AI Visibility · 12 min read · 20 May 2026

Google I/O 2026: what changed for marketing teams, and what to do in the next 90 days.

AI Mode became the default search surface with Gemini 3.5 Flash. Universal Cart launched for agentic checkout. Personal Intelligence expanded to nearly 200 countries. The marketing playbook from 2024 covers two thirds of the work that matters now. Here is the rest.

▸ Bottom line up front

The visibility surface that decides whether your brand enters a buyer's consideration set has shifted under the surface that decides whether your bid wins. Three protocol-level changes (AI Mode, Universal Cart, Personal Intelligence) mean technical eligibility now precedes spend. The 90-day fix is schema, entity, llms.txt, and a citation-share baseline.

On this page

What Google actually announced

At Google I/O 2026 on 19 May 2026, four announcements moved the load-bearing pieces of Search at the same time.

AI Mode runs on Gemini 3.5 Flash as the default model, globally. AI Mode crossed one billion monthly active users and queries are tripling in length relative to traditional Search. More than one in six United States searches now use voice or images, and image searches are growing about 40 percent month over month (Google, 19 May 2026).

An intelligent AI-powered Search box. The Search box is the biggest upgrade Google has made to it in 25 years. It expands dynamically, accepts text, image, file, video, or Chrome-tab inputs, and offers AI-powered suggestions beyond autocomplete. Conversational follow-up from an AI Overview into AI Mode is now one continuous flow across desktop and mobile worldwide.

Universal Cart, Universal Commerce Protocol (UCP), and Agent Payments Protocol (AP2). Universal Cart is an intelligent shopping cart that lives across Search, Gemini, YouTube, and Gmail. UCP is the common-language protocol between agents, merchants, and processors. AP2 is the secure payment rail with tamper-proof digital mandates. Launch partners on Universal Cart include Nike, Sephora, Target, Ulta Beauty, Walmart, Wayfair, and Shopify merchants such as Fenty and Steve Madden.

Personal Intelligence expanded to nearly 200 countries across 98 languages, no subscription required. Users can securely connect Gmail and Google Photos, with Google Calendar coming. For brands, this is the new in-market signal layer.

Information agents (operating 24/7 in the background) and agentic booking for local services were also announced, with agentic coding inside Search via Antigravity. The compounding effect of those four announcements is what this post is about.

One day later, on 20 May 2026, Google Marketing Live extended the picture into ads and measurement. The table below collapses both events into one scannable list (Google Marketing Live 2026 collection).

Every 2026 announcement at a glance

Thirteen announcements that moved buyer-facing surfaces, ads stack, and measurement layer at the same time. The third column ties each to a concrete 90-day action; the full sequencing rationale sits in the 90-day priority list further down.

Google I/O 2026 (19 May) plus Google Marketing Live 2026 (20 May): every marketing-relevant announcement and the 90-day action it implies.
Announcement What changed 90-day action
AI Mode default on Gemini 3.5 Flash Conversational Search is the default surface; AI Mode queries are about three times the length of traditional Search; one in six United States searches uses voice or images; image search up about 40 percent month over month. Audit citation share across AI Mode, ChatGPT, Claude, Gemini, Perplexity, and Bing Copilot. Lock the pre-implementation baseline on a 30-prompt buyer-intent set.
Intelligent AI-powered Search box Biggest Search-box upgrade in 25 years. Accepts text, image, file, video, or Chrome-tab input. Conversational follow-up flows from AI Overview into AI Mode without losing context. Ship ImageObject and VideoObject schema with captions, alt text, and transcripts on every primary visual or video asset.
Universal Cart (US, summer 2026) Intelligent cart that lives across Search, Gemini, YouTube, and Gmail. Launch partners: Nike, Sephora, Target, Ulta Beauty, Walmart, Wayfair, Shopify merchants including Fenty and Steve Madden. Canada, Australia, then UK in the following months. Audit product feed for UCP eligibility. Wire order-status callbacks to GA4, Google Ads Enhanced Conversions, and Meta Conversions API before the traffic arrives.
Universal Commerce Protocol (UCP) Common-language layer between agents, merchants, and processors. Expanding to new verticals with new product-discovery capability for conversational AI. Make catalogue UCP-speaking inside the 12-to-18-month APAC readiness window. Brands without UCP-ready feeds sit outside the cart layer while direct competitors are inside it.
Agent Payments Protocol (AP2) Tamper-proof digital mandates produce a shared digital record across user, merchant, and processor. Returns and disputes operate on one shared record, not two separate logs. Wire the AP2 audit trail into existing KYC and AML workflow. The brand stays merchant of record; the work is plumbing, not policy.
Personal Intelligence (200 countries, 98 languages) AI Mode grounds answers in the user's own Gmail and Google Photos (Google Calendar coming). Opt-in, disconnectable, and Google does not train models on the connected data itself. Improve email-receipt template quality (structured data, complete merchant identity). Complete Google Business Profile depth. Tighten consent records to cover agent-as-processor relationships.
Information agents User-defined background agents that monitor a query 24/7 and resurface matching content. The new push surface for category-watching buyers. Add named-author bylines, primary-source citations with publisher and date inline, visible publish date, JSON-LD dateModified, and stable URLs on every page.
New AI Search ad formats Ad units sit inside conversational AI Search responses. Google's stated framing is "helpful additions to the conversation" rather than interruption-led placement. Test creative built for conversational placement. Longer copy, declarative claims, primary-source citations, and brand-safe phrasing replace headline-and-pun creative.
AI Max for Shopping Retail-specific version of AI Max for campaign optimisation on modern Search. Calibrated for shopping-feed accounts. Pilot on the top two commercial-intent campaigns where last-30-day spend is high enough that a matched-control holdout produces a readable signal inside one calibration cycle, and re-poll citation tracking pre and post so the AI Max contribution is isolated from organic AI Mode movement on the same query set.
AI Max for Travel Vertical-specific AI Max for travel campaigns. Workflow simplification for booking-led ecommerce. If the account runs hotels, OTA inventory, flights, or experiences, pilot before peak booking season so the calibration cycle completes ahead of demand. Use the same matched-control holdout pattern as AI Max for Shopping, and measure cost-per-booked-stay rather than cost-per-click; the latter rewards the bid algorithm for chasing low-quality traffic.
Ask Advisor (live 20 May 2026) Unified AI agent spanning Google Ads, Analytics, Merchant Center, and Marketing Platform. One conversational interface across the stack. Grant the account team access on day one and pilot weekly diagnostic queries on the lowest-performing campaigns. Document the prompts that produce useful answers in a shared library, because Ask Advisor reward-shapes to specific question patterns and tribal knowledge inside the team compounds faster than ad-hoc usage.
Meridian inside Google Analytics 360 Open-source Marketing Mix Modelling brought into GA 360. The MMM toolchain lands inside the analytics product rather than as a separate workflow. If MMM is already in place, plan the consolidation. If not, evaluate Meridian against the existing measurement stack before committing.
YouTube Demand Gen with Gemini Gemini-powered creator discovery and AI-driven ad formats inside YouTube Demand Gen. Refresh Demand Gen audience seeds. Test Gemini-generated creative variants alongside human-led creative. Compare cost per qualified outcome, not view rate.

The grouping logic: I/O 2026 reshaped the buyer-facing surface (Search box, AI Mode, Universal Cart, Personal Intelligence). Google Marketing Live 2026 reshaped the operator surface (AI Max verticals, Ask Advisor, Meridian, Asset Studio multimodal). Both are 2026 work, not 2027 work.

Why this resets your visibility surface

Three things shifted at the protocol level.

Query fan-out. AI Mode breaks one user question into many sub-queries against the open web in parallel. Deep Search can issue hundreds of sub-queries per session. The implication: a brand surfaces only if its schema, content extractability, and entity graph hold up across the full fan-out, beyond the literal user query. A page that ranks #1 for the headline query but does not have FAQPage breadth, primary citations, and a named-author byline can be skipped on the sub-queries the engine actually answers from.

Multimodality is now load-bearing. One in six United States searches uses voice or images. Image searches grew about 40 percent month over month over the last reporting period. ImageObject schema, VideoObject schema, video transcripts, and caption density on every primary asset all move from optional to required. The brand whose imagery is undescribed loses the multimodal-input share by default.

Conversational follow-up changes the buyer journey. Users can now ask a follow-up question from an AI Overview and flow into AI Mode without losing context. The funnel from initial query to conversational depth is one continuous experience, not two surfaces. Pages that hold context across the sub-questions a buyer is likely to ask next (BLUF + supporting chunks + an FAQ block) get cited in the conversational continuation; pages that answer the headline question only do not.

Universal Cart, UCP, and AP2: checkout moves into Search

Universal Cart is the protocol layer for agentic commerce. It runs across Search, Gemini, YouTube, and Gmail, finds deals and price drops in the background, flags incompatibilities, and accepts checkout with Google Pay in a few taps. The brand stays the merchant of record. Order data, customer record, fulfilment, and post-purchase service remain with the merchant.

For an ecommerce buyer, three practical changes follow.

  1. UCP-compatible product feed. Universal Cart eligibility requires the catalogue to speak UCP. Real-time availability, accurate pricing, and order-status callbacks all need to be wired in. The merchants on the launch list are already there. Brands without UCP-ready feeds will sit outside the cart layer while their direct competitors are inside it.
  2. Attribution will undercount Universal Cart orders. Checkout-on-Google means a portion of orders never touch your site checkout. Existing GA4 and platform-tagged attribution will miss those orders unless the order-status callback is wired into GA4, Google Ads Enhanced Conversions, and Meta Conversions API. The fix is a measurement-stack audit before the traffic arrives, not after.
  3. AP2 changes the dispute path. Tamper-proof digital mandates produce a shared digital record between user, merchant, and processor. Returns and disputes operate on that shared record, not on two separate logs. For merchants who run high chargeback or return rates, this is a measurable operational gain. For finance leaders, it is also a new compliance surface that needs to sit inside your existing KYC and AML workflow.

Geographic rollout: Universal Cart launches in the United States in summer 2026, with YouTube and Gmail to follow. UCP-powered checkout expands to Canada and Australia in the coming months and then the United Kingdom. APAC ecommerce buyers should treat this as a 12-to-18-month readiness window.

Personal Intelligence: the new in-market signal

Personal Intelligence lets AI Mode ground answers in the user's own context: Gmail, Google Photos, and (soon) Google Calendar. Users opt in, can disconnect at any time, and Google does not train on the connected Gmail or Photos data directly (only on prompts and model responses, per Google's own disclosure).

For a brand, the in-market signal layer changed. The brands that show up in personalised AI Mode answers are the ones already present in the user's own context: receipts in Gmail, addresses on Google Maps, content in Google Photos, calendar invitations referencing the brand. Generic interest-based targeting weakens. Owned-data presence wins.

Three operational consequences.

  • Email receipts are now a SEO surface. Order confirmations, booking acknowledgements, and subscription receipts in Gmail are now ground-truth signal for AI Mode's personalised answers. Email-template quality (clean structured data, complete merchant identity, calendar entries) materially affects whether an agent recommends the brand back to its existing user.
  • Google Business Profile depth matters more. GBP completeness (categories, hours, photos, reviews, booking endpoint, products, services) feeds the personalised local-search layer that AI Mode pulls from when a user with Personal Intelligence asks about a place near them.
  • First-party data hygiene compounds. Newsletter subscription quality, transactional email coverage, and CRM-to-platform feedback loops now have an upstream visibility benefit on top of the downstream retention benefit.

Information agents: the new push surface

Information agents run 24/7 on a user-defined query. The example Google gave was an apartment-hunting agent that monitors listings against a user's exact requirements. The pattern generalises: any buyer can stand up an agent for a category, a price band, an event, a regulatory deadline, or a product drop.

For brands, this is the new push surface. Pages that summarise cleanly (declarative answers, dated information, named-author bylines, primary citations) get re-surfaced repeatedly to users whose agent is watching the category. Pages that need to be argued with do not.

What works for information agents:

  • Named-author bylines on every published page
  • Visible publish date and dateModified in JSON-LD
  • Primary-source citations with dated publisher names inline
  • Declarative claim statements that an agent can lift verbatim
  • Stable URLs (information agents revisit; broken URL chains kill the resurface)

What does not work: hedge-heavy copy, marketing-claim language without sources, undated content, anonymous publishing. Pages with those traits do not give the agent a clean summary to push back to the user.

The 90-day priority list

Sequenced so the first month closes technical eligibility gaps and the next two earn share.

Days 0 to 30: technical eligibility.

  1. Audit citation share across AI Mode, ChatGPT, Claude, Gemini, Perplexity, and Bing Copilot on a 30-prompt buyer-intent query set. Note where you appear, where you do not, and which competitors are cited in your place.
  2. Ship schema architecture: @graph with cross-referenced @id covering Organization, WebSite, WebPage, Person (author), BreadcrumbList, and the page-specific entity (BlogPosting, TechArticle, Service, FAQPage, ImageObject, VideoObject, HowTo).
  3. Update robots.txt with an explicit allow list for inference bots (OAI-SearchBot, PerplexityBot, Claude-Web, ChatGPT-User, GoogleOther). Block training bots only if that is your policy. The mistake we audit out of most accounts: a single blanket disallow that blocks the inference crawlers that decide whether AI Mode cites you.
  4. Ship llms.txt at root with brand identity, key pages, editorial policy, and compliance posture.

Days 30 to 60: entity and commerce.

  1. Entity graph: register on Wikidata, claim Crunchbase, complete Google Business Profile depth, build the sameAs ladder in your Organization schema.
  2. Universal Cart readiness: if you are an ecommerce brand, audit the product feed for UCP eligibility, wire order-status callbacks to GA4 and platform measurement APIs, plan the Google Pay merchant connection.
  3. Image and video schema: ImageObject on every primary visual with caption and alt, VideoObject with transcript on every video asset.

Days 60 to 90: earn share.

  1. Primary-cited content on the highest-priority unbranded queries: statistic density, named-expert quotation, outbound citation, self-contained 50-to-150-word chunks. Princeton GEO research shows statistic addition lifts AI impressions about 37 percent, expert quotation about 27 percent, outbound citation about 22 percent (arXiv:2311.09735).
  2. Brand-mention campaigns into named publishers. Ahrefs (March 2026) studied 75,000 brands and found brand-mention correlation with AI Overview visibility at 0.664 versus 0.218 for backlinks.
  3. Citation-tracking baseline: lock in the post-implementation baseline across the same 30-prompt set. Re-poll every 30 days.

Three lenses: B2C, B2B, regulated

For B2C brands. Universal Cart is the headline. If you sell physical goods, your catalogue's UCP readiness will determine which of your competitors gets bought from inside Search and Gemini in 2026 to 2027. The measurement audit (order callback to GA4 and Conversions APIs) is the first thing to ship.

For B2B brands. AI Mode + information agents are the headline. B2B buyers are running longer, more complex queries (AI Mode queries average three times the length of traditional Search) and a meaningful share are setting up information agents on category-watch. Schema breadth, FAQPage depth, named-author bylines, and primary-cited content are the load-bearing investments. Universal Cart is mostly not your problem unless you sell self-serve.

For regulated brands. Personal Intelligence is the surface that needs a compliance read. Personal Data Protection Act in Singapore, Australian Privacy Act, Canadian Personal Information Protection and Electronic Documents Act, and GDPR in the EU all reach the data inputs to Personal Intelligence. Consent records need to reflect that an agent acting on a user's behalf is a processor relationship. Universal Cart and AP2 keep the brand as merchant of record, so existing know-your-customer and anti-money-laundering processes attach to the order as normal. The work is wiring the audit trail through the new rail, not avoiding the rail.

Across all three: the 2024 marketing playbook covered the spend-side decisions. The 2026 marketing playbook needs to cover protocol eligibility first, then spend. The order matters.

Questions, answered.

What is AI Mode and how is it different from AI Overviews?

AI Overviews is the AI-generated summary at the top of a traditional Search results page. AI Mode is the end-to-end conversational Search surface that uses query fan-out and Deep Search to issue many parallel sub-queries against the open web, then synthesises an answer with cited sources.

As of Google I/O 2026 (19 May 2026), AI Mode runs on Gemini 3.5 Flash by default and crossed one billion monthly active users.

What is Gemini 3.5 Flash and what does it change for Search?

Gemini 3.5 Flash is Google's newest Flash model, announced 19 May 2026 as the default model for AI Mode globally. It runs the query fan-out, agentic-coding, and generative-UI capabilities inside Search.

For marketing teams, the practical change is that single user queries now expand into many parallel sub-queries. Your site's schema, content extractability, and entity graph need to hold up across the full fan-out, beyond the literal user query.

What is the Universal Commerce Protocol (UCP) and do we need to support it?

UCP is the open standard Google and retail partners co-developed earlier in 2026 for agentic commerce. It standardises how product data, availability, pricing, and order information move between an agent, a merchant, and a payment processor.

If you want your catalogue inside Universal Cart (Search, Gemini app, soon YouTube and Gmail), it needs UCP-compatible feeds. Launch partners include Nike, Sephora, Target, Ulta Beauty, Walmart, Wayfair, and Shopify merchants such as Fenty and Steve Madden.

What is the Agent Payments Protocol (AP2)?

AP2 is the payment-rail protocol Google launched so agents can make secure purchases on behalf of a user. It uses tamper-proof digital mandates, a verifiable link between user, merchant, and payment processor, and privacy-preserving identity. AP2 is rolling into Google products in the coming months, starting with Gemini Spark.

For merchants, AP2 changes the dispute-and-return workflow: user, merchant, and processor all see the same digital record.

How do information agents differ from regular search crawlers?

Regular search crawlers index pages once and re-crawl on a schedule. Information agents (announced at I/O 2026, launching summer 2026 for Google AI Pro and Ultra subscribers first) monitor the open web 24/7 against a user-defined query and surface a synthesised update when something relevant changes.

For brands, this is a new push surface. Pages that summarise cleanly (declarative answers, named author, dateModified, primary citations) get re-surfaced. Pages that need to be argued with do not.

Does Personal Intelligence work outside the United States?

Yes. As of 19 May 2026, Personal Intelligence in AI Mode is rolling out to nearly 200 countries and territories across 98 languages with no subscription required. Users can securely connect Gmail and Google Photos, with Google Calendar coming.

For brands, the practical implication is that generic in-market signals weaken and brands already present in a user's inbox (receipts, past orders, newsletter subscriptions) and Google Business Profile become channel inputs.

What schema changes should marketing teams make this quarter?

Five priorities.

  1. FAQPage schema on every page with buyer-intent questions, with visible HTML matched 1:1.
  2. ImageObject schema with caption and alt text on every primary image.
  3. VideoObject schema with transcript on every video asset.
  4. HowTo schema on any process content with structured steps.
  5. @graph cross-referenced @id linking Organization, WebSite, WebPage, Person (author), BreadcrumbList, and the page-specific entity.
Should we still invest in traditional SEO?

Yes. Traditional SEO still drives blue-link traffic, branded search, and the surface that AI Mode crawls. The shift is in the share of work, not the existence of the work.

Schema, entity graph, content extractability, and primary citations now matter more than keyword density or link velocity. Treat traditional SEO and AI visibility optimisation as one programme with two measurement layers: ranking share for blue-link queries, citation share for AI-engine queries.

What is the 90-day priority list for marketing teams?

Six items, sequenced. Audit citation share across six AI engines, ship @graph schema, allow inference bots in robots.txt, ship llms.txt at root, wire order callbacks for any ecommerce surface that may touch Universal Cart or AP2, and register on Wikidata, Crunchbase, and Google Business Profile.

Detail in the section above.

How do regulated industries handle agentic checkout and Personal Intelligence?

With the same disclosures and consent flows the human-checkout path already carries. For Singapore financial services, the Personal Data Protection Act applies to data inputs to Personal Intelligence; consent records must reflect that an agent acting on a user's behalf is still a processor relationship. The Australian Privacy Act and the Canadian Personal Information Protection and Electronic Documents Act apply the same logic.

Universal Cart and AP2 keep the brand as merchant of record, so existing know-your-customer and anti-money-laundering processes attach to the order as normal. The work is wiring the audit trail through the new rail.

Read next