Google has started pushing generative UI into AI Overviews, meaning search results can now generate an interactive tool on the spot instead of just linking to one. For any site that earns traffic from an embedded calculator, converter, or simulation, that is a direct competitive threat, and Google’s own research suggests the overlap goes well beyond simple study aids.
What Generative UI Actually Does
Generative UI builds a custom visual interface whenever Google’s model decides one would help answer a query better than plain text. Search “pH scale” and instead of a wall of text, the result might render an interactive scale. Ask a follow-up and the experience shifts into AI Mode with a more specific version — plotting citrus fruits on that same scale, for instance.
Google has framed the rollout around education, adding practice quizzes alongside the interactive visualizations. But the underlying system is not limited to study tools. Google has already demonstrated a mortgage calculator inside AI Mode that lets users compare loan options without leaving the results page — the exact kind of utility a finance site would normally build to attract links and repeat visits.
As of this rollout, generative UI is available worldwide in English inside AI Mode and is beginning to appear in AI Overviews, though Google has not said when the AI Overviews rollout will be complete or which languages beyond English are included.
From A Paid Subscriber Perk To A Global Default
Generative UI first appeared alongside Gemini 3 in November 2025, limited to AI Pro and Ultra subscribers using the Gemini app and AI Mode in the U.S. At I/O 2026 in May, Google said it would become free for everyone in Search that summer. The current expansion — global English in AI Mode, plus a phased introduction into AI Overviews — is the next step in that plan.
The scale involved is worth noting. Sundar Pichai said at I/O that AI Overviews now reaches more than 2.5 billion monthly active users, with AI Mode past 1 billion. The two surfaces behave differently: AI Overviews appear automatically on the standard results page, while AI Mode requires someone to actively switch into it. That distinction matters for click behavior — AI Mode tends to send a different kind of visitor than a standard search result page does, and both surfaces show measurably lower click-through than a page without an AI summary. A Pew Research Center analysis found only 8% of visits that included an AI summary led to a click, against 15% for visits without one. Summaries have already been competing with the pages people read; generative UI now extends that competition to the pages people use.
What Google’s Own Testing Found
The research behind the November launch, a paper titled “Generative UI: LLMs are Effective UI Generators,” reached arXiv in February. Google tested how people respond when a generated interface goes head-to-head with a real website by showing participants five results for the same prompt: a human-built custom site, the top organic search result, a generated interface, and two plain-text LLM outputs. All results were pre-cached, so generation delay was not a factor in participants’ judgments.
On the main prompt set — drawn largely from LMArena, a chatbot evaluation platform, and skewed toward conversational-style requests — participants preferred the generated interface over the top organic search result in 90% of head-to-head comparisons. Custom-built sites held up much better, winning 50% of comparisons against generative UI’s 35.3%, with the remainder rated neutral. On a second, narrower prompt set built specifically around information-seeking queries, generative UI still won 73.5% of the time versus 19% for the top search result, though Google notes the smaller margin likely reflects the difference in prompt type.
It’s worth being precise about what the “custom sites” in this test actually were. Google paid freelance developers between $100 and $130 per site for roughly three to five hours of work, and told them to prioritize user experience over SEO. These were quick builds, not mature, refined tool pages from an established business — the comparison measured generative UI against a rough baseline, not a company’s best work.
Google also reports that each newer model generation performed better on this benchmark: on the LMArena set, the output-error rate dropped from 29% with Gemini 2.0 Flash to zero with Gemini 3, suggesting today’s numbers understate where the gap will sit a year out.
Which Tools Generative UI Can Actually Reproduce
Generative UI assembles a single page with scripts, built from the query plus search and image tools available to the model. That constrains what it can realistically match. Tools built on generic public inputs and standard formulas — mortgage calculators, unit converters, learning simulations — are the easiest targets, and Google has already demoed several of exactly this type.
Tools that depend on proprietary data, logged-in state, or backend integrations the page can’t reach are harder to reproduce on the fly. Google’s paper describes these generated interfaces as “ephemeral” — built fresh for a single prompt rather than stored and reused, standing in for a compressed product, design, and engineering effort compiled in about a minute. Google is explicit that this isn’t as capable as a real team, but the line between “formula-based tool” and “proprietary tool” is already shifting — mini apps and dashboards previewed for subscribers at I/O suggest generated pages are edging toward becoming standing tools rather than one-off answers.
Why This Matters For Site Owners And Agencies
Education sites face the most direct overlap, since Google built the first wave of this feature around interactive explainers, simulations, and practice tests — content that competes head-on with test-prep businesses. Publishers and businesses running tool sections face a broader version of the same problem: a finance site’s payment calculator or a health site’s BMI tool can now be matched by an interface generated inside the results page itself, intercepting the click before it ever reaches the site.
SaaS companies that use free tools as a lead-generation front door are similarly exposed — a free converter or grader that used to be the first touchpoint with a prospective customer can now be replicated at the query stage, before the user ever lands on a branded page.
For agencies, this changes the calculus on tool-building as a content strategy. A calculator built purely to attract links and traffic from generic, publicly available inputs is now competing with something Google can generate instantly and for free. The stronger pitch is a tool built around a client’s proprietary data or workflow — something a generated interface, built only from the query and public tools, cannot reproduce.
The Measurement Gap Nobody Has Solved Yet
Search Console’s Generative AI performance report, which has been rolling out gradually since June, breaks down impressions from AI Overviews and AI Mode by page, country, device, and date — but it does not separate a generated interface from a standard AI Overview, and it does not report clicks at all. That report has already drawn criticism for how easily its numbers get misread, and generative UI adds another layer of ambiguity: if a tool page loses clicks because a generated interface replaced it, that decline will look identical to ordinary AI Overview cannibalization in the data. There has been talk of more dedicated AI search reporting inside Search Console, but nothing yet separates this specific effect out.
Google also hasn’t disclosed which query types trigger generative UI or how often it appears, so site owners can only estimate exposure based on the kind of tools they run. The preference data itself comes from a controlled lab setting, too: raters compared finished, pre-cached pages side by side, with the generation wait — one to two minutes in the research version, roughly halved by streaming — removed from the equation. Whether real-world users are as patient is untested.
The pattern is familiar to anyone who has watched click data shift as AI surfaces expand — the same way AI Overviews have made it harder to isolate exactly where lost clicks are going, or how new ad formats inside AI Mode are reshaping what a “result” even looks like before a user reaches a website at all.
What To Do Before The Rollout Reaches Your Site
Site owners don’t need trigger data to prepare. The useful exercise is an inventory: list every tool on the site, then sort each one into “buildable from public information and public formulas” versus “dependent on data or workflows only this business has.” The first group is squarely in generative UI’s reach today. The second is safer for now, and it’s also the stronger area to invest in going forward, since it’s the one category a generated interface, built only from a search query and public tools, structurally cannot copy.
Frequently Asked Questions
What is Google’s generative UI feature?
Generative UI is a Google Search capability that builds a custom interactive interface — such as a calculator, simulation, or visualization — directly inside AI Mode or AI Overviews, generated on the fly for a specific query instead of linking out to an existing tool page.
Where is generative UI available right now?
It is available worldwide in English inside AI Mode, and is being gradually introduced into AI Overviews. Google has not announced a completion date for the AI Overviews rollout or confirmed additional languages.
Which types of tools are most at risk from generative UI?
Tools built on generic public inputs and standard formulas — mortgage calculators, unit converters, and learning simulations — are the easiest for generative UI to reproduce. Tools that depend on proprietary data, account logins, or backend systems are harder for it to match.
Can Search Console show whether generative UI is affecting my traffic?
Not precisely. The Generative AI performance report in Search Console shows impressions by page, but it does not distinguish a generated interface from a standard AI Overview and does not report clicks, so any impact currently blends in with general AI Overview effects.
How should agencies adjust tool-building pitches because of this?
Favor tools built around a client’s proprietary data or unique workflow rather than tools that only run public formulas on generic inputs. The latter category is exactly what generative UI is built to replicate instantly and for free.