8 Must-Know Google AI Updates for May 2025 (Expert Guide)
The Google AI updates May 2025 aren’t just changing search – they’re completely reshaping how users find information online.
Traditional search results have seen click-through rates drop from 32% to 16% as AI features take center stage in the search experience.
We’ve watched AI Overviews grow at unprecedented rates since March. Entertainment queries now trigger 528% more AI Overviews.
Restaurant queries jumped by 387%, with travel queries close behind at 381%. Google’s rollout of these features coincides with their new Ironwood TPU – the most energy-efficient processing unit they’ve created to date.
College students throughout the U.S. now have free access to Gemini Advanced and NotebookLM Plus through spring 2026, fundamentally shifting how research and education work with AI tools.
For businesses appearing as sources in Google’s AI search results, the benefits are clear. One website saw a 658% increase in monthly visitors after focusing on quality backlinks.
These eight key updates aren’t minor technical tweaks – they’re reshaping the entire digital landscape. We’ll examine what each one does and how they impact your online presence.
Gemini 2.5 Pro Expansion
Image Source: Blankboard Studio
Google’s flagship AI model got a major upgrade with Gemini 2.5 Pro. This powerful addition pushes AI forward with impressive reasoning abilities and extended context handling that outperforms many competitors.
Gemini 2.5 Pro key features
Gemini 2.5 Pro stands out with enhanced reasoning capabilities, topping key industry benchmarks.
The model achieves exceptional video understanding, scoring 84.8% on the VideoMME benchmark.
It leads on math and science benchmarks like GPQA (84%) and AIME 2025 (86.7%) without requiring costly test-time techniques.
One of its most impressive capabilities is the massive context window—currently 1 million tokens with an expansion to 2 million tokens coming soon.
This lets the model process enormous amounts of information, equivalent to reading the entire “Lord of The Rings” series in a single prompt.
The model excels at coding tasks, ranking #1 on the WebDev Arena leaderboard for building attractive and functional web applications.
It shows superior performance on SWE-Bench Verified, the industry standard for agentic code evaluations, scoring 63.8% with a custom agent setup.
Notably, Gemini 2.5 Pro is natively multimodal, able to process and understand:
- Text and code
- Images (up to 3,000 per prompt)
- Video (approximately 45 minutes with audio, 1 hour without)
- Audio (up to 8.4 hours per prompt)
Gemini 2.5 Pro pricing
Google offers both free and paid access with different limitations and pricing structures:
For prompts up to 200,000 tokens:
- Input pricing: $1.25 per million tokens
- Output pricing: $10.00 per million tokens (including thinking tokens)
For prompts exceeding 200,000 tokens:
- Input pricing: $2.50 per million tokens
- Output pricing: $15.00 per million tokens
This makes Gemini 2.5 Pro more expensive than some competitor models like OpenAI’s o3-mini ($1.10/$4.40 per million input/output tokens) but more affordable than others such as OpenAI’s GPT-4.5 ($75.00/$150.00).
The free tier uses the model ID “gemini-2.5-pro-exp-03-25” with limits of 5 requests per minute and 25 per day. Unlike the paid version, data from the free tier “is used to improve our products”.
Gemini 2.5 Pro best use cases
We see Gemini 2.5 Pro shining in several key application areas:
Advanced coding and web development:
The model excels at creating visually compelling web applications, agentic code applications, and performing code transformation and editing. Developers can quickly turn concepts into functional, attractive web apps with features like responsive design and subtle UI effects.
Video understanding and transformation:
The model analyzes video content and generates interactive applications based on that understanding. It can transform educational videos into interactive learning tools or create p5.js animations representing video content.
Complex problem solving:
With its reasoning capabilities, Gemini 2.5 Pro handles tasks requiring logical reasoning, particularly in mathematics and science. It scored 18.8% on Humanity’s Last Exam, a benchmark designed by subject matter experts to test frontier knowledge.
Multimodal content summarization:
The model effectively processes and summarizes information across text, images, audio, and video, making it valuable for research and content creation.
Agentic workflows:
Its advanced reasoning, function calling, and tool use capabilities make it ideal for creating sophisticated agentic systems that can automate complex multi-step tasks.
Gemini 2.5 Pro is available through Google AI Studio and the Gemini API, with enterprise customers able to access it via Vertex AI. For individuals, it’s accessible through the Gemini Advanced subscription at $20 per month.
Multimodal Search in AI Mode
Image Source: Rampiq
Google’s May 2025 update brings visual intelligence to search with multimodal capabilities in AI Mode. Users now search with images and text together – a fundamental shift from traditional keyword-only queries.
Multimodal Search capabilities
AI Mode combines Google Lens with a custom Gemini version to create a search system that truly sees and understands visual content.
The technology doesn’t just identify objects – it analyzes entire scenes, recognizing materials, colors, shapes, and the relationships between elements.
This visual intelligence works through what Google calls “query fan-out,” where multiple questions are asked about the complete image and specific objects within it.
The system processes various inputs, including text, images, audio, and video, transforming them into different output formats.
What makes this technology truly powerful is its contextual understanding. When you upload an image and ask a question about it, you get nuanced, contextually relevant responses with links for deeper exploration.
For online sellers, this means uploading product images to receive suggestions for item names, descriptions, and pricing based on similar listings.
Multimodal Search user experience
Searching feels different with these new capabilities. Instead of typing keywords, you snap photos or upload images and ask complex questions about what you see.
Your bookshelf photo might return detailed information about each title with similar recommendations.
Users are embracing this new way of searching – AI Mode queries average twice the length of traditional search queries.
People use it for open-ended questions and complex tasks like comparing products or planning trips.
The system provides comprehensive responses with links to learn more.
When confidence in quality is low, Google falls back to traditional web search results instead of AI-generated responses.
Multimodal Search impact on SEO
This shift fundamentally changes SEO strategies.
AI Overviews have expanded dramatically across specific industries – growing by 528% for entertainment queries, 387% for restaurant queries, and 381% for travel queries between March 13-27, 2025.
Traditional keyword optimization isn’t enough anymore. Effective SEO now requires:
- Multi-format content optimization across text, images, videos, and audio to match how users prefer to consume information
- Visual search optimization as tools like Google Lens become central to the search experience
- Entity-driven content that focuses on context and detailed descriptions rather than keyword density
- Core Web Vitals improvement since visual stability and loading speed directly impact engagement
Responsive design isn’t optional – pages must display correctly on any device. Keywords continue evolving into entities, where quality comes from context rather than density.
While multimodal search represents a technological leap, it creates new challenges.
Many user journeys now follow a simplified pattern: query → search → answer, potentially reducing website visits as AI-generated responses appear directly in results.
Websites must ensure perfect crawlability to maximize chances of being selected as sources for these AI-generated responses.
AI Overviews Surge in Travel, Food, and Entertainment
Image Source: Search Engine Land
Google’s AI Overviews have hit unprecedented growth across key industries during the March 2025 core update. This surge changes search visibility for thousands of businesses.
We’re seeing dramatic shifts in how users find and interact with content in travel, food, and entertainment sectors.
AI Overviews expansion data
Between March 13-27, AI Overviews grew at remarkable rates across three key industries.
Entertainment queries jumped by 528%, with 76% of new keywords focusing specifically on movies (such as “Jennifer Love Hewitt movies” and “Bruce Dern movies”). Travel queries rose by 381%, with an overwhelming 93.78% of new travel AI Overviews targeting location-specific activities. Restaurant queries followed closely, increasing by 387% during this same period.
This shift is especially significant because these industries historically triggered AI Overviews less frequently than healthcare and education.
Just months ago, Google’s AI Overviews appeared for only about 2-5% of travel-related queries. By the end of October 2024, that number reached 30% – a 700% increase in just one month.
The “Things to do” trend now dominates search, creating full AI-generated destination guides for queries like “things to do in Buffalo, NY” or “things to do in Providence this weekend”.
AI Overviews’ impact on industries
For the travel sector, AI Overviews now create complete trip planning experiences beyond basic information.
Starting in May 2025, users will receive entire country or regional itineraries by typing prompts like “create an itinerary for Costa Rica with a focus on nature”. These AI-generated guides include:
- Day-by-day itineraries with activities and restaurant recommendations
- Photos and reviews from other users
- Expandable maps with key locations
- Export options to Docs, Gmail, or saved as custom lists in Google Maps
Restaurants feel similar impacts. AI Overviews now function as modern replacements for traditional business listing sites, showing menus, pricing, location details, and customer reviews.
We’re seeing 87% of UAE restaurant operators and 79% of US operators now using AI technologies for marketing, data analytics, and customer service.
The entertainment industry sees AI Overviews primarily in movie-related queries, with recommendation systems refining content suggestions based on user preferences. These AI tools analyze vast data sets to create tailored content for specific audiences, similar to how Netflix uses machine learning algorithms for recommendations.
AI Overviews optimization tips
To maximize visibility in this new landscape, focus on specific optimization strategies.
First, optimize for featured snippets and question-based keywords – these are the strongest indicators for triggering AI Overviews. For travel, education, and entertainment sectors specifically:
- Create decision-supporting content – Develop comprehensive guides around activities and ideas for “things to do” in specific locations [USER GUIDELINES]
- Track and refresh seasonal content – Update content regularly to match shifting seasonal interests and trends [USER GUIDELINES]
- Optimize for entity-driven searches – Especially in entertainment or around prominent people in your organization [USER GUIDELINES]
- Include citations and statistics – Content with authoritative tone, quotes, and statistics appears more frequently in large-language model outputs
Remember that AI Overviews adapt to specific search queries with various formats including mini informational articles, step-by-step guides, comparison tables, and interactive results.
Structure your content to align with these formats and user intent to increase your chances of appearing as a source in these increasingly common AI-generated responses.
Gemini Live with Camera and Screen Sharing
Image Source: The Decoder
Gemini Live with camera and screen sharing stands out as one of Google’s most significant AI updates for May 2025.
This feature moves AI conversations beyond text-only exchanges into dynamic visual interactions where Gemini sees and interprets your surroundings in real-time.
Gemini Live features
Gemini Live takes AI interaction to a new level with comprehensive multimodal capabilities. The system supports natural conversations in over 45 different languages. What makes this update powerful is its real-time visual input options:
- Camera sharing lets you point your phone at objects or locations for immediate feedback
- Screen sharing allows Gemini to analyze anything displayed on your device
- File and image uploads provide detailed explanations of documents and visual content
Gemini Live adapts to how you speak, handling interruptions smoothly and switching topics as needed. Users can move between front camera, rear camera, or screen sharing during a single conversation.
Gemini Live supported devices
The camera and screen sharing features are available across several platforms:
Initially launched on Pixel 9 series and Samsung Galaxy S25 devices, access has expanded to all Android devices with at least 2 GB of RAM running Android 10 or newer. For other devices, a Gemini Advanced subscription was first required, though Google is now making the feature available to more users at no cost.
To use Gemini Live, long-press the power button to open Gemini, then tap the “Live” icon at the bottom of your screen. For screen sharing, navigate to the screen you want to share, long-press the power button, and select “Share screen with Live”.
Gemini Live use cases
Gemini Live works well in many everyday situations:
Space organization and home improvement – Point your camera at cluttered spaces for real-time organization suggestions, including what to keep and what to donate. When renovating, get paint color advice or step-by-step guidance for fixing household items.
Creative brainstorming – Share your screen to analyze images of textures, colors, or environments that spark design ideas, writing concepts, or craft projects. Content creators find this valuable for inspiration.
Shopping assistance – When browsing online stores, screen sharing turns Gemini into a personal shopping assistant that compares products, offers style advice, or suggests items that complement your existing wardrobe.
Learning and skill development – Get feedback on creative work by sharing screens with blog posts, social media campaigns, or photo collections. Students can take pictures of textbooks for help understanding complex topics.
Gemini Live with camera and screen sharing shows how Google’s May 2025 AI updates are moving beyond traditional AI interactions. By using cameras and screens as inputs, Gemini becomes an AI assistant that truly sees and understands your physical world.
AI for Education: Free Tools for Students
Image Source: Google One
College students across the U.S. now get free access to Google’s cutting-edge AI tools. This May 2025 update removes financial barriers for students wanting to use AI for their academic work.
Google AI tools for students
The Google One AI Premium plan for students includes several powerful tools:
- Gemini Advanced, powered by Gemini 2.5 Pr,o with Deep Research features that explore complex topics and create readable reports
- NotebookLM Plus offering five times more Audio Overviews for learning on-the-go
- Gemini integration in Google Docs, Sheets, and Slides to help with writing, data analysis, and presentations
- Veo 2 for creating detailed eight-second video clips from text descriptions
- Whisk from Google Labs that turns text prompts into animated images
Google also offers “Help me read” on Chromebook Plus devices to help students better understand content in essays, PDFs, and research materials.
Eligibility and pricing
This offer is only available to U.S. college students with verified .edu email addresses. Students need to sign up before June 30, 2025, to get free access through spring 2026. After registering, they’ll need to reverify their student status before the next school year.
For K-12 education, Google provides AI resources through Google Workspace for Education, with Gemini available as an add-on.
Benefits for learning and research
These tools make studying and research much more efficient. Gemini Advanced handles complex problem-solving and lets students upload entire textbooks (up to 1,500 pages) to ask questions that span multiple chapters.
NotebookLM turns course materials into quick summaries, FAQs, timelines, and briefings that highlight important information.
Students can create custom practice tests based on their notes or previous exams to focus on areas where they need more practice.
For research projects, Deep Research browses and analyzes hundreds of websites in real-time, creating comprehensive research reports in minutes instead of hours.
The package also includes 2 TB of storage for course materials, assignments, and personal files.
Google maintains its commitment to security and privacy with these tools, using the same protection levels as their other services while following educational standards like FERPA and COPPA.
AI-Powered Grid Optimization
Image Source: Data Center Frontier
Google’s new initiative brings AI to electric grid management through a partnership with PJM Interconnection, North America’s largest grid operator, and Tapestry, an Alphabet moonshot company.
AI in energy infrastructure
The US electricity grid struggles with a massive backlog of more than 2,600 gigawatts of potential capacity waiting to connect to organized power grids.
This backlog exceeds double the size of the total installed US power fleet capacity. Where grid operators once handled a few dozen interconnection requests yearly, they now face thousands.
We’re seeing Tapestry, powered by Google Cloud and Google DeepMind, develop specialized AI tools that address these challenges head-on.
Their technology will optimize power generation connections to PJM’s electric grid across 13 states and the District of Columbia.
The system brings dozens of existing databases into a unified model, creating a secure platform where grid planners and project developers can collaborate effectively.
Partnerships and implementation
The initiative unfolds in phases starting in 2025.
Tapestry begins by streamlining the interconnection application and verification process.
Their tools automate time-intensive processes that currently take years to complete.
PJM’s move to a cluster-based interconnection process gains momentum from these innovations as they work through the final 67 GW of projects in their transition phase. The technology focuses on:
- Accelerating capacity additions through automation
- Unifying scattered databases into a comprehensive grid model
- Supporting faster integration of variable energy resources like wind and solar
Impact on sustainability
This AI partnership does more than improve efficiency.
The Federal Energy Regulatory Commission has updated its five-year demand forecast, now expecting US peak load to increase by 128 GW before 2030, triple previous estimates.
With this technology, power generation projects connect to the grid faster, making electricity more affordable and reliable for the 67 million people PJM serves.
This matters especially as data centers—which power our AI-driven world—consume more electricity than ever before.
The collaboration shows how tech companies can positively impact power infrastructure, positioning AI not just as an energy consumer but as a solution to our growing energy challenges.
Sec-Gemini v1 for Cybersecurity
Google unveiled Sec-Gemini v1 in April 2025, an AI model built specifically for cybersecurity challenges. This specialized version of Gemini marks a shift toward security operations enhanced by AI capabilities.
Sec-Gemini v1 capabilities
Sec-Gemini v1 pairs Gemini’s reasoning abilities with real-time cybersecurity knowledge and tools.
The model achieves superior results by integrating with Google Threat Intelligence (GTI), the Open-Source Vulnerabilities database, and Mandiant Threat Intelligence data.
The performance numbers speak for themselves. Sec-Gemini v1 scores at least 11% higher than competing models on the CTI-MCQ threat intelligence benchmark.
It also outperforms competitors by 10.5% on the CTI-Root Cause Mapping benchmark, showing its skill in understanding vulnerability descriptions and classifying them using Common Weakness Enumeration.
What makes this model stand out is how it connects vulnerabilities to threat actors.
When asked about Salt Typhoon, Sec-Gemini v1 identifies it as a threat actor and provides detailed descriptions of related vulnerabilities.
Cybersecurity use cases
The model excels in several key security workflows thanks to its integration with multiple data sources:
- Incident root cause analysis to track security breaches to their source
- Threat analysis to identify and understand attack patterns
- Vulnerability impact assessment to determine risk profiles
Security teams can use Sec-Gemini v1 to understand risks associated with specific vulnerabilities much faster.
For example, when responding to queries about Salt Typhoon, the model provides both vulnerability details and places them in context within the broader threat landscape.
Future of AI in security
Google plans to make Sec-Gemini v1 freely available to select organizations, institutions, professionals, and NGOs for research purposes.
This collaborative approach acknowledges that no single organization can fight global cyber threats alone.
Before widespread adoption, we need to view AI as a tool that enhances human decision-making rather than replacing it.
The most effective approach will likely combine AI capabilities with human expertise in a human-in-the-loop model.
Looking forward, AI-powered security tools may help address the fundamental challenge in cybersecurity – attackers only need to find one vulnerability, while defenders must secure entire systems.
AI could help balance this asymmetry by multiplying the effectiveness of cybersecurity professionals.
DolphinGemma: AI for Marine Research
Image Source: AllAboutAI.com
Google’s latest AI model swims into new territory with the National Dolphin Day 2025 launch of DolphinGemma.
This specialized tool shows how language model technology works beyond human communication.
DolphinGemma model overview
DolphinGemma works as an audio-in, audio-out model built specifically to analyze dolphin sounds – from clicks to whistles to burst pulses. Built on Google’s Gemma framework, this ~400 million parameter model runs efficiently on Pixel smartphones that researchers use in the field.
The system uses Google’s SoundStream tokenizer to compress and process complex dolphin audio signals. Similar to how human language models predict next words in sentences, DolphinGemma processes dolphin sound sequences to spot patterns and predict likely following vocalizations.
Research collaboration details
This project grew from a partnership between Google, Georgia Tech researchers, and the Wild Dolphin Project (WDP). WDP has studied a specific community of wild Atlantic spotted dolphins in the Bahamas since 1985 – the world’s longest-running study of its kind.
Their non-invasive methods have created decades of underwater video and audio matched with individual dolphin identities, life histories, and observed behaviors. This extensively labeled dataset provided the essential training material for DolphinGemma. The technology also works with the Cetacean Hearing Augmentation Telemetry (CHAT) system to enable direct interaction between humans and dolphins.
Implications for wildlife studies
DolphinGemma opens significant possibilities for wildlife conservation beyond analyzing dolphin communications. Google plans to release DolphinGemma as an open model in mid-2025, making it available to researchers studying other cetacean species.
While trained specifically on Atlantic spotted dolphin sounds, researchers studying bottlenose or spinner dolphins can fine-tune it for their work. This approach to sharing tools matches broader trends in AI for conservation.
AI applications for marine wildlife now include monitoring marine mammals through acoustic systems, detecting illegal fishing, studying coral reefs, conducting biodiversity surveys, and predicting environmental changes.
FAQs
Q1. What are the key features of Google’s Gemini 2.5 Pro?
Gemini 2.5 Pro boasts a massive 1 million token context window (expanding to 2 million), multimodal processing capabilities, and enhanced reasoning abilities. It excels in advanced coding, video understanding, and complex problem-solving tasks.
Q2. How does Google’s new Multimodal Search work?
Multimodal Search combines Google Lens technology with Gemini to allow users to search using both images and text simultaneously. It can analyze entire scenes, recognizing objects and their relationships, and provide contextually relevant responses to complex queries.
Q3. What is the impact of AI Overviews on travel and entertainment searches?
AI Overviews have dramatically expanded in these sectors, with entertainment queries increasing by 528% and travel queries by 381%. This shift is changing how users discover content and plan trips, with AI generating comprehensive guides and recommendations.
Q4. What is Gemini Live and how can it be used?
Gemini Live is a feature that enables real-time camera and screen sharing with Google’s AI. It can be used for tasks like space organization, creative brainstorming, and shopping assistance by allowing the AI to visually analyze and interact with your surroundings.
Q5. How is Google applying AI to cybersecurity?
Google has introduced Sec-Gemini v1, an experimental AI model specifically designed for cybersecurity. It integrates threat intelligence data and excels in tasks like incident root cause analysis, threat assessment, and vulnerability impact evaluation, outperforming other models on key benchmarks.
