monthly users reached by AI Mode about one year from its debut, with query volume more than doubling every quarter since launch. Overall searches reached an all-time high.
Google I/O 2026: Search stops answering and starts acting.
On May 19, 2026, during Google I/O, Search changed its skin once again. Gemini 3.5 Flash becomes the default model of AI Mode worldwide, the search bar receives its biggest update in 25 years, and most importantly, agents make their appearance: software that monitors the web 24/7, makes bookings, calls businesses, and builds custom interfaces on the fly. A year after AI Overviews, the question for businesses shifts again: no longer just "am I cited in the answer?", but "am I reachable by the agent acting on the user's behalf?".
Six data points to put the leap into perspective.
All announced or confirmed by Google during the May 19, 2026 keynote and in official Search materials.
is the timespan Google attributes to the biggest update ever made to the search bar: now expandable, multimodal (text, images, files, videos, Chrome tabs) and with generative suggestions beyond autocomplete.
is the schedule of Information Agents: they work continuously in the background, reasoning across web sources and Google's real-time data, and alerting the user when something relevant emerges, ready to propose an action.
countries and territories where Personal Intelligence in AI Mode is extending, in 98 languages and without a subscription. Optionally connect Gmail and Google Photos, and soon Google Calendar.
for everyone, from summer 2026, generative interfaces in Search: layouts, visual tools, tables, charts, and simulations assembled on the fly by Antigravity and the agentic coding of Gemini 3.5 Flash.
in three years: from an index of links (pre-2024) to synthesized answers (AI Overview, 2025) to agentic action (I/O 2026). Each leap raises the bar for what is needed to be visible.
From answer to action: what Search does today.
In our previous case study, we described the first major shift: SERPs ceasing to be an index of links and becoming a synthesized answer. AI Overviews had reached 54% of Italian searches and organic click-through had plummeted by 61% on informational queries. I/O 2026 adds a second, qualitatively different shift: Search no longer limits itself to composing an answer, but performs actions on behalf of the user.
The starting point is the model. Gemini 3.5 Flash becomes the default model of AI Mode for all users worldwide: it is the engine designed for agent functions and coding, and it is what makes everything else possible. Around this model, Google redesigns the only element that remained almost unchanged for two decades: the search bar. The new one expands as you type, accepting text, images, files, videos, and even open Chrome tabs as input, and suggests how to continue the question with generative suggestions that go beyond autocomplete. In practice, Google is encouraging longer, conversational, and multimodal queries: the three-keyword dry query becomes the exception, not the rule.
Upon this foundation, agents are born. Information Agents are configurable software systems that remain active 24/7: they monitor blogs, publications, social media, and Google's real-time data on finance, shopping, and sports, then alert the user when something matching their criteria emerges, proposing an action. Google's example is telling: home buyers can describe their requirements and let the agent scan the market and flag the right listings as they appear.
Then comes actual action. Agentic bookings extend to local experiences and services: you describe what you want ("a karaoke room for six, Friday night, serving food late") and Search gathers prices and availability with direct links to finalize. For some categories — home repairs, beauty, pet care — Google can even make phone calls to businesses on the user's behalf. Finally, generative interfaces: Search writes code on the fly, with Antigravity, to build the tool best suited to the query — a simulation, a comparison, or even a mini-app like a workout tracker drawing from local reviews, maps, and weather. Personal Intelligence closes the loop by connecting personal context (Gmail, Photos, soon Calendar) to make all this relevant to the individual.
What concretely changes.
| Search with AI Overview (2025) | Agentic Search (post I/O 2026) | |
|---|---|---|
| What the engine does | Synthesizes an answer from sources | Acts: monitors, books, calls, builds tools |
| Query input | Text, partially voice and images | Text, images, files, video, Chrome tabs — multimodal and conversational |
| Result format | Generated text block + links below | Tailor-made generated interface: tables, charts, simulations, mini-apps |
| Interaction duration | Single query, potential follow-up | Continuous: the agent works in the background 24/7 over time |
| Brand touchpoint | User reads and potentially clicks | Agent reads, compares, and can book/call on behalf of the user |
| What is needed for visibility | Being cited as a reliable source | Being citable and actionable: price, availability, hours, booking data readable by a machine |
| Personalization | Limited, mainly query-based | Personal Intelligence: context from Gmail, Photos, Calendar (opt-in) |
| Primary metric | Share of voice in AI citations | Share of voice + agent-readiness: presence in actions executed by agents |
What it means, feature by feature, for those who want to be found.
Every announcement of I/O 2026 has a direct consequence on how a business becomes (or stops being) visible. Here we translate features into operational implications for SEO and GEO.
Information Agents
Freshness becomes a selection factor
A 24/7 active agent rewards sources that update and remain reliable over time. A static page, published once and never touched again, has little chance of being retrieved by an agent looking for the latest relevant update. Regular publishing, with explicit dates and verifiable data, ceases to be an editorial habit and becomes an agentic indexing requirement.
Agentic bookings and calls
Actionability matters as much as reputation
For repairs, beauty, pet care, dining, and local services, Google's agent can book or call on behalf of the user. It is crucial that hours, availability, prices, booking methods, and phone numbers are correct, updated, and structured (Google Business Profile, schema markup, booking data). Local SEO is transforming into local agent-readiness: if the agent cannot read you or book you, it chooses the competitor it can manage.
Generative interfaces
Fewer clicks, more value as a data source
If Search builds the tool on the fly (comparison, simulation, mini-app) by pulling real-time data, the user gets the functional answer without visiting a site. The dynamic already seen with AI Overviews is accentuated: less traffic, but the site becomes the data provider feeding the generated interface. Public, structured, and citable data are worth more than a closed, beautiful landing page.
Personal Intelligence
Relevance beats average positioning
With personal context (Gmail, Photos, soon Calendar), results adapt to the individual. The "average position" valid for everyone loses weight: the winner is whoever is clearly relevant to a precise intent and specific context. For businesses, this means factual content, segmented by real-use case, instead of generic pages optimized for broad keywords.
Multimodal search bar
Long queries and conversational as the norm
Inputs from text, images, files, videos, and Chrome tabs push towards long, specific questions. Optimization for dry keywords gives way to optimization for intents and entities: covering real questions, formulated as a person would ask them, and structuring content to answer conversational variations of the same need.
Gemini 3.5 Flash by default
Scale: AI Mode is now mainstream
With one billion monthly users and queries more than doubling each quarter, AI Mode is no longer an experiment for early adopters. Optimizing only for the "ten blue links" means designing for an experience that a growing portion of users no longer sees as primary. The working assumption must be reversed: AI-first, not AI-as-well.
What businesses need to do now.
Classic SEO remains the foundation: what ranks in the top-10 today is still the most likely candidate to be cited and used by agents. But a new, agentic layer must be built on top. Three concrete focus areas.
Make the site agent-readable
Expose structured and updated data that an agent must be able to act on: services, prices, availability, hours, served areas, booking and contact methods. The page must be readable by a machine even before a person.
- Schema Service, Offer, LocalBusiness, OpeningHours
- Complete and verified Google Business Profile
- Exposed booking/availability data
- llms.txt and robots open to AI bots
Maintain freshness and monitorability
Information agents look for the latest relevant update. Key content must be treated as live data: regularly updated, explicitly dated, with primary sources and verifiable figures. Constant publishing becomes an indexing signal, not just an editorial one.
- Update schedule for top-50 contents
- Visible and marked publication and modification dates
- Verifiable and cited data and sources
- Feeds/updates for strong-freshness topics
Strengthen entity identity
Agents and models trust recognizable and reliable entities. You need a solid profile in the knowledge graph: structured organization and authors, verifiable authority signals, consistent presence across sources. It is what makes you selectable when an agent must choose just one option.
- Schema Organization + Person + sameAs
- Recognizable authors with bio and profile
- Verifiable reviews and case studies
- Real thematic topical authority
What to measure and monitor after I/O 2026.
Generative search metrics remain valid; I/O 2026 adds a family linked to agent action. A guide to understanding where you are ready and where you are not.
| Dimension | What to monitor after I/O 2026 | Tool / signal |
|---|---|---|
| Citability in AI responses | Share of voice in AI Overviews and AI Mode on relevant prompts | Otterly, Profound, BrandRank.AI |
| Local actionability | Bookable data: correct and structured hours, availability, prices, phone | Google Business Profile + schema LocalBusiness/Offer |
| Freshness | Update frequency of key content and marked dates | Editorial audit + dateModified in schema |
| Machine readability | Schema coverage and accessibility to AI bots (robots, llms.txt) | Rich Results Test + check llms.txt/robots |
| Entity identity | Recognition in the knowledge graph of organization and authors | Google Knowledge Graph Search API + brand monitoring |
| Qualified traffic | AI-referred visits and conversions from ChatGPT, Perplexity, Gemini, Copilot | GA4 + source filter + custom segments |
| Conversational intent | Coverage of long and multimodal questions, not just dry keywords | AI Mode query analysis + content audit for intents |
The question is no longer "am I first?". It is "can the agent use me?".
I/O 2026 confirms a trajectory that has lasted for three years and accelerates at each stage. First, the SERP was an index and the synthesis work was done by the user. Then the generated response arrived and the work was done by the model. Now action arrives, and a growing part of the work — searching, comparing, booking, calling, building tools — is done by the agent. With each step, the point of contact between company and customer moves one degree further away from the traditional website.
This does not make the website useless, but it changes its role once again. After AI Overviews, we said the site goes from destination to source. After I/O 2026, a layer must be added: the site also becomes an endpoint for agents — a place from which software reads data and, where possible, performs actions. A page that converts little human traffic but is read correctly by agents across thousands of relevant intents holds a value that no session-based KPI can measure.
The good news is that much of the work is the same that makes a site robust for GEO: structured data, factual and updated content, clear entity identity, verifiable authority. The difference is that now these elements do not just serve to get cited in an answer, but to get selected and executed by an agent. The most advanced features will land in the US and for Pro subscribers first, but the direction is set and timelines are measured in months.
As with AI Overviews, the greatest risk is not the change itself: it is not noticing it. A business can continue to see "normal" KPIs while, underneath, agents begin to systematically choose the most readable and actionable competitors. Those who build the agentic layer now — while the market is still focused on the ten blue links — get ahead on a lead that, this time too, will be hard to catch up with later.
On Google I/O 2026 and agentic search.
What did Google announce for Search at I/O 2026?
At Google I/O on May 19, 2026, Google presented five main new features for Search: Gemini 3.5 Flash as the new default model for AI Mode globally; the most updated search bar in 25 years, multimodal and with generative suggestions; Information Agents, AI agents that monitor the web 24/7 and alert the user; the expansion of agentic bookings and phone calls to businesses on behalf of users; generative interfaces built on the fly by Search; and the extension of Personal Intelligence to nearly 200 countries and 98 languages.
What are Google's Information Agents?
They are user-configurable AI agents that work in the background 24 hours a day, 7 days a week. They continuously monitor the web (blogs, publications, social posts) and Google's real-time data on finance, shopping, and sports, to flag updates on a specific topic and propose an action — for example, alerting you when a real estate listing matching your criteria appears. Google plans to launch them for AI Pro and Ultra subscribers in summer 2026.
What does "agentic search" mean for SEO and GEO?
It means Search no longer just answers, but acts: it monitors, books, calls, and builds interfaces. For business visibility, the touchpoint shifts from the user to the agent. It is no longer enough to be positioned or cited: you must be readable and actionable by an agent, with structured data on availability, prices, hours, bookings, and a clear entity identity in the knowledge graph. GEO thus extends to agent-readiness: making your content and services consumable by software agents.
Will Google's generative interfaces reduce website traffic even further?
That is the most likely trajectory. With generative interfaces, Search assembles tables, charts, simulations, and custom mini-apps based on the user's query on the fly, pulling real-time data. The user gets the tool directly on the results page, further reducing the need to visit a site. This reinforces the trend already documented with AI Overviews: the site loses clicks but gains value as a data source feeding responses and tools generated elsewhere.
Does Personal Intelligence change the way we rank on Google?
In part, yes. With Personal Intelligence extended to nearly 200 countries and 98 languages, and with the optional connection of Gmail, Google Photos, and soon Calendar, AI Mode results become more sensitive to the user's personal context. The "average" ranking valid for everyone loses weight in favor of relevance to a specific person and their context. For businesses, being clear, factual, and relevant to a precise intent matters more than optimizing for a generic keyword.
What should businesses do after Google I/O 2026?
Three priorities. (1) Agent-readiness: expose structured and updated data on services, prices, availability, hours, and bookings, so that agents and models can read and act on them. (2) Freshness and monitorability: regularly update key content, because information agents reward sources that change and remain reliable over time. (3) Entity identity: strengthen your profile in the knowledge graph with complete schema markup, recognizable authors, reviews, and verifiable authority signals. Classic SEO remains the foundation; an agentic layer must be built on top.
Is your business ready for agentic search?
Free agent-readiness audit: we measure where you are already cited by AIs, whether your service and booking data are readable by agents, and what concrete actions to take in the next 90 days to avoid remaining invisible as Search learns to act.
Request free GEO audit
Document type: market case study — analysis of the new features presented at Google I/O 2026 and their SEO/GEO implications.
Published: May 20, 2026 — Last updated: May 20, 2026
Author: Di Marco Nardino, Startupweb.io
Primary source: official Google I/O 2026 announcements (keynote and Google Blog, "A new era for Search with AI", May 19, 2026). Market data on AI Overviews and zero-clicks are sourced from our previous case study; interpretations and operational implications are original Startupweb.io analyses.
Related resources.
AI Overviews and the new role of the website
The previous case: how generative search changed the Italian SERP and transformed the website from destination to source.
Read analysis ServiceGEO & AI Search
The operational service: how we make you citable and actionable by ChatGPT, Perplexity, AI Overviews, AI Mode, and Google's agents.
Discover the service IndexAll case studies
Return to the complete list of industry analyses and client cases of Startupweb.io.
See all