google i o 2026 for geo and seo

“Search Agents” vs. AI Overviews and AI Mode: what Google I/O 2026 shift means for GEO and SEO

André Pitì

20 de May de 2026

Google’s I/O 2026 promises to upgrade Search from a place you go to ask questions into an always-on agent that finds answers, books appointments, and completes tasks before you even think to ask.

But what businesses is this update mostly affecting?

The ecommerce has the most actionable near-term response, local services have the most urgent behavioral change to make (phone responsiveness, booking integration, operational data quality), and content/publisher businesses have the most structural rethinking to do with the least clear path forward.

TL;DR on the Google I/O 2026

Key Findings from the I/O 2026 panel

  1. Search Agents are a category, not a feature
    Liz Reid’s framing at I/O 2026 was explicit: “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.”

The first sub-product is information agents, persistent background processes that monitor blogs, news, social, and Google’s real-time finance/shopping/sports data 24/7 and push synthesized updates. Information agents launch summer 2026 for U.S. Google AI Pro & Ultra subscribers first.

  1. The reactive-to-proactive shift is the defining architectural change
    AI Overviews and AI Mode both wait for a query. Information agents don’t.

As The Next Web put it on May 19, 2026: “Traditional search is reactive: it waits for you to show up with a question. Information agents are designed to operate continuously in the background, around the clock.”

  1. Agentic booking + voice calling collapses the local-SEO funnel into Google itself
    Google will now call businesses on a user’s behalf in select U.S. categories (home repair, beauty, pet care) and book local experiences and services on multi-criteria prompts (the famous “private karaoke room for six on a Friday night that serves food late” demo). For local SEO, this means a business’s operational data ( hours, availability, prices, policies, phone responsiveness) becomes a ranking and conversion factor, not just a discovery factor.

  2. Personal Intelligence destroys universal rankings
    Now live in nearly 200 countries and 98 languages free of charge, connecting Gmail, Google Photos, and (soon) Calendar to AI Mode, Personal Intelligence means two users typing the same query get materially different AI answers. Glenn Gabe (GSQi) called this “another big step toward Jarvis.” For SEOs, traditional rank-tracking becomes structurally inadequate; brand presence in user-frequented sources (inbox, photo library, calendar, repeat visits) becomes a personalization signal.

  3. Generative UI + mini-apps further reduce the need to click out
    Powered by Gemini 3.5 Flash and Antigravity, Search can now build interactive simulations, dashboards, and persistent “mini apps” (e.g., a custom fitness tracker, wedding planner) on the fly. For utility-style content (calculators, comparison tables, how-to guides), the substitution risk is now nearly total: Search itself becomes the app.

  4. Google’s own May 15, 2026 guidance frames the GEO debate
    Four days before I/O, Google published its first optimization guide for generative AI features. The community read it as a deflation of GEO hype: no llms.txt needed, no chunking needed, structured data not specially required, and “inauthentic mentions” (paid listicle seeding) flagged as risky.

Lily Ray on X and LinkedIN (May 15, 2026): “SEO is still the foundation for AI search; Create non-commodity, ‘people-first’ content; Ignore most ‘GEO/AEO hacks like chunking and llms.txt.”

Aleyda Solis (X, May 15, 2026): “The best practices for SEO continue to be relevant because their underlying logic still applies.”

  1. Query fan-out is the technical mechanism that ties all three (AIO, AI Mode, Search Agents) together
    AI Mode decomposes a single user query into dozens of synthetic sub-queries executed in parallel (query fanout), then synthesizes a response, which makes SEOs question what kind of content succeeds in AI search engines (not just Google, ChatGPT and other models too). Search Agents extend this temporally: not just fan-out once, but fan-out continuously, on a schedule, against fresh data.

    Optimization at the passage level (what Mike King, iPullRank, calls “Relevance Engineering”).

  2. Traffic data confirms the structural shift, but Search is not collapsing
    AI Mode crossed 1 billion monthly users one year after launch with queries doubling every quarter.

Per Sundar Pichai at the I/O 2026 keynote, “AI Overviews now has over 2.5 billion monthly active users.”

Independent measurement still shows AI Mode usage well below traditional Google search in absolute terms (Rand Fishkin/Datos Q1 2026: AI Mode at less than 0.2% of total search share).

But click-through rate research consistently shows steep declines on AIO-present queries: in July 22, 2025, Pew Research Center, studied 68,879 real searches across 900 U.S. adults and found an 8% click rate with AIO vs. 15% without (a 46.7% relative decline).

Ahrefs‘ Ryan Law measured a 34.5% position-1 CTR drop across 300,000 keywords in April 2025, worsening to 58% by December 2025; Amsive‘s analysis of 700,000 keywords across 10 sites found a 15.49% average CTR decline, spiking to 37.04% when AIO was stacked with a featured snippet.

Per the Reuters Institute for the Study of Journalism’s Journalism and Technology Trends 2026 report (January 2026), citing Chartbeat data from approximately 2,500 news sites for November 2024–November 2025, organic Google search traffic to publishers fell 33% globally and 38% specifically in the U.S.

1. AI Overviews vs. AI Mode: the baseline you need to know before understanding Search Agents

AI Overviews (AIO) launched broadly in May 2024. They are AI-generated summary blocks that appear above the traditional ten blue links on roughly 25.8% of U.S. queries as of January 2026 (rising to ~48% on pure informational intent, per Stackmatix/Conductor 2026 data).

Source: Stackmatix

They are:

AI Mode launched broadly to U.S. users in May 2025 and is now globally available with Gemini 3.5 Flash as the default model. It is:

Ahrefs’ study put the difference clearly: AIO and AI Mode are not “short” and “long” versions of the same answer; t hey are two distinct systems with 86% semantic agreement but very different source selection.

Per Ahrefs’ Despina Gavoyannis study (December 2025 on 730,000 query pairs): “If your brand gets mentioned in AI Overviews, there’s a 61% chance it’ll also appear in AI Mode’s longer response.”

The implication for SEOs: optimizing for one does not guarantee visibility in the other. It’s a meaningful overlap, but not deterministic.

2. What Search Agents actually are (and why they’re different from both)

Search Agents, announced at Google I/O 2026 on May 19, are an entirely new layer that sits on top of AI Mode and changes search along three orthogonal axes:
Dimension AIO AI Mode Search Agents
Trigger User query User query User-set intent, then autonomous
Duration One-shot Single session Persistent (days to weeks)
Surface SERP block Chat tab Push notifications + AI Mode history
Goal Summarize Explore Complete tasks / monitor changes
Data scope Web index Web + fan-out Web + finance/shopping/sports real-time feeds + social posts
Action capability None (links only) Limited (some bookings in Labs) Call businesses, complete bookings, build mini-apps, transact via Universal Cart

The Search Agents stack as announced has five interlocking components:

  1. Information agents (summer 2026, AI Pro/Ultra first): user describes a topic (e.g., “keep me updated on nearby movie tickets for The Mandalorian and Grogu”) and the agent monitors continuously, sending push notifications.
  2. Expanded agentic booking: across home repair, beauty, pet care, and local experiences, including direct phone calls Google places to businesses on the user’s behalf (U.S., summer 2026 rollout).
  3. The new intelligent Search box: Liz Reid calls it “the biggest upgrade to our iconic search box since its debut over 25 years ago.” It dynamically expands for long queries, accepts text/images/files/videos/Chrome tabs as inputs, and offers AI-powered query suggestions that go beyond autocomplete.
  4. Generative UI and Antigravity mini-apps: Search itself codes interactive widgets, calculators, simulations, and persistent dashboards on the fly. Generative UI is free for all users this summer; mini-apps roll out to U.S. AI Pro/Ultra subscribers in the coming months.
  5. Personal Intelligence at global scale: Gmail and Google Photos connections (with Calendar coming) are now available free in ~200 countries/98 languages.

3. What this means for GEO (Generative Engine Optimization)

Does traditional SEO still matter?

Yes, more than ever, but it is now the floor, not the ultimate ceiling.

Google’s May 15, 2026 official guide (written before the I/O announcement but published as deliberate context for it) is the single most important text for SEOs to internalize.

Its consolidated message, reinforced by Lily Ray, Aleyda Solis, Glenn Gabe, Marie Haynes, and Pedro Dias on X the same day: AEO and GEO are not separate disciplines. They are foundational SEO applied to AI surfaces.

Google explicitly listed five areas it considers important and pushed back on five tactics it considers wasted or risky.

Google says support visibility:

Google says waste budget or carry downside risk:

Mike King at iPullRank goes further and argues Google’s guidance understates how different the technical work actually is. His “Relevance Engineering” framework calls for passage-level vector embedding, query fan-out simulation (his free Qforia tool), and content engineered across “constellations” of synthetic sub-queries. Both positions can be true simultaneously: nothing fundamentally changes about what makes content good, but a lot changes about how you measure and engineer it.

What shifts for content creators and SEOs

  1. From keywords to topic and entity coverage across the journey
    Aleyda Solis’s #SEOFOMO 2026 trends survey of practitioners showed convergence on this: from “keyword-first” to “intent, entity & problem-solving models,” and from “rankings & clicks” to “visibility, mentions & citations.”

  2. From single-page optimization to “passage portfolios” across multiple pages and properties
    From Mike King quote on Digiday: “Trying to compete for volume keywords is not happening anymore. We have to focus on the specific journey our customers might be on.” His raffle-ticket analogy: more relevant passages across more pages equals more opportunities for AI to select your content.

  3. From owned-media-only to reciprocal mentions across third parties
    Ethan Smith ( Graphite): “Reciprocal mentions is the new reciprocal backlink.”

Per Yahoo!Finance publication in March 2026: “YouTube accounted for 31.8% of citation volume. Reddit led at 46.4%. Together, the two platforms represent 78.2% of all social media citations observed across the six AI search platforms analyzed.”

Reddit, YouTube, Wikipedia, LinkedIn, and reputable trade press seems to be the most-cited sources in AI answers across most verticals.

  1. From traffic to citation share + brand visibility
    Aleyda Solis’ deck for Women in Tech (Feb 2026): SEO success now means tracking visibility, citations, sentiment, share of voice — plus conversions, revenue, CAC, ROI. Rankings remain a leading indicator, but they are no longer the KPI itself.

What changes for content that gets cited / surfaced

Kevin Indig’s research on 1.2 million search results (March 2026) and 815,000 query-page pairs (April 2026) is the clearest empirical work on what AI cites:

A Search Engine Land fan-out analysis by André Pitì & Ben Tannenbaum (April 30, 2026), testing 90 prompts across beauty, legaltech, and IT, found that commercial prompts triggered ChatGPT web searches 78.3% of the time, while informational prompts did so just 3.1%.

Of the 42 fan-out sub-queries generated, 39 were commercial, expanding into comparisons, shortlists, feature filters, and brand evaluations. The practical implication: purely educational content that avoids product names, tradeoffs, pricing logic, or selection criteria is unlikely to appear on the fan-out branches AI systems actually retrieve from. Search Engine Land

Practical implication: lead the page with a 50–70 word, fact-dense answer module; stack 4–6 follow-up modules covering the natural sub-queries (definition, comparison, decision rule, pricing, FAQ); ensure structured data on entities, authors, products; and earn citation from credible third parties.

What agentic booking and calling mean for local SEO and local businesses

The shift here is severe and concrete.

When Google itself calls a business on the user’s behalf, three things stop mattering as much: phone-prominence in the local pack, branded paid search, and (over time) website-form conversions. Three things start mattering more:

  1. Operational data integrity Hours, availability, pricing, service area, capacity windows, and policies ( as exposed to Google) become a precondition for being selected by an agent.

  2. Phone responsiveness and “agent-friendliness”
    A business that doesn’t answer, takes 30 seconds of IVR before reaching a human, or refuses to provide a quote over the phone, will be deprioritized in the agent’s selection logic over time. Local Service Ads-style velocity signals (review velocity, responsiveness, visual trust) become more important than review count alone.

  3. Booking-platform integration
    Restaurants on OpenTable/Resy, salons on connected scheduling platforms, home-service providers on integrated booking systems (these become preferred because the agent can actually complete a booking). Businesses with phone-only intake become slower-path options. Property managers, OTAs, and aggregators face displacement risk if they don’t expose machine-readable inventory.

For local SEO practitioners: Google Business Profile fundamentals (NAP consistency, hours accuracy, photos, reviews, categories) remain core, but the next layer of work is integrating booking APIs, ensuring real-time availability is exposed to Google, and treating phone responsiveness as a marketing KPI.

How the Personal Intelligence layer changes visibility

This is the most subtle and arguably the most consequential change. Personal Intelligence connects AI Mode to a user’s:

For an SEO, this means two users in the same city searching “ best dentist near me” can get different AI Mode answers depending on whether one has a Gmail receipt from your practice, a Calendar appointment, a Photos image at your location, or a YouTube watch history of your educational videos!

Visibility now depends on a web of touchpoints:

Glenn Gabe on this point (paraphrasing his GSQi post): “When the results are hyper-personalized, companies will need to do real marketing and not rely on some tricks or gimmicks.”

Content and site categories at most/least risk

Highest risk:

Lowest risk:

4. Recap list of expert comments by industry leaders on the I/O 2026 panel

Worth noting

A meaningful tension worth flagging: Google’s numbers (“1 billion MAU in AI Mode,” “queries doubling every quarter”) and Fishkin/Datos’ panel data (AI Mode at <0.2% share of searches) describe the same world differently.

Both are likely directionally correct: AI Mode is huge as an installed base but still small as a fraction of total search occasions. For decision-making, treat AI Mode adoption as a trajectory you need to position for, not as today’s traffic source.

5. How is this impacting different business models

Ecommerce feels it first and most tangibly

The entire stack (Universal Cart, UCP, AP2, agentic booking, real-time inventory exposure) is built with transactions in mind. An agent that can find, compare, and buy a product without the user ever visiting a website is a direct revenue event.

The good news for ecommerce is that Google still needs merchants in the loop (they remain Merchant of Record). The bad news is that if your feed data, pricing, returns policy, and availability aren’t clean and machine-readable, you simply won’t be selected. It becomes an operational/data quality problem as much as a marketing one.

Local service businesses face the most disruptive single feature

This is the agentic calling capability.

A plumber, dentist, or nail salon that gets called on a user’s behalf and books an appointment without the user ever visiting their website or even Google Maps has effectively lost the discovery touchpoint entirely. Local SEO was always about getting found; now getting found may mean being acted upon by a machine. That’s a qualitatively different relationship.

Publishers and content-led businesses arguably face the deepest structural threat

At least over the medium term.

A blog, news site, or affiliate publisher that existed to answer questions are now at risk (especially if filled with AI slop). Search Agents and generative UI can now build custom dashboards and interactive tools on the fly, directly substituting for utility content.

B2B / SaaS businesses are the most insulated right now

As software buying cycles involve demos, security reviews, legal, procurement, an agent isn’t booking a $200K enterprise contract. But awareness and consideration stages (comparison content, analyst mentions, community reputation) are a bsolutely affected by where AI citations come from.

Practical recommendations driven by Google I/O 2026 Keynote

Stage 1: to do immediately  in the next 30 days

  1. Audit operational and feed data
    For ecommerce: feeds in Merchant Center accurate, return policies defined, agentic checkout eligibility signals set, identifier coverage complete. For local services: GBP fully complete, hours/categories/photos current, booking integrations live wherever possible.

  2. Audit your pages for information gain
    Lily Ray and Aleyda Solis converge here: commodity content that any LLM can re-synthesize is being silently demoted. Identify pages that don’t include proprietary data, first-person experience, original photography, or expert credentials; these are your biggest exposure.

  3. Set up AI-citation tracking
    Even if volume is small, track trajectory. It can be based on Microsoft Clarity + GA4 referrals (that’s the easy, yet limited view), or you can set up a complete, precise and zero-part-data-based view for your business by contacting us. Ore you can choose an enterprise platform like Profound, Peec AI, Semrush Enterprise AIO, or others. Without this you cannot manage what you cannot see.

  4. Build a brand-mention strategy across third parties
    Reddit AMAs, YouTube content, expert podcast appearances, Wikipedia/data sources, trade press contributions, and industry research. This is the new link-building.

Stage 2: to-dos in the next 30–90 days

  1. Restructure top pages for passage-level extraction
    Lead with 50–70 word answer modules. Stack 4–6 follow-up modules covering the natural fan-out (definition, comparison, when-to-use, cost, FAQs).

  2. Invest in original research and proprietary data
    This is the single highest-leverage SEO/GEO investment of 2026 (Kevin Indig’s data, Lily Ray’s repeated argument, Cyrus Shepard’s “proprietary assets” factor at 0.357 correlation).

  3. For ecommerce/SaaS: join the UCP waitlist and structure your site to be agent-readable
    Real-time pricing, machine-readable availability, conversational commerce attributes (compatible accessories, product substitutes, usage scenarios), unique product imagery.

  4. For local businesses: integrate with booking platforms agents can transact through
    Phone-only intake is becoming a competitive disadvantage. Audit phone response time and treat it as a marketing metric.

  5. Build a branded search moat
    Branding is the opposite of dead. Branded queries remain the most defensible channel: PR, community, email, app, and direct relationships are the assets to compound.

Stage 3: reassessment with benchmarks/ thresholds

Final caveats from Google’s I/O 2026

The I/O 2026 announcement is fresh
Information agents, generative UI mini-apps, and the expanded agentic booking calling feature are all scheduled for “this summer” U.S. rollouts. Real-world performance and selection criteria are not yet observable. Treat all current advice as informed extrapolation from Google’s stated direction plus existing AI Mode/AIO empirical research.

Named-expert reaction cycle to I/O 2026 specifically is still building
Most substantive named commentary in the days immediately surrounding May 19 focused on Google’s May 15 official “Optimizing for generative AI in Search” guide, not on the I/O announcements themselves. Lily Ray, Mike King, Kevin Indig, Aleyda Solis, and Rand Fishkin have all written extensively on the underlying trends (agentic search, query fan-out, GEO discipline) but published opinion pieces specifically dissecting the I/O 2026 Search Agents announcement should be expected over the following 1–4 weeks.

Sourcing tension on AI Mode adoption
Google claims 1 billion monthly active users in AI Mode with queries doubling every quarter. Rand Fishkin/Datos panel data shows AI Mode at <0.2% of total search share. Both can be true simultaneously: large installed base, small share of total search occasions. Treat AI Mode and Search Agents as a future-state strategic priority, not yet a today-state traffic source for most businesses.

Personal Intelligence creates measurement opacity
Two users searching the same phrase get different answers; rank tracking as we know it cannot measure this. Several incumbent SEO platforms have not yet shipped reliable tooling. Expect a period of measurement uncertainty during 2026.

Regulatory risk
The European Commission’s April 2026 Digital Markets Act measures require Google to share anonymized search data with rival search and AI providers, with a compliance deadline of July 27, 2026. Future product changes in the EU may diverge from U.S. behavior. Also expect antitrust scrutiny to intensify as Search increasingly becomes self-contained AI rather than a gateway to the open web.

“Mount AI” risk
Lily Ray’s repeated warning is worth taking seriously: aggressive GEO tactics that produce rapid gains often precede equally rapid demotions when search engines update against them. Anything that looks like manipulation (inauthentic mentions, listicle farms, programmatic content scaling, llms.txt-style chunking, paid forum seeding) is likely to be classified as spam over the next 1–2 years.