It’s been twenty-five years since Google redesigned its search box. Last Tuesday, it did it again — and the changes go a lot deeper than a new coat of paint.
At Google I/O 2026, the company unveiled what it calls the biggest transformation in Search since the product launched. Not AI Overviews tweaked, not a new tab bolted on — a fundamental rethinking of what search actually is. Here’s what happened, what it means for how people find things on the internet, and — critically — what it means for anyone who depends on search traffic for their business.
Google I/O ’26 Keynote — full replay on YouTube. The Search announcements begin around the 43-minute mark.
01
The search box gets its first real redesign
For a quarter of a century, the Google search box has been a white rectangle. You typed some words, pressed Enter, got a list of links. Simple, familiar, unchanging.
That’s over now.
The new “intelligent search box” — powered under the hood by Gemini 3.5 Flash — expands dynamically as you type, accommodating longer, more conversational queries without the awkward horizontal scrolling of old. But the bigger story is what’s now inside the box: an AI-powered suggestion system that goes well beyond autocomplete, actively reading your intent and helping you shape better questions.
More significantly, there’s a new + button on the left side that turns the search box into a genuine multi-modal input. Drag a Chrome tab into it and ask a question about the page you’re reading. Drop in an image, a PDF, a video — the system reasons across all of them simultaneously.
Google’s reimagined search box expands to fit conversational queries. The + button (left) accepts files, images, and Chrome tabs. Source: Google Blog
“This new search box puts our most powerful AI tools right at your fingertips, and you can ask across modalities with text, images, files, videos — and Search reasons across them all.”
— Liz Reid, VP of Google Search, Inc. Magazine, 19 May 2026
The conversation also carries forward now. Ask a follow-up directly from an AI Overview and flow seamlessly into AI Mode. Your context persists, the links become more relevant as you explore, and the experience starts to feel less like a tool and more like a knowledgeable colleague you’re thinking alongside.
AI suggestions, memory, multimodal input — these aren’t features bolted onto the old search box. They’re a new paradigm: keyword search is becoming conversational search, and the interface has finally caught up.
Further reading: Google Search is getting a complete AI overhaul — here’s what’s actually changing (MakeUseOf) · Google Search as you know it is over (TechCrunch)
02
Agents move into the search bar
The other enormous announcement: Google is wiring autonomous AI agents directly into Search — no separate app, no API key, no switching context. Just ask.
The headline feature is information agents: persistent, background AI systems that continuously scan the web — news sites, blogs, social posts, real-time finance and sports data — and surface updates the moment something relevant changes. Think of Google Alerts, rebuilt with a frontier language model’s capacity for nuance and inference. Set an agent to monitor a competitor’s pricing, a regulatory development, or a niche hobby market. It watches, reasons, and alerts you.
These agents will launch this summer for Google AI Pro and Ultra subscribers in the US — so they’re initially gated — but the direction is clear: Search is evolving from a place you visit into a system that proactively works for you.
Information agents are in many ways the natural evolution of search intent. The traditional model asks: “what does the user want right now?” Agents ask: “what does the user want to know about, continuously, at the moment it matters?” That’s a fundamentally different product.
Beyond information agents, Google also announced the ability to create, customize, and manage multiple AI agents for different tasks, all from within Search. The era of the general-purpose search is giving way to a portfolio of specialised assistants — your travel researcher, your market watcher, your supplier tracker — all living inside the same interface you’ve used for decades.
Further reading: Google Search’s I/O 2026 updates: AI agents and more (Google Blog) · Google replaces the search box with AI agents at I/O 2026 (The Next Web)
03
Generative UI: Search builds its own interface
Perhaps the most technically striking announcement: Search can now generate its own custom user interfaces on the fly, in response to your query.
Using Google’s Antigravity platform and the agentic coding capabilities of Gemini 3.5 Flash, Search can build interactive widgets, data visualisations, simulations, and mini-apps — tailored precisely to what you asked. Ask how the planets orbit the sun and you might get an interactive solar system. Ask about planning a wedding and Search could generate a personalised tracker dashboard you can return to over weeks.
Google calls this “generative UI” and it represents a meaningful shift in what a search result even is. It’s no longer just a list of links to resources that might have what you need. It can be a bespoke interactive experience built for this exact question, this exact person, this exact moment.
Generative UI will be free for all users when it rolls out this summer. The more capable agentic mini-apps — which can pull in live maps, weather, reviews, and external data — will be restricted to AI Pro and Ultra subscribers initially.
The implications for publishers are significant and uncomfortable. If Search can synthesise information and build interactive tools directly, the incentive to click through to a source website diminishes further. This is the trajectory that has defined the last two years — and it just accelerated.
Further reading: Google reimagines search with AI agents and generative interfaces (SiliconAngle) · Beyond blue links: Google redesigns Search around AI agents and Gemini (Interesting Engineering)
04
The Universal Cart: Wheeling your trolley across the internet
The most commercially consequential announcement — and the one that will reshape e-commerce — is the Universal Cart.
Imagine pushing a shopping trolley around a giant physical market. You pick up a pair of trainers from one stall, a kitchen gadget from another, a book from a third. The trolley doesn’t care who you bought each item from — it just holds everything in one place while you browse. You pay at a single till on the way out.
That’s exactly what Google’s Universal Cart does, except the market is the entire internet and the trolley follows you across every surface Google touches.
You can add items while browsing Search results, chatting with Gemini, watching YouTube, or reading Gmail. The cart persists across all of them. It spans multiple merchants simultaneously — products from Amazon, Walmart, Nike, and independent retailers sitting side by side in one unified view. And when you’re ready to check out, the whole process goes through Google, with Google Wallet recommending the best payment method for rewards and savings.
Under the bonnet: Universal Cart is built on two pieces of infrastructure Google has been assembling. The Universal Commerce Protocol (UCP) is an open standard that lets AI agents speak to merchant systems in a common language. AP2 is Google’s payment protocol that lets agents make purchases autonomously within user-defined spending limits — specific brands, specific price caps, specific conditions. When those conditions are met, the agent completes the transaction without you touching it.
Crucially, the cart is intelligent, not just a container. It tracks price histories so you can see whether a “discount” is real. It monitors stock and alerts you when an out-of-stock item returns. It flags when the items in your cart don’t quite match what you actually need — and suggests better alternatives. Gemini can recommend compatible components when you’re assembling something complex, like a PC build.
UCP will expand from the US to Canada and Australia in the coming months, with the UK to follow. For retailers, the integration question is pressing: those that speak UCP’s machine-readable language will be visible to AI-powered shopping flows. Those that don’t may increasingly find themselves outside the funnel entirely.
For consumers, it’s genuinely transformative. The friction of maintaining a dozen different accounts, comparing prices across tabs, managing separate carts on every website — that’s being eliminated. One cart, everywhere, with an AI doing the comparison work for you.
Further reading: Google’s new Universal Cart wants to follow your entire shopping journey across the internet (TechCrunch) · Google is transforming your online shopping cart into something much smarter (Android Authority) · Google Announces New Universal Cart At I/O (Search Engine Journal)
05
Google's AI search guide just blew up 18 months of GEO advice
On 15 May, Google quietly published its first official AI Search Optimisation guide. For anyone who’s been following the “Generative Engine Optimisation” space on LinkedIn, it makes for uncomfortable reading.
The guide contradicts, point by point, a significant portion of what’s been sold as expert advice for the past year and a half. Here’s what Google actually said:
| What GEO ‘experts’ said | What Google actually found |
|---|---|
| Create an llms.txt file — it’s like a sitemap for AI | A study of 300,000 domains found zero correlation between llms.txt files and LLM citation rates. Google’s crawler treats it like any other text file. Decorative. |
| AI Overviews use a separate AI index — optimise differently | AI Overviews run on the regular Search index. Same crawl, same ranking signals, same E-E-A-T. RAG pulls from what’s already indexed. No separate layer to game. |
| Target as many “AI query variants” as possible with individual pages | Creating pages targeting every fan-out variation is now an explicit spam policy violation. Not clever. Spam. |
| Top 10 rankings are what matter for AI citations | 68% of pages cited in AI Mode rank outside the organic top 10. The system pulls strong passages, not top-ranked pages. A genuinely useful section at position 30 beats a thin guide at position 3. |
The “query fan-out” finding is particularly worth sitting with. When someone asks a complex question, AI Mode breaks it into dozens of sub-queries and runs them simultaneously. That means a single, genuinely deep page — one that covers a topic properly from multiple angles — has a far better chance of being cited than fifty thin pages each targeting a keyword variant. Depth wins. Coverage wins. Quality at the passage level is what’s being rewarded.
What actually works in 2026, per Google’s own guide:
- One genuinely useful page beats fifty thin ones
- Passage-level quality, not page-level keyword stuffing
- E-E-A-T signals — experience, expertise, authoritativeness, trustworthiness — still matter enormously
- Write for humans. Crawl-friendly HTML. That’s it.
Simultaneously with the guide’s release, Google updated their spam policy to explicitly cover AI manipulation tactics. The message couldn’t be clearer: trying to game the AI layer with the same tactics that gamed the organic layer won’t work — and now actively risks a penalty.
Official sources: Google’s AI Search Optimization guide · Google Spam Policies (updated May 15, 2026) · How Google Search works
06
What this means for your traffic — and how to think about it
Here’s the uncomfortable truth that no amount of reframing changes: the new features announced at I/O 2026 will cut into organic click-through rates. Generative UI means fewer reasons to leave Google. Information agents mean some queries never get asked at all. The Universal Cart means purchase journeys happen inside Google’s ecosystem.
And yet — note this carefully — AI Mode is not becoming the default in Google Search. The standard results page isn’t being replaced. Traditional blue links still exist. For the foreseeable future, you’re dealing with a blended environment: AI Overviews, AI Mode, and conventional results coexisting. The ratios will shift, but the mix remains.
Users who want the old-fashioned list of links can still get it: clicking the “Web” tab that appears beneath the search bar filters out AI Overviews entirely and returns a clean page of traditional organic results. It’s a manual step — you have to click it on every search — and Google isn’t exactly advertising it. But it exists, and it matters. The platform is architected as a spectrum, from fully traditional to fully agentic, with users choosing where they sit on it. For now.
That blended reality demands a blended measurement approach. If you’re still evaluating search performance purely at the channel level — organic traffic, ranking positions — you’re getting an increasingly inaccurate picture. Someone who gets their answer from an AI Overview and never clicks has still been influenced. Someone who sees your brand cited in AI Mode but converts via direct later has still been touched by your search presence.
The frame needs to expand: look at brand search volume, look at assisted conversions, look at blended share of search, look at citation frequency in AI-generated answers.
From Google Search’s perspective, optimising for generative AI search is optimising for the search experience — and thus still SEO.
The fundamentals haven’t changed. Make genuinely excellent content. Be authoritative. Go deep on topics you know. Structure your pages so they’re easy to crawl and passage-level quality shines through. E-E-A-T still prints.
What has changed is the definition of success. “Ranking #1” matters less than “being cited in the AI answer.” “Organic traffic” matters less than “search-influenced revenue.” And the tactics that many consultants have been charging for — llms.txt files, separate AI index strategies, fan-out keyword farming — have been officially called out as either useless or counterproductive.
The playbook is simpler and harder than it’s ever been: be the best answer to the question, wherever Google looks for it.
