Reports

Google has signed the White House's "Ratepayer Protection Pledge" and unveiled a five-point approach for responsible energy growth amidst surging power demand driven by AI and other technologies. The plan focuses on ensuring that data center expansion does not burden other electricity customers. Key commitments include: 1. **Paying Its Own Way**: Committing to cover 100% of its data centers' power consumption and related infrastructure costs, utilizing a "Capacity Commitment Framework" to guarantee funding for new power needs. 2. **Bringing New Energy Online**: Pledging to add net-new energy to the grid, having already contributed over 22 gigawatts globally. It will continue investing in advanced nuclear, geothermal, and long-duration storage, and explore new rate structures like the Clean Transition Tariff. 3. **Contributing to Grid Resilience**: Investing in infrastructure modernization, such as advanced conductors that can double transmission capacity affordably. 4. **Creating Jobs**: Its data centers generate nine additional community jobs for every direct hire, surpassing industry standards. It aims to increase the electrical workforce pipeline by 70% within five years. 5. **Investing in Efficiency and Reliability**: Maintaining a leading Power Usage Effectiveness of 1.09 (vs. industry average of 1.56) and deploying Demand Response programs. **Comment**: Google's move translates corporate energy responsibility into actionable frameworks and quantifiable metrics, setting a new industry benchmark for cost-sharing, job creation, and efficiency. Its innovative mechanisms like the Capacity Commitment Framework offer a replicable model for other energy-intensive sectors to contribute fairly to grid stability and expansion.

Google has announced the general availability of its “Canvas in AI Mode” feature for all English users in the United States. This feature provides a dynamic, dedicated space for organizing plans and projects over time, with newly added support for creative writing and coding tasks. Users can now draft documents or create custom, interactive tools directly within Google Search. To start a project, users can select the new Canvas option from the tool menu (+) in AI Mode and describe what they want to create. The system will generate a working prototype in the side panel, pulling together the latest information from the web and Google's Knowledge Graph. Users can test the functionality, toggle to view the underlying code, and refine it through conversational follow-ups until it meets their needs. An example from early testers was a dashboard to visualize and track academic scholarship information, including requirements, deadlines, and amounts. The feature is designed to help users jumpstart various projects, from building custom tools to studying for exams or planning trips. <b>Comment</b> This move further evolves Google Search from an information retrieval tool into a productivity platform. By integrating code generation and real-time data fetching capabilities, Canvas lowers the barrier for non-technical users to create personalized tools, directly entering the low-code/no-code application arena. It represents a significant practice in deeply integrating AI with search and may reshape user expectations and habits regarding search engines.

Google has introduced a major update called "Cinematic Video Overviews" to its AI-powered note-taking application, NotebookLM. This feature moves beyond traditional narrated slides to generate unique and immersive personalized video content for users. The core technology integrates several of Google's advanced AI models, including Gemini 3, Nano Banana Pro, and Veo 3. These models work in concert to produce fluid animations and rich, detailed visuals, aiding users in learning and engaging with topics of interest. In this process, the Gemini model acts as a "creative director," making hundreds of structural and stylistic decisions based on the user's source materials to best tell the story. It autonomously determines the optimal narrative, visual style, and format, and continuously refines its own output to ensure consistency. The feature is now available in English for Google AI Ultra subscribers (aged 18+) on web and mobile platforms. **Comment**: This update signifies the evolution of AI-driven knowledge content creation from static/linear presentations towards dynamic, multimodal, and highly personalized immersive experiences. Positioning Gemini as the "creative director" is key, elevating AI from a content generation tool to a content architect. It is advisable to monitor the performance of its model combination (particularly Veo 3's video generation capabilities) in practical applications and whether this feature will be extended to broader user bases or enterprise/education scenarios in the future.

Google has partnered with Taiwan's National Health Insurance Administration (NHIA) to launch an innovative AI public health solution. The centerpiece is the "AI-on-DM (Artificial Intelligence on Diabetes Mellitus)" model. By digitizing clinical logic and leveraging Google Cloud's concurrency, the model reduces the average time for a single diabetes risk assessment from 20 minutes to just 25 seconds, achieving a staggering 14,400x efficiency boost. This enables the system to evaluate 20,000 people in under 90 minutes, making population-scale risk screening a scalable tool. Furthermore, NHIA will integrate a Gemini-powered health assistant into its government health app used by 10 million people in Taiwan this month. The assistant generates personalized health insights and advice based on clinical guidelines to support daily care. The collaboration framework plans to expand to manage conditions like hypertension and hyperlipidemia. To ensure equitable access, Google.org has awarded a $1 million grant to the Digital Humanitarian Association (DHA). The funding aims to bring diabetes management services and digital training to 300 community centers, supporting 240,000 health check-ins and training 200 local caregivers. <b>Comment</b>: This is a benchmark case that integrates large language models (Gemini/MedLM) with domain-specific AI models (AI-on-DM) deeply into an existing universal healthcare system. Its core value lies in using AI to automate time-consuming manual clinical logic analysis, enabling a paradigm shift from "treatment" to "prediction and prevention." The community program also attempts to address the "digital divide," providing a replicable technological and managerial blueprint for global public health systems.

Brief: Google announced on its official blog a collaboration with Taiwan to build an artificial intelligence blueprint for public health. The core objective of this project is to leverage AI technology to deliver predictive diabetes care for millions within its population-wide healthcare system. The initiative is technically founded on combining two decades of accumulated health data from Taiwan with Google's generative AI model, Gemini. By analyzing this vast historical dataset, the AI model aims to identify early risk factors and development patterns for chronic diseases like diabetes, enabling earlier intervention and personalized care plans. This effort seeks to shift the healthcare paradigm from "reactive treatment" to "proactive prediction and prevention." This collaboration is positioned as a scalable "blueprint," demonstrating how large-scale historical health data can be integrated with advanced generative AI to address broad public health challenges. Its model could potentially be adopted by other regions or for other diseases in the future. <b>Comment</b>: This project is a landmark case for AI implementation in public health, with its core value lying in the deep integration of "data + generative AI." Utilizing 20 years of real-world data for model training is crucial for improving the accuracy of chronic disease prediction. It is advisable to monitor subsequent disclosures regarding specific technical architecture (e.g., data processing methods, model fine-tuning details) and actual clinical validation results to assess its replicability and potential for other disease areas.

Google has introduced a new "share item location" feature in its Find Hub app, designed to help users recover lost checked luggage faster. This feature allows users to generate a secure URL link for Find Hub-compatible tracker tags or network accessories and share it directly with partner airlines, enabling them to view the item's updated location in real-time. Users can stop sharing at any time from the app, links automatically expire after seven days, and sharing is disabled as soon as the phone detects the item is back with the user, prioritizing privacy and user control. Google has partnered with over 10 major global airlines, including Air India, China Airlines, the Lufthansa Group, Saudia Airlines, and Turkish Airlines, which now accept Find Hub locations as part of their baggage recovery processes. Furthermore, Google has collaborated with SITA and Reunitus to integrate this technology into the industry-leading baggage-tracing systems, WorldTracer and NetTracer, covering thousands of airports worldwide. Additionally, Google is working with Samsonite to embed Find Hub technology directly into their latest suitcase designs for out-of-the-box pairing. **Comment**: This feature deeply integrates personal item tracking networks with aviation industry infrastructure, addressing a key pain point in luggage loss scenarios through ecosystem partnerships. Its technical approach (secure links, system integration) and extensive industry collaboration significantly enhance the solution's practicality and coverage, serving as a typical example of IoT technology implementation in a vertical industry (travel).

Google released its March 2026 “Pixel Drop” software update, introducing a suite of new AI-centric features and personalization enhancements for Pixel phones and watches. **Enhanced Core AI Capabilities**: The Circle to Search feature has been upgraded with multi-object image recognition, allowing users to identify multiple elements within an image (e.g., plants, clothing). A new “Try It On” feature enables virtual try-ons directly within search results. The Gemini AI assistant, now in beta, can handle everyday tasks in the background by working with other apps, such as ordering groceries or booking rideshares. The Magic Cue feature intelligently suggests restaurant recommendations within chats, allowing users to access Gemini-powered suggestions without leaving the conversation app. **Personalization and Experience Optimization**: The music recognition tool “Now Playing” has been upgraded to a standalone app for easier tracking of music history. The “At a Glance” feature now provides richer real-time information, including transit delays, sports scores, and financial portfolio updates. Users can also generate one of five AI-powered icon pack styles to create a consistent aesthetic for their home screen. **Pixel Watch Security and Convenience Upgrades**: The watch introduces new security联动 features, automatically locking the paired phone when it moves out of range and sending instant alerts if left behind. The Find Hub feature allows users to locate devices directly from their wrist. Gesture controls, first introduced on Pixel Watch 4, are now expanding to Pixel Watch 3, supporting actions like answering calls with a double pinch or wrist turn. A new express pay feature enables payments with a simple tap without first opening the Google Wallet app. For safety, standalone earthquake alerts have been added, and Satellite SOS coverage has been expanded to Canada, Europe, Alaska, and Hawaii. **Commentary**: This update deeply integrates generative AI with core device functionalities, from information search and task automation to device ecosystem联动, reflecting Google's strategy of positioning AI as a core interaction paradigm. The practical advancements in Circle to Search's “Try It On” and Gemini's background task handling are noteworthy, potentially driving mobile AI assistants towards a more proactive and seamless service model. For competitors, the experience壁垒 built through on-device AI and wearables ecosystem integration is值得关注.

Google has released a series of new feature updates for its Android platform in March 2026. Key updates include: 1. **Real-time Location Sharing**: Integration of "Find Hub" within Google Messages allows users to share live locations directly during chats with a real-time map view, with the flexibility to start or stop sharing at any time. 2. **Luggage Tracking Assistance**: Via "Find Hub", users can share the location link of luggage equipped with tracker tags with participating airlines to help recover lost bags and receive updates, maintaining full control over location sharing permissions. 3. **App Discovery Innovation**: Introduction of "shorts" short-form video feeds in Google Play, enabling users to scroll through videos (covering wellness, photography, shopping, etc.) to see apps in action and evaluate them without leaving the store. 4. **Personalized Calling Experience**: Launch of custom "Calling Card" feature, allowing users to design personalized caller IDs with chosen photos, fonts, and colors, applicable to all contacts or saved contacts only. 5. **In-car Entertainment Expansion**: New "Teacher-Approved" kids' games on Android Auto offer educational play experiences for ages 3-12, such as *Kids Games: For Toddlers 3-5* and *Disney Coloring World*. 6. **Emoji Innovation**: Updated "Emoji Kitchen" combinations in Gboard provide new sticker mash-up options. **Comment**: This update focuses on enhancing daily utility and personal expression. The deep integration of "Find Hub" strengthens scenario-based location sharing (from social meetups to luggage tracking), while the short-form video format innovates the app discovery process, potentially improving Google Play conversion rates. It is advisable to monitor the privacy control details of location data sharing and the actual impact of "Play shorts" on distribution for small-to-medium developer apps.

Google recently shared detailed usage tips for its experimental research prototype, "Project Genie," via its official blog. Project Genie is an AI tool that allows users to create, explore, and remix their own interactive worlds. Users can construct virtual worlds with characters and environments using text prompts or a combination of text and images, and navigate within them in real-time. The briefing highlights four core prompting techniques: First, describe the environment in detail, including the specific setting (e.g., forest, city), detailed characteristics (e.g., lush or wintry), objects/structures, and weather/atmosphere, with the option to set a photo-realistic or cartoon/video game style. Second, freely choose a navigation character, which can be any entity (e.g., a tiny blue giraffe or a giant pixelated doll), and define its movement and special effects (e.g., emitting smoke when hopping). Third, support for world creation starting from the user's own images, requiring the character to be centered with sufficient background to define the environment. Finally, it recommends using short, direct, action-oriented language for prompts and leveraging the Gemini app for assistance in optimization. After creation, users can preview the world and character via Nano Banana Pro and make real-time adjustments. Additionally, users can switch between first-person and third-person perspectives for exploration. Currently, Project Genie is only available to Google AI Ultra Subscribers in the U.S. over the age of 18, with plans for further expansion. **Commentary**: Project Genie demonstrates Google's cutting-edge exploration in the fusion of generative AI and interactive content creation. Its core value lies in lowering the technical barrier to building high-quality virtual worlds, enabling non-professional users to quickly create immersive experiences through intuitive prompts and image inputs. This provides a prototype for new creative tools in future scenarios such as game development, virtual socializing, and education/training. It is advisable to monitor its subsequent rollout plans and its potential for ecosystem integration with other Google AI products like Gemini.

Google has officially launched the Gemini 3.1 Flash-Lite model, the fastest and most cost-effective model in the Gemini 3 series, designed to meet developers' large-scale, high-frequency workload demands. The model is now available in preview via the Gemini API in Google AI Studio (for developers) and Vertex AI (for enterprises). In terms of cost and performance, Gemini 3.1 Flash-Lite is competitively priced at $0.25 per million input tokens and $1.50 per million output tokens. According to the Artificial Analysis benchmark, its performance significantly surpasses the previous-generation Gemini 2.5 Flash, achieving a 2.5X faster Time to First Answer Token and a 45% increase in output speed while maintaining similar or better quality. The model scored an impressive 1432 Elo on the Arena.ai Leaderboard and outperformed other models of similar tier on reasoning and multimodal understanding benchmarks, including 86.9% on GPQA Diamond and 76.8% on MMMU Pro, even surpassing the larger previous-generation Gemini 2.5 Flash model. The model comes standard with "thinking levels" in AI Studio and Vertex AI, allowing developers to flexibly adjust the model's "thinking" depth based on task requirements, which is critical for managing high-frequency workflows. Its application scenarios are broad, including large-scale translation, content moderation, user interface generation, creating simulations, and following complex instructions. Companies like Latitude, Cartwheel, and Whering have already used the model as early testers to solve complex problems at scale. <b>Comment</b>: The launch of Gemini 3.1 Flash-Lite marks a key step for Google in balancing large model performance and cost. Its extremely low latency and affordable pricing make it an ideal choice for building real-time, responsive AI applications, especially in high-throughput scenarios. For enterprises and developers seeking to deploy AI capabilities at scale while strictly controlling costs, this is an option worth evaluating closely.

Google has announced the general availability of its VRC Non-Skip ads globally. This product is designed to help brands reach the massive audience watching YouTube on connected TVs (CTV) in living rooms. YouTube has been the #1 streaming service in the U.S. for three consecutive years, providing a substantial big-screen audience base for advertisers. The core of this launch is the deep integration of Google's AI technology. VRC Non-Skip ads leverage Google AI for dynamic optimization, intelligently selecting and allocating between 6-second Bumper ads, 15-second standard non-skippable ads, and a 30-second CTV-only non-skippable ad format. This AI-powered optimization aims to ensure campaigns reach the right audience at the right time, thereby improving overall marketing efficiency. Compared to manually mixing single-format campaigns, Google emphasizes that its AI-driven precision can deliver greater efficiency and more unique reach and impact. The solution is specifically optimized for the CTV big-screen environment, ensuring brand messages are delivered in their entirety. <b>Comment</b>: This move signifies Google's systematic effort to capture the fast-growing big-screen advertising market using its AI advantage. By dynamically optimizing multi-format ad mixes with AI, Google offers advertisers a smarter and more efficient CTV advertising solution, positioning it as a core tool for brands targeting living room audiences. It is advisable to monitor subsequent performance data releases to evaluate its claimed efficiency gains.

Google has released its latest feature update, the "Gemini Drop," for the Gemini app in February 2026. This update focuses on enhancing AI's creativity and problem-solving capabilities. Regarding creative tools, Google launched its most advanced music model, Lyria 3 (in beta), allowing users to quickly generate 30-second custom soundtracks using text or image prompts. Concurrently, the image model has been upgraded to Nano Banana 2, which can generate images with higher fidelity and "incredible speed," supports adding text in any language with "real-world accuracy." The video creation tool Veo now offers a template gallery where users can browse, select a style, and remix it with their own details to quickly produce polished, personalized videos. In terms of core AI capabilities, Google released Gemini 3.1, which features significantly improved intelligence, particularly excelling at solving complex problems. This version includes Gemini 3.1 Pro for demanding workflows and a specialized reasoning mode called "Deep Think" designed for modern science and engineering, currently available only to Google AI Ultra subscribers. Furthermore, Gemini now provides direct links to verified scientific paper citations, helping users quickly access high-quality research data. **Comment**: This update indicates Google is deepening its focus from general AI conversation to verticalized, professional creation and productivity tools. The launch of Lyria 3 and Nano Banana 2 signifies its technological integration and experience optimization in the AIGC multimodal generation (audio, image, video) field. The introduction of Gemini 3.1, especially the "Deep Think" mode, aims to compete for the high-end scientific research and engineering market, forming a differentiated competition against rivals like OpenAI. It is advisable to monitor the actual performance of its professional reasoning mode and the impact of its subscription strategy on the market.

Google has introduced a time-limited new feature within the Gemini app, enabling users to create and share highly personalized musical greeting cards for the 2026 Lunar "Year of the Fire Horse." This feature leverages the newly released Lyria 3 model to transform user inputs into a 30-second high-fidelity audio track accompanied by custom cover art. Users simply provide prompts within the Gemini app, including the recipient's name, a personal message, hobbies, and music genre preference (e.g., Rock, Ballad, Chinese Classical, R&B, Jazz, Mandopop-Trap, or Luk Thung). The system then synthesizes original lyrics, composes the music, and generates cover art featuring traditional red and gold motifs with dynamic horse imagery. The finished personalized greeting can be exported with one tap to major messaging platforms like WhatsApp, WeChat, KakaoTalk, and iMessage. The feature is primarily accessible via a time-limited discovery banner in the app for users in Singapore, Thailand, Malaysia, Indonesia, South Korea, Vietnam, the Philippines, Taiwan, Brunei, and Mongolia until March 3rd (coinciding with the Lantern Festival). Users in other regions can manually initiate the experience by copying and pasting a specific prompt. **Comment**: This is a precise marketing and productization effort that combines a cutting-edge AI audio model (Lyria 3) with a specific cultural festival (Lunar New Year). It demonstrates the maturity of AI in personalized content generation (music + visuals) and employs a "time-limited + regional" launch strategy to test market response and create a sense of exclusivity. For vendors focused on AI application deployment, this "lightweight, scenario-based" functional innovation model is worth noting.

Google Arts & Culture has launched a new online digital hub called "Canvas Legends" in deep collaboration with Berlin's Gemäldegalerie. This platform represents Google Arts & Culture's largest single-museum art digitization project to date, having digitized over 1,100 masterpieces from the collection in ultra-high resolution, resulting in more than 1,100 Gigapixel images. Technically, the project utilizes Google's "Art Camera" technology, allowing users to explore painting details with "digital magnifying glass" precision online, such as the fine brushstrokes in a Vermeer or cracks in the varnish, details often missed by the naked eye during in-person visits. The platform also integrates AI technology through the "Mice in the Museum" interactive experience. Leveraging the Google Gemini model, it generates playful narrative dialogues between two fictional mice commenting on the artwork, based on academic metadata and image analysis of the paintings, offering users a novel interpretive perspective. Furthermore, the platform's content is curated around the theme of "life stages" and offers over 50 online exhibitions that delve into the motifs and stories behind the works, aiming to demonstrate the continued relevance of classical art today. **Comment**: This project is a representative case of combining cultural institution digitization with AI application. Its core value lies in using technology (Gigapixel imaging and generative AI storytelling) not only to achieve the permanent preservation and global accessibility of cultural heritage but, more importantly, to create entirely new, interactive ways of engaging with and interpreting art. This enhances public participation depth and趣味性, providing a reference solution path for the digital transformation of other cultural institutions.

Google has introduced new AI-powered features in Google Translate, driven by the Gemini model, designed to enhance contextual accuracy and expressive richness in translations. The core objective of this update is to help users precisely capture the tone of conversations across various scenarios, from informal hangouts to professional meetings. The new feature offers "helpful alternatives" for translations, particularly excelling at handling idioms and colloquial phrases. For instance, when translating "It's raining cats and dogs," the system provides not only a literal translation but also different expression options accompanied by tips on when and why to use them, aiding users in selecting the most appropriate phrasing. To explore linguistic nuances in depth, users can tap "understand" for an overview or use the "ask" function to follow up with questions about specific scenarios, such as expressions used in a particular country or dialect. This new experience is now available on the Translate mobile app (Android and iOS) in the U.S. and India, with a web version coming soon. **Comment**: This update signifies a shift in machine translation from "literal correspondence" to "contextual understanding and expression adaptation." Its core value lies in leveraging the multilingual capabilities of the large language model (Gemini) to address the long-standing pain point of "cultural adaptation" in translation, offering practical benefits for business communication and cross-cultural exchange. It is advisable to monitor the expansion of its supported dialect range and its handling of terminology in specialized fields.

Google DeepMind has launched its latest image generation and editing model, Nano Banana 2 (Gemini 3.1 Flash Image). The model is designed to deliver high-fidelity image generation and faster advanced editing, enabling developers to deploy sophisticated visual creation at scale via the Gemini API and Google AI Studio, with an emphasis on excellent price-performance ratio. Key technical enhancements include: 1. **Improved World Knowledge**: Leverages the Gemini model's extensive knowledge base combined with web image search to generate more detailed depictions inspired by real-life references. 2. **Advanced Text Rendering and Localization**: Provides more reliable text rendering compared to previous Flash image models and supports generating or translating text across multiple languages directly within the image. 3. **Greater Creative Control and Consistency**: Adds native support for extreme aspect ratios like 4:1, 1:4, 8:1, and 1:8; introduces a new 512px resolution tier to optimize efficiency and minimize latency; improves adherence to complex, multi-layered developer prompts; and introduces configurable "thinking levels" (Minimal/High/Dynamic), allowing the model to reason through complex prompts before rendering, significantly improving output quality and prompt adherence. The model is now available via the Gemini API in Google AI Studio and for enterprise deployment on Vertex AI, as well as in Google Antigravity and Firebase. **Commentary**: The launch of Nano Banana 2 marks a significant step in Google's productization and refinement within the image generation AI space. Its core highlights, such as "configurable thinking levels" and "improved instruction following," shift more control over the generation process to developers for complex, professional use cases. The addition of extreme aspect ratios and a low-resolution option demonstrates targeted optimization for the efficiency of different production pipelines (e.g., ad banners, rapid iteration). This represents not just a performance boost but a strategic shift towards deeper integration into developer workflows.

Google DeepMind has officially launched its latest image generation model, Nano Banana 2 (Gemini 3.1 Flash Image). This model aims to combine the advanced capabilities of its predecessor, Nano Banana Pro, with the generation speed of the Gemini Flash model. Technically, Nano Banana 2 features several core upgrades. It integrates Gemini's real-time world knowledge base for more accurate rendering of specific subjects and supports the creation of infographics and converting notes into diagrams. The model demonstrates precise text rendering and translation, suitable for marketing mockups and greeting cards. In terms of creative control, it supports maintaining the resemblance of up to five characters and the fidelity of up to 14 objects within a single workflow, facilitating storyboard creation. The model's ability to follow complex instructions has been enhanced, and it can generate assets in various aspect ratios and resolutions from 512px to 4K. The model is now being widely deployed across the Google ecosystem, including the Gemini app (replacing Nano Banana Pro in Fast, Thinking, and Pro models), AI Mode and Lens in Search, AI Studio and API preview, Vertex AI, Flow, and Google Ads. **Commentary**: The launch of Nano Banana 2 marks a key step for Google in balancing the "quality" and "speed" of AI image generation. By bringing professional-grade features to a high-speed model, it is expected to significantly improve iteration efficiency in creative workflows. Its broad ecosystem integration strategy will rapidly expand user reach, posing direct competitive pressure. It is advisable to monitor its actual performance and user feedback in specific application scenarios such as advertising and design.

Google has announced that it will introduce the "Gemini advantage" to its Google Marketing Platform at the Google NewFront event on March 23. This initiative aims to provide advertisers with a unified, AI-powered marketing platform through its latest Gemini AI model ecosystem, driving marketing transformation. The core enhancement lies in leveraging Gemini models to boost platform capabilities. The solution is designed to allow advertisers to gain deeper insights into the consumer journey, remove data fragmentation, and activate data in real-time. Google claims this is a "transformational shift" with proven success. Leading brands will share how the platform helps them uncover untapped value. Google positions this launch as "Our best AI, for your best ROI," emphasizing the direct link between its AI capabilities and business outcomes. The event will be livestreamed. <b>Comment</b>: This move signals Google's acceleration in deeply integrating its most advanced Gemini large models into its core commercial product suite. It aims to address pain points like marketing data silos and decision-making latency through a unified AI platform, directly targeting "high-value results" and "ROI" to strengthen its Marketing Cloud competitiveness. Advertisers should focus on its specific integration features and practical efficacy validation.

Google has announced the global expansion of beta access to its "text guidelines" feature within the AI Max advertising platform, effective immediately. This functionality allows advertisers to guide AI-generated creatives using natural language instructions to ensure brand compliance. Within AI Max for Search and Performance Max campaigns, advertisers can now define specific terms to exclude or concepts to avoid, such as "don't imply our products are cheap" or "don't use language like 'only for'." This integrates AI's creative generation with advertisers' unique insights, ensuring ad content matches user intent while strictly adhering to brand voice. Google states the feature now offers full language and vertical industry support. Brands like BYD are already scaling creatives using these controls. They reportedly increased leads by 24% at a 26% lower cost while using text guidelines to safeguard their brand standards. Google emphasizes that high-quality creatives drive performance and is exploring more ways for users to guide AI using everyday language. <b>Comment</b>: This feature represents a significant iteration in the application of generative AI in advertising, shifting from pure content generation to "controlled generation." It directly addresses core enterprise concerns regarding the brand safety and consistency of AI output, embedding brand governance capabilities into the creative production phase. This is likely to enhance the usability and trust in AI-generated advertisements.

Google announced on its official blog an early preview of a new multi-step task automation feature in the Gemini app for Android. This feature aims to free users from tedious, repetitive daily tasks. Scheduled as a beta release, the feature will initially launch in the U.S. and Korea, compatible with the Pixel 10, Pixel 10 Pro, and Samsung Galaxy S26 series devices. Users can simply long-press the power button to activate Gemini and give voice commands, such as "help book me a ride home" or "reorder my last meal on DoorDash." Gemini will execute the task seamlessly in the background, allowing users to continue using their phones. Technically, Gemini automates tasks by running the required app within a secure, virtual window on the phone. This means it can only access a limited set of authorized apps and not the rest of the device, ensuring privacy and security. The feature emphasizes user control: automations start with a user command and stop upon task completion; users can monitor progress via live notifications and can view, jump in, or stop the task at any time. Currently, the beta will support select apps in categories like food, grocery, and rideshare. Google states this is just the beginning and looks forward to user feedback.

Google has announced a major update to its "Circle to Search" feature, introducing multi-object image search capabilities. Users can now circle multiple objects within a single image for simultaneous identification and querying. For example, circling all fish in a photo allows the system to identify each species, explain their ecological relationships, and provide further web links. This feature is particularly impactful for fashion shopping. Users can circle an entire outfit from social media, and the system will automatically deconstruct and identify each clothing item, shoes, and accessories, offering links to similar products for purchase. Furthermore, in countries where virtual try-on is already available, users can now enter a virtual dressing room directly from "Circle to Search" results on the Samsung Galaxy S26 series and Pixel 10 devices, enabling "see it, try it" functionality. The core driver of this upgrade is the agentic planning and reasoning capabilities of the Gemini 3 model. It employs a visual query fan-out technique, automatically identifying key parts of an image, running multiple searches in parallel, and cross-referencing results to compile a final response. The feature launches today on the Samsung Galaxy S26 series and Pixel 10 devices, with plans to expand to more Android devices soon. **Comment**: This upgrade transforms "Circle to Search" from a single-point identification tool into a contextual exploration and shopping assistant, significantly enhancing its interactive depth and commercial potential. By leveraging Gemini 3's complex task planning, Google is blurring the lines between image recognition, information retrieval, and e-commerce discovery, opening new pathways for mobile search and discovery-based shopping. It is advisable to monitor its subsequent impact on user engagement, ad monetization, and e-commerce conversion rates.

At the 2026 Galaxy Unpacked event, Google announced a collaboration with Samsung to bring a series of new Android features powered by Google AI to the upcoming Galaxy S26 series. These updates signify Android's evolution from an operating system into an "intelligent system." The core updates comprise three main AI features. First, the Gemini assistant will introduce a beta feature allowing users to delegate multi-step tasks (like ordering a ride, food, or groceries) to Gemini running in the background via a long press of the side button, while users retain full use of their phone. Initially available in the US and Korea, it will support select apps in food, grocery, and rideshare categories. Second, the "Circle to Search" feature is enhanced with multi-object image recognition, enabling it to identify an entire outfit or decor set on screen at once and offering virtual try-on capabilities. Finally, Google's "advanced Scam Detection" is integrated directly into the Samsung Phone app on Galaxy S26 devices. Leveraging the on-device Gemini model, it can detect potential scams during a call and provide instant audio and haptic alerts, with all analysis performed locally for privacy. Powered by the next-generation Gemini 3 series models, these features aim to understand user intent and proactively suggest next steps, marking a new beginning for mobile intelligence. <b>Comment</b>: This collaboration deeply integrates Google's cutting-edge AI capabilities into Samsung's flagship hardware. The combination of the on-device Gemini model with privacy protection and multi-task delegation showcases a new paradigm of "cloud-device collaborative, device-centric" AI implementation, potentially reshaping interaction and security standards for premium smartphones.

Google has recently introduced Project Genie, an experimental research prototype powered by "world model" technology. Unlike large language models that predict the next word, a world model simulates an environment and predicts what happens next based on a sequence of actions taken by a user (or agent). It reacts to visual observations (like images) and text prompts, simulating environmental dynamics end-to-end—such as light reflection and object collisions—without relying on a backend game engine. Project Genie is currently available to Google AI Ultra subscribers in the U.S. over 18. Users can create and explore dynamic, interactive scenes (e.g., alien planets or underwater environments) by uploading images (like those generated by Nano Banana) accompanied by text descriptions. The Google team highlights several potential applications for world model technology: 1) Providing a safe, low-cost simulated training ground for AI agents to learn real-world skills; 2) Transforming education and vocational training, enabling immersive experiences like exploring ancient Rome or simulating a firefighter's role; 3) Assisting game development and filmmaking by exploring new creative environments, potentially leading to a new medium that blends viewing and interactive gameplay. <b>Comment</b>: Project Genie represents an evolution in AI from static content generation to dynamic, interactive environment simulation. The "world model" concept opens new avenues for AI training, educational technology, and content creation. However, as an early-stage prototype, its realism, complexity, and scalability for broader applications remain key areas to watch for future development.

Google announced a major update to its AI creation platform, Flow, in February 2026, aiming to provide a seamless, unified workflow for creating and refining content. The core of this update is the deep integration of image generation capabilities into the video creation process. **Key Feature Upgrades**: 1. **Multimodal Creation Integration**: Google is migrating the best features from its image generation experiments, Whisk and ImageFX, directly into Flow. Users can now generate, edit images, and use them directly as ingredients or keyframes for Veo video generation within a single workspace without switching tools. The Nano Banana technology is now fully built-in, supporting high-fidelity image generation. 2. **Flexible Asset Management**: A new asset grid view is introduced, supporting search, filtering, sorting of images and videos, and grouping related assets into "Collections". Users can quickly reference specific assets in their library using the "@" symbol. 3. **Precision Editing Controls**: A new lasso tool allows users to precisely select areas of an image and make edits using natural language instructions (e.g., "remove the man" or "add Koi fish in the water"). It also supports seamlessly extending clip length, adding/removing objects in videos, and orchestrating precise camera movements like pans and zooms. **Migration & Cost**: For early users' convenience, starting March 2026, users can opt to transfer all projects and assets from Whisk and ImageFX directly into their Flow library. Additionally, generating images in Flow will be free. **Commentary**: This update signals Google's effort to consolidate its disparate AI generation tools (image, video) into a powerful, coherent creation platform. By breaking down barriers between tools and offering granular editing control, Google aims to attract professional creatives from concept design to short film production, strengthening its competitiveness in the AIGC creation tool market. It is advisable to monitor the efficiency gains of its "image-as-video-ingredient" workflow in practical applications.

Google has introduced a novel "agent step" in its Opal platform, upgrading static workflows to dynamic, interactive intelligent experiences. The core of this new feature is that the agent step understands the user's objective and autonomously determines the best path, tools, and models needed to complete the task, such as invoking web search for research or using Veo for video generation, thereby reducing manual configuration. Key technical enhancements include: 1. **Memory**: Opal can now remember user information (e.g., name, preferences) across sessions, making experiences more personalized and continuous. 2. **Dynamic Routing**: Users can define multiple execution paths based on custom logic, and the agent intelligently transitions to the correct step when conditions are met. 3. **Interactive Chat**: The agent can proactively initiate conversations, asking users for missing information or offering choices, enabling more natural collaboration. This upgrade marks a shift for Opal from predefined, rigid model calls to dynamic narratives driven by real-time creative decisions. For instance, in Opals like the "Visual Storyteller" or "Room Styler," the agent can autonomously suggest plot points or refine design proposals through iterative dialogue, making it function more like a collaborative partner than a one-way tool. **Comment**: This move significantly enhances Opal's intelligence and usability by seamlessly embedding agent capabilities into workflow steps, striking a good balance between automation and user control. It provides a new paradigm for building more complex and personalized AI applications, worthy of developers' attention to explore its potential in new scenarios.

Google has released a new set of best practice guidelines for its Demand Gen advertising campaigns on its official blog. According to Google's internal data analysis, advertisers who adopted at least 3 out of the 4 best practices saw an average increase of over 40% in conversions. The core of this solution revolves around four key practices: 1. **Audiences**: Leverage Google AI to find potential customers through optimized targeting, lookalike audiences, and new customer acquisition goals. 2. **Bid and Budget**: Drive efficiency using tCPA or tROAS bidding strategies and set adequate budgets with the Performance Planner tool. 3. **Creative**: Achieve strong performance by attaining an "Excellent" Ad Strength rating and using Google AI to produce a comprehensive set of high-quality assets. 4. **Data Strength**: Implement sitewide tagging via the Google tag gateway and connect offline data sources through Data Manager to ensure proper setup. A case study shows that Cropp achieved a 50% uplift in Return On Ad Spend (ROAS) for online sales by applying "Excellent" ad strength and recommended audience solutions. **Comment**: Google has systematically packaged its AI, automated bidding, and data management capabilities into a quantifiable "best practice" solution, aiming to lower the barrier to entry for advertisers and increase platform stickiness. The provided 40% average conversion lift data is compelling. It is recommended that digital marketing teams use this framework to audit and prioritize optimization of their audience targeting and data integration setups.