Google spent 27 years training the world to type a few words into a box and scan ten blue links. That era is now officially over. At Google I/O 2026 on May 19, the company made Gemini 3.5 Flash the default model behind AI Mode, a feature that had rocketed from roughly 75 million daily users less than a year earlier to the milestone Alphabet confirmed that June: more than 1 billion monthly users. Similarweb’s own traffic data told a parallel story, recording 279 million visits to AI Mode in May 2026 alone, more than double the 126 million visits the feature logged in May 2025. By July 10, according to trade outlet TechTimes, the shift had gone further still: nearly every query returning an AI-written answer as the primary result, with ranked links pushed below the fold or gone from the page entirely. Google’s head of Search, Liz Reid, called it the biggest change to the search box in more than 25 years, and neither of the two words in “search engine” quite describes what Google Search does now.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
Google Confirms the Switch: Gemini 3.5 Flash Now Runs AI Mode Search
The official announcement was narrower than the headlines that followed it. Google’s own Search blog post from I/O 2026 states that Gemini 3.5 Flash became “the new default model in AI Mode for everyone globally” starting May 19, a claim Search Engine Journal’s own May 2026 reporting independently corroborated. That is a specific, bounded claim: one feature, one model, rolled out worldwide. Google framed Gemini 3.5 Flash as a model built for “sustained frontier performance for agents and coding,” deployed simultaneously across the Gemini app, Search, the Antigravity 2.0 coding environment, and the Gemini API.
What makes this different from Google’s earlier AI Overviews push is scale. AI Mode, the company said in its I/O 2026 recap, had already surpassed 1 billion monthly users roughly one year after the feature’s debut and before the Gemini 3.5 Flash swap, with query volume “more than doubling every quarter since launch.” Alphabet’s own February 2026 disclosures filled in the behavioral picture: daily AI Mode queries per US user had doubled since launch, those queries ran roughly three times longer than a traditional search, and nearly one in six now arrived as voice or image input rather than typed text. The trajectory was visible early: data from SE Ranking, Profound and Datos, summarized by Memeburn, showed AI Mode had already topped 100 million monthly active users across just the US and India by July 2025, barely two months after its wider US rollout. Google also said total Search query volume hit an all-time high in the quarter before I/O. Growth in AI-generated answers is not cannibalizing search usage. It is riding alongside it, and increasingly replacing what used to be a results page.
The Numbers: 1 Billion Users and a Rollout Timeline
Tracking exactly what happened and when matters here, because Google’s framing and the trade press’s framing diverge in an important way. The table below lays out the confirmed public timeline from Google’s own posts alongside independent reporting.
| Date | Milestone | Source |
|---|---|---|
| May 19-20, 2026 | Google I/O 2026; Gemini 3.5 Flash named default model in AI Mode, globally | Google Search blog |
| May 19, 2026 | Liz Reid calls it the biggest search box change in over 25 years | AP News, TechTimes |
| May 19, 2026 | “Google Changes Its Search Box for the First Time in 25 Years” headline runs | New York Times |
| Summer 2026 (rolling) | Information agents begin rolling out to Google AI Pro and Ultra subscribers | Sundar Pichai I/O post |
| Summer 2026 (rolling) | Generative UI in Search opens to all users at no charge | Google blog |
| July 10, 2026 | Reported completion of AI-first rollout; blue links no longer the default output | TechTimes |
| July 21, 2026 | Gemini 3.6 Flash and Gemini 3.5 Flash-Lite added to Search’s model lineup | Industry reporting |
Notice the gap between May and July. Google announced a model swap inside a specific tab in May. Independent reporting in July described something closer to a wholesale replacement of the classic results page. Both things can be true at once: a feature that starts inside AI Mode can expand to cover most queries within eight weeks once a company decides to push it there, especially once it is already handling a billion users a month.
From Ten Blue Links to AI Overviews to AI Mode
None of this happened overnight, and the shift did not start in 2026. Google has been walking toward an AI-first results page for several years, in three distinct phases.
The AI Overviews Era (2024)
Google’s Search Generative Experience, an opt-in experiment, graduated into AI Overviews in 2024, placing an AI-written summary above the traditional link list for a subset of queries. The coverage climbed fast from there: BrightEdge data cited by Axis Intelligence found AI Overviews appearing on just 6.49% of tracked queries in January 2025, a share that had grown to roughly 48% by February 2026, and Alphabet CEO Sundar Pichai said AI Overviews itself had passed 2 billion monthly users across more than 200 countries and over 40 languages by Q1 2026, according to figures cited by Arvow. That single change already reshaped how publishers thought about click-through rates, even though the ten-blue-links format sat untouched just below it.
AI Mode’s Build-Up (2025 into early 2026)
AI Mode arrived quietly, first as a Search Labs experiment Google opened to US users running a Gemini 2.5 model in March 2025, then expanded to every US Search user without a Labs sign-up on May 20, 2025, the same day Google folded a custom-tuned Gemini 2.5 model into both AI Mode and AI Overviews at once. Growth compounded quietly from there until Google disclosed the billion-user figure at I/O 2026. By the time Gemini 3.5 Flash took over as the default model, AI Mode was no longer a side experiment. It was already a mainstream habit for a meaningful share of Google’s user base.
The July 2026 rollout is best read as the point where that separate tab stopped being separate. Instead of choosing between “Search” and “AI Mode,” users increasingly get the AI Mode-style answer whether they ask for it or not.
What Actually Changed on the Results Page
According to AP News’s coverage of the I/O 2026 keynote, the redesigned search box now accepts text, images, video, files, and even dragged-in Chrome tabs as input, not just typed keywords. The Verge’s rollout coverage described the new box as shipping worldwide on both desktop and mobile starting the Tuesday after the announcement, in every region and language where AI Mode was already active. That international footprint was already substantial well before the July rollout: Google says AI Mode had expanded to 200-plus countries and territories and added more than 35 new languages by October 2025. Language support kept expanding from there – as of August 2026, Google says AI Mode covers nearly 200 countries and territories across 98 languages total, with the feature remaining free to use and carrying no subscription requirement.
TechTimes’ July 13 report is the source for the most dramatic claim in this story: that as of July 10, every query returns an AI-generated answer as its primary output, composed in real time by Gemini 3.5 Flash, with ranked links demoted below the fold “when they appear at all.” That is a stronger claim than anything in Google’s own blog posts, which describe Gemini 3.5 Flash as the default model “in AI Mode” rather than announcing the end of the classic results page across all of Search. Readers should treat the July 10 “completion” date as trade-press reporting on the scope of the rollout, not a verbatim Google announcement, even though Google is quoted in that same reporting confirming Gemini 3.5 Flash now powers search broadly, not just the AI Mode tab.
Information Agents and the “Intelligent Search Box”
The feature Google leaned on hardest in its own messaging is not the answer format. It is “information agents,” described in Sundar Pichai’s I/O 2026 post as personalized AI agents that run in the background around the clock, watch for the information a user cares about, and surface updates with links attached. Google said these agents would begin rolling out through the summer, starting with subscribers to its paid AI Pro and Ultra tiers before reaching free users later.
Generative UI is the second half of the pitch: instead of a static page of text snippets, Search can assemble interactive modules, comparison layouts, or step-by-step walkthroughs on the fly, depending on the query. Google said this capability would open to everyone, free, over the summer. Combined with information agents, the goal Google is describing is a search box that behaves less like an index and more like a standing assistant that checks in on its own schedule.
Google’s Line: “Best of the Web With the Best of AI”
Google has been careful, in its own language, to avoid saying it killed the open web. Its May 2026 AI updates post describes the goal as combining “the best of the web with the best of AI,” and specifically frames information agents as tools that “monitor information on your behalf and send detailed updates with links to dive deeper.” Links, in Google’s own wording, are still part of the product.
That framing sits awkwardly next to the July reporting that blue links now show up “below the fold, when they appear at all.” Both descriptions can be accurate for different query types. A recipe search or a factual lookup may get a self-contained AI answer with no obvious need to click through. A complex research task, or anything an information agent is tracking over time, may still surface a stack of source links. The practical effect for most simple queries, though, is fewer reasons to leave Google’s page at all, which is exactly the trend publishers have been warning about since AI Overviews launched in 2024.
The SEO and Publisher Industry Reckoning
The SEO industry’s read on this rollout has been blunt. Independent analysis from Passionfruit, an SEO and search analytics outlet that published a detailed breakdown of the I/O 2026 announcements, framed the takeaway for site operators around one idea: showing up inside AI Mode’s answers matters more now than ranking in a link list that fewer users will scroll down to. That outlet’s coverage ran to roughly 3,500 words with its own FAQ section, itself a sign of how much operational detail publishers are trying to absorb about a single product change.
What is missing, as of this writing, is hard third-party measurement. No Similarweb, Pew Research, Chartbeat, or Digital Content Next study quantifying the referral-traffic hit from the July rollout had been published at the time of writing, and no publisher lawsuit or antitrust complaint specifically targeting AI Mode’s July expansion had surfaced either. That is worth stating plainly rather than guessing at a percentage: the anecdotal alarm across the SEO industry is running well ahead of any independently audited traffic data. Expect that gap to close within a quarter or two, given how closely publishers tracked the 2024 AI Overviews rollout.
The practical advice circulating among SEO practitioners in the days after the rollout has shifted noticeably. Instead of chasing position one in a link list, site owners are being told to focus on structured data, clear sourcing, and content that an AI system can cite and summarize accurately, since citation inside an AI Mode answer may now matter more than ranking below it. That is a real change in what “ranking well” means, even without a hard traffic number attached to it yet.
Search Engine Market Share in 2026
Google’s dominance of search has not moved much despite years of AI-search competitors chasing it. StatCounter’s June 2026 global figures put Google at 89.9% of all search traffic, with the rest of the market split among multiple smaller players.
| Search Engine | Global Market Share (June 2026) | AI Search Feature |
|---|---|---|
| 89.87% | AI Mode (Gemini 3.5 Flash), AI Overviews | |
| Bing | 4.68% | Copilot / Copilot Search |
| Yahoo | 1.28% | Bing-powered backend |
| Yandex | 0.79% | Regional AI assistant features |
| DuckDuckGo | 0.67% | Opt-in AI chat features |
| Baidu | 0.43% | ERNIE-based AI search (China) |
Several independent trackers cited in the same period show Google’s global share drifting in a narrow band between roughly 89% and 91% across 2026 rather than sitting at one fixed number, which is normal for month-to-month StatCounter data. The direction, though, is consistent across every source: a slow, multi-year erosion measured in single percentage points, not a collapse. Google can push its riskiest search-page redesign in a generation and still control roughly nine in every ten searches on the planet. That is the leverage that let it move this fast in the first place.
Competitive Landscape: AI Mode vs. Bing Copilot, Perplexity and ChatGPT Search
Google is not the only company betting search’s future is conversational. Microsoft folded Copilot directly into Bing years ago. Perplexity built its entire product around cited, AI-generated answers instead of a link list. OpenAI added ChatGPT Search as a standing feature rather than a side experiment. What Google’s July rollout changes is the scale of the comparison: none of those competitors touch anywhere close to a billion monthly AI-answer users, let alone Google’s 89.87% share of total query volume.
Our recent comparison of Perplexity, ChatGPT, and Gemini pricing found an enterprise cost gap of roughly $295 between the cheapest and most expensive plans among the three, which underscores a different competitive axis: Google is not really trying to out-price Perplexity or OpenAI for enterprise AI search contracts. It is trying to make the free, default consumer experience so thoroughly AI-native that switching search engines stops feeling like an upgrade. Meanwhile, OpenAI’s own consumer position has been softening on a separate front. Our analysis of ChatGPT’s user base found its share of the consumer AI chatbot market slipping to 46.4% as Gemini, Grok, and other rivals gained ground, a dynamic covered in our ChatGPT market share report. A weaker ChatGPT consumer position makes Google’s search-box overhaul look even more like an attempt to close off the front door before rivals can widen it.
The Gemini Model Family Behind the Shift
Gemini 3.5 Flash is doing the heavy lifting for now, but it was never going to be the only model in Search for long, and the pace of churn since has proven that out. Industry reporting on Google’s July 21 update described three models landing inside Search’s infrastructure around the same time: Gemini 3.6 Flash, Gemini 3.5 Flash, and a new lightweight variant, Gemini 3.5 Flash-Lite, aimed at cheaper, simpler queries where a full frontier model is overkill. Google kept iterating from there: Gemini 3.7 Flash rolled out inside AI Mode for AI Pro and Ultra subscribers in August 2026, and by September 2026 Gemini 3.8 Flash had become a selectable model in AI Mode worldwide for those same paid tiers.
Gemini 3.5 Flash vs. Gemini 3.6 Flash vs. Gemini 3.5 Flash-Lite
The pattern is familiar from every major model family: a flagship Flash model handles the default experience, a newer numbered release (3.6 Flash, which we covered in detail in our Gemini 3.6 Flash launch report) pushes pricing and benchmark scores further, and a Lite variant trims cost and latency for high-volume, low-complexity traffic like simple factual queries. Running three tiers inside one product lets Google route a “what time is the Super Bowl” query to the cheapest possible model while reserving the priciest compute for multi-step research tasks. Some mid-July reporting also pointed to an internal target date of July 17 for a separate, more powerful Gemini 3.5 Pro release, though Google had not confirmed that date publicly, and it should be treated as unconfirmed until Google says otherwise.
Market Impact: Advertisers, Publishers and Alphabet’s Bottom Line
Alphabet has not broken out AI Mode’s specific effect on advertising revenue in isolation, but the top-line trend undercuts any fear that AI answers are starving Google’s core business: Search revenue grew 17% year-over-year to $63.07 billion in the fourth quarter of 2025, a figure Alphabet disclosed in February 2026, even as AI Mode usage climbed. What can be said with more confidence is structural. Search advertising has always depended on a results page with room for both organic links and paid placements above them. An answer-first page compresses that visual real estate, and Google will have to decide how ads sit alongside an AI Mode-generated answer without making the product feel cluttered again in the way that drove some users toward AI Mode in the first place.
For publishers, the calculation is more immediate. A site that depended on informational, “what is” or “how to” style queries for traffic is exactly the category of query AI Mode handles best without a click-through. Sites built around transactional or highly specific technical queries, where an AI summary is a poor substitute for the source, are better insulated for now. That split is likely to accelerate consolidation in the publishing industry around exactly the kind of content Google’s models cannot easily answer on their own.
Investors have not punished Alphabet for any of this so far, in part because there is no competing default search product anywhere near large enough to make switching costly for Google. That is the paradox of this rollout: a change dramatic enough to worry an entire publishing industry barely registers as a risk to Google’s own business, precisely because the market share table above shows just how little competitive pressure Google is actually under.
Where This Leaves the Broader AI Model Race
Search is the highest-traffic surface any AI lab controls, and Google just used it to put a Gemini model in front of more daily users than any single chatbot app can claim. That matters for the wider frontier-model competition covered across our ongoing model comparisons, including Sonnet 5 vs. GPT-5.6 vs. Gemini 3.1 Pro and Grok 4.5 vs. GPT-5.6 vs. Gemini 3.1 Pro. Anthropic, OpenAI and xAI are all fighting for developers and enterprise seats one contract at a time. Google is fighting for consumer search habits it already mostly owns, which is a very different, and arguably easier, battle to win. OpenAI’s own retirement of four legacy GPT models this year shows a company trying to simplify its lineup under competitive pressure, while Google is doing the opposite: stacking more Gemini variants into one product surface to cover every price point at once.
What Google and Industry Watchers Are Saying
Beyond Liz Reid’s characterization of the update as Search’s biggest change in a quarter-century, Google’s own written statements have stayed carefully measured, repeatedly describing the goal as blending web and AI results rather than eliminating the web from Search. The New York Times and Wired both framed the I/O announcement as Google’s most significant search redesign in decades, while The Verge emphasized the simultaneous worldwide rollout across desktop and mobile as unusual for a change this large. None of the coverage reviewed for this piece includes an on-record statement from a named Google executive disputing the more dramatic July framing of blue links disappearing, which leaves a real gap between the company’s cautious public language in May and the sweeping description of the July rollout in trade press.
5 Predictions for Search, SEO and AI Models Through 2027
- Hard traffic data becomes the story. Once Similarweb, Chartbeat or Pew Research publish audited referral-decline numbers tied specifically to the July rollout, expect that single data point to dominate coverage more than the rollout itself did.
- More Gemini tiers, not fewer. Look for Google to keep adding narrower model variants (Flash-Lite for simple queries, a higher-end Pro tier for agentic research) rather than settling on one model to power all of Search.
- Rivals chase publisher partnerships. Bing, Perplexity and ChatGPT Search are likely to lean harder into licensing and citation deals with publishers to differentiate themselves as a friendlier alternative to Google’s AI Mode.
- Regulators take notice. Existing antitrust scrutiny of Google’s search dominance in the US and EU is likely to expand to explicitly cover default AI-generated answers, not just link ranking.
- Publishers diversify faster. Sites reliant on informational search traffic will accelerate the move toward newsletters, apps and direct AI-licensing deals rather than betting on organic search referrals recovering.
Frequently Asked Questions
What is Google AI Mode?
AI Mode is a Search feature that answers queries with an AI-generated response built for longer, conversational, multimodal questions, rather than a list of ranked links. Google said it had surpassed 1 billion monthly users as of its May 2026 I/O announcement.
Did Google actually remove all blue links from search results?
Google’s own statements describe Gemini 3.5 Flash as the new default model “in AI Mode,” not a total removal of the classic results page. Trade press reporting from July 2026, notably TechTimes, describes a broader rollout in which ranked links appear below the fold or not at all for many queries. Both can be true depending on the query type.
What is Gemini 3.5 Flash?
Gemini 3.5 Flash is the Google model that became the default engine behind AI Mode and the redesigned search box starting May 19, 2026. Google describes it as built for sustained performance on agentic and coding tasks, and it also runs inside the Gemini app, Antigravity 2.0, and the Gemini API.
When did AI Mode become the default in Google Search?
Google announced Gemini 3.5 Flash as AI Mode’s new global default model on May 19, 2026, at Google I/O. Trade press reported a broader rollout affecting most queries by July 10, 2026.
How many people use Google AI Mode?
Google said AI Mode passed 1 billion monthly users by May 2026, roughly a year after the feature’s debut, with query volume more than doubling every quarter since its launch, according to the company’s own I/O 2026 recap.
What are “information agents” in Google Search?
Information agents are background AI agents, described in Sundar Pichai’s I/O 2026 post, that monitor topics a user cares about around the clock and send updates with links attached. They began rolling out over summer 2026, first to Google AI Pro and Ultra subscribers.
How many countries and languages does Google AI Mode support?
As of August 2026, Google says AI Mode is available in nearly 200 countries and territories across 98 languages, at no subscription cost. Premium features like information agents remain limited to paid AI Pro and Ultra subscribers for now, but core AI Mode access is free and broadly international.
How does Google AI Mode compare to Bing Copilot and ChatGPT Search?
Bing Copilot and ChatGPT Search offer similar AI-generated answers, but neither approaches Google’s scale. StatCounter put Google at 89.87% of global search traffic in 2026, versus 4.68% for Bing, meaning Google’s AI-search rollout reaches a vastly larger existing user base than any single competitor.
What does this mean for SEO and website traffic?
No independently audited study had quantified the July 2026 rollout’s effect on publisher referral traffic at the time of writing. SEO analysts broadly recommend optimizing for visibility inside AI-generated answers rather than only for link rankings, since informational queries are the category most likely to be answered without a click-through.
