pool of analyzed citations reached 36,000, the dataset distilled down to approximately 25,000 unique URLs, as many top-performing pages were cited across multiple platforms. The findings were staggering. Of the 25,000 unique URLs reviewed, half were formatted as listicles. When looking at the broader scope of nearly 400 million citations across all models, 63% pointed to listicles. This suggests that while traditional long-form articles and deep-dive essays still have a place, the “Best of” or “Top 10” format is the current king of the AI-driven web. Listicles possess several inherent qualities that make them ideal for model consumption. First, they are tightly focused on a single topic, such as “best laptops for gamers” or “top CRM software for small businesses.” This topical density makes them highly relevant to specific user prompts. Second, their structured nature—often featuring clear headers, bullet points, and consistent formatting—makes them exceptionally easy for an AI to parse, summarize, and reproduce in a chat interface. The Comparison Advantage For brand-related queries, listicles do the heavy lifting for LLMs. Rather than the model having to scan ten different individual product pages to understand the differences between them, a single listicle provides a head-to-head comparison of features, price points, materials, and pros and cons. This structured comparison is exactly what ChatGPT now features prominently in its specialized shopping widget, which prioritizes clear, data-rich product comparisons over nebulous marketing copy. How Different AI Models Prioritize Content While the preference for listicles is a universal trend, different AI models exhibit unique behaviors in how they select and present citations. The Evertune analysis showed that listicles accounted for 40% to 65% of the most-cited URLs depending on the specific model. Gemini and the Google Ecosystem Google’s Gemini models—including Gemini, Google AI Mode, and Google AI Overviews—showed the highest reliance on listicles, sitting at the top of the range. There is also a significant amount of overlap within the Google ecosystem. More than half of the URLs cited in Google AI Mode also appeared in Google AI Overviews. This suggests that Google’s various AI implementations likely share a core index or a similar set of ranking signals that favor highly structured, authoritative list content. Copilot and Perplexity Microsoft’s Copilot sat at the lower end of the listicle spectrum, though listicles still represented a massive 40% of its citations. Interestingly, Copilot is the most “independent” of the models, sharing only 4% to 6% of its top URLs with other models. This indicates that Microsoft’s search algorithms and training data prioritize different authority signals than Google or OpenAI. Perplexity, often dubbed the “answer engine,” shares more than 20% of its URLs with Google’s models. This overlap suggests that as Perplexity crawls the web, it is identifying the same high-value, highly-structured pages that Google’s traditional and AI search engines favor. Breaking Down the Listicle Format Not all lists are created equal. The Evertune study categorized listicles into several types to see which ones the AI models preferred. The vast majority of cited listicles featured ranked lists—content like “Top 5 CRM Tools” or “10 Best Running Shoes for Marathons.” Depending on the specific AI model, ranked lists made up between 71% and 86% of all listicle citations. Ranked vs. Unranked Content Unranked lists, such as “7 Ways to Save on Groceries” or “12 Ideas for a Backyard Garden,” were a distant second. These provide value but lack the definitive “winner” or hierarchy that AI models often look for when answering a direct recommendation prompt. Institutional rankings, such as the data-heavy “Best Colleges” rankings from U.S. News & World Report, accounted for a surprisingly small portion of citations, ranging from only 1.4% to 4.7%. The Rise of Earned Media and Affiliate Domains The study also looked at the domains providing these listicles. Corporate sites, earned media (news and industry publications), and affiliate domains were the dominant sources. Forbes.com emerged as a powerhouse in this category. While Forbes is traditionally considered an earned media domain, its expansion into affiliate segments like Forbes Advisor and Forbes Vetted has made it a top-three source for listicles across every single AI model analyzed. This highlights a critical lesson for marketers: appearing in a “Best of” list on a high-authority domain like Forbes or TechRadar is often more valuable for AI visibility than having the #1 spot for a keyword on your own corporate blog. The Risks: Google

Direct Traffic & Popularity – Correlation, Not Causation via @sejournal, @TaylorDanRW The SEO industry has long been obsessed with the “secret sauce” behind Google’s ranking algorithms. For years, practitioners have debated which metrics are genuine ranking signals and which are merely indicators of a site’s overall health. A recent discussion sparked by an AI citation study has brought one of these age-old debates back to the forefront: the relationship between direct traffic, brand popularity, and search engine rankings. Specifically, the industry is once again grappling with the distinction between correlation and causation. When high-ranking websites consistently show high levels of direct traffic, it is easy to assume that Google uses that traffic as a direct ranking factor. However, as experts like Taylor Danvers have pointed out, the reality is far more nuanced. High direct traffic is often a symptom of a successful brand rather than the cause of its high search visibility. Understanding this distinction is critical for SEOs and digital marketers who want to build sustainable strategies rather than chasing phantom metrics. The AI Citation Study: A New Lens on an Old Problem The latest iteration of this debate was triggered by research into how AI search engines and Large Language Models (LLMs) choose their sources. As tools like Perplexity, ChatGPT, and Google’s Search Generative Experience (SGE) become more prominent, SEOs are desperate to understand how to earn citations within these AI-generated responses. The study in question noted a strong correlation between websites that receive significant direct traffic and those cited most frequently by AI models. At first glance, this might suggest that AI models—and by extension, traditional search engines—prioritize sites that people visit directly. The logic seems sound: if many people go directly to a website, that website must be an authority, and therefore it should be cited. However, this interpretation misses the underlying mechanism. AI models are trained on massive datasets that represent the “best of the web.” A site with high direct traffic is typically a site with a massive brand presence, extensive backlinks, and a long history of providing value. It is these foundational elements that lead to both high direct traffic and AI citations, rather than the traffic itself driving the citations. Defining Direct Traffic in the Modern SEO Era To understand why direct traffic is often misunderstood, we must first define what it actually is. In the simplest terms, direct traffic occurs when a user arrives at a website without clicking a link on another website or a search engine result page (SERP). This usually happens when a user types a URL directly into their browser, clicks a bookmark, or clicks a link in a non-web-based application like a PDF or a private messaging app. However, “Direct” traffic in Google Analytics is often a “catch-all” bucket. It includes “dark traffic” from sources where the referrer data is lost, such as: Links shared via Slack, WhatsApp, or Discord. Clicks from mobile apps (like Facebook or Twitter) that don’t pass referrer data properly. Visitors moving from an HTTPS site to an HTTP site. Users browsing in Incognito or Private mode. Because direct traffic is often a “noisy” metric, it is highly unlikely that Google would use it as a primary, weighted ranking factor. Doing so would make the algorithm vulnerable to manipulation through bot traffic and would reward sites for traffic that Google cannot fully verify. The Correlation vs. Causation Fallacy In data science and SEO, correlation means that two variables move together. Causation means that one variable directly influences the other. A classic example used to explain this is the relationship between ice cream sales and shark attacks. Both increase during the summer months. Does eating ice cream cause shark attacks? No. The hidden variable is the warm weather, which causes more people to buy ice cream and more people to swim in the ocean. In SEO, high rankings and high direct traffic are the ice cream and the shark attacks. The “warm weather” is brand authority and user satisfaction. When a brand provides an exceptional service or high-quality information, two things happen simultaneously: users bookmark the site (leading to direct traffic), and other websites link to it (leading to higher search rankings). The direct traffic doesn’t cause the ranking; the quality of the site causes both. Why Popularity Looks Like a Ranking Factor Google’s goal is to provide the most relevant and authoritative result for a user’s query. Popularity is a powerful proxy for authority. If millions of people search for “Amazon” or “The New York Times,” Google recognizes these as authoritative entities. This leads to what many call the “Brand Halo Effect.” When a brand is popular, it benefits from several signals that Google *does* explicitly track: 1. Higher Click-Through Rates (CTR) If a user sees a well-known brand in the search results alongside an unknown site, they are more likely to click the known brand. Google has confirmed through various disclosures (and the recent DOJ vs. Google trial documents) that user interaction signals, often referred to as Navboost, play a massive role in how results are re-ranked. Popularity drives clicks, and clicks drive rankings. 2. Branded Search Volume When users search for a specific brand name (e.g., “Nike running shoes” instead of just “running shoes”), it sends a clear signal to Google that the brand is a leader in its space. This increases the site’s overall “entity” strength in the Knowledge Graph, which can indirectly boost the rankings of its non-branded pages. 3. Natural Link Acquisition Popular websites are cited more often by bloggers, journalists, and researchers. A site with 50,000 direct visitors a day is much more likely to be linked to naturally than a site with 50 visitors. These backlinks are the primary currency of SEO causation. While the direct traffic itself isn’t the signal, the backlinks generated by that popularity certainly are. The Role of Navboost and User Intent The discussion around direct traffic often touches on “Navboost,” a system within Google’s infrastructure that uses click data