Google's New AI Search Box

On a Tuesday morning in Surry Hills, a customer does not type “best physio for runners near me” into Google and scroll. They photograph their swollen ankle, drop it into the search box, add “can I still run the City2Surf in three weeks”, and read a single, confident answer that names two clinics and books a Thursday appointment. No scrolling. No ten blue links. That is not a hypothetical — that is the search box Google just shipped.

At I/O 2026, Google called it the biggest upgrade to its Search box in over 25 years. The box now takes text, images, files, videos, and even your open Chrome tabs as input. It runs on Gemini 3.5 Flash by default inside AI Mode. And according to Google, AI Mode has passed 1 billion monthly users within a year, with queries more than doubling every quarter. For Australian businesses, this is the moment the search box stopped being a doorway to your website and became the destination itself.

“The search box used to be where the journey to your website began. Now it is increasingly where it ends.”

What Google actually shipped

Strip away the keynote theatre and the upgrade comes down to three changes that matter for anyone who wants to be found in search.

It is multimodal. The box is no longer a text field. You can hand it a photo, a PDF, a video, or the tabs you have open in Chrome, and it reasons across all of them at once. A search is now a situation, not a string of keywords.

It is conversational. AI Mode, running Gemini 3.5 Flash by default, holds context across a thread the way a chat does. The follow-up question is the search now, and Google has built the box to expect it.

It is becoming agentic. Google announced Information agents that work 24/7, launching in summer 2026 for Pro and Ultra subscribers, alongside generative UI — on-the-fly “mini apps” the box builds to answer a query — and agentic booking that can complete a reservation inside the answer. As Elizabeth Reid, Google’s VP of Search, put it: “We shared the next step in our journey to bring together the best of a search engine with the best of AI.”

Google also said its Personal Intelligence layer is expanding to nearly 200 countries and 98 languages. That last detail is the one most Australian operators skim past, and it is the one that should change their plans this quarter.

Our read: the box is now an answer surface

Here is the part the announcement will not say out loud. When the search box answers in place — photographs the ankle, reads the situation, names two clinics, books the appointment — the click that used to land on your website does not happen. The zero-click search has been creeping up for years. The AI Search Box does not nudge it; it makes the answer the product.

This is not a reason to panic. It is a reason to change the scoreboard. The question that mattered for two decades was “do we rank in the top three for this keyword.” The question that matters now is different.

“Winning is no longer ranking first for a keyword. It is being the source the answer cites — across every engine that writes answers.”

That work has a name. It is answer engine optimisation and generative engine optimisation — AEO and GEO — and it is a different discipline to blue-link SEO. Blue-link SEO competed for a position on a page. AEO and GEO compete to be the quotable, attributable source inside a generated answer. The two overlap, but they are not the same job, and the AI Search Box just made the difference impossible to ignore.

What to do now

There is no secret here, which is the good news. The signals that earn a citation are concrete, auditable, and mostly under your control. We organise the work into five moves.

A concrete example: a Melbourne allied health clinic

Take a multi-site physiotherapy and sports clinic with rooms in Richmond and Carlton. Under the old model, its growth came from ranking on page one for “physio Richmond” and “sports physiotherapy Melbourne,” then converting the click on a well-built location page. That motion still has value. It is no longer the whole game.

Under the AI Search Box, a prospective patient hands Google a photo of a knee, a screenshot of a GP referral, and the question “which Melbourne clinic does ACL rehab and bulk bills.” Google reasons across all three inputs and writes one answer. The clinic either appears in that answer as a named, linked, trusted source — or it is invisible, no matter how well it ranks on the classic results page.

Getting named is concrete work for that clinic:

1. MedicalClinic schema per location with medicalSpecialty, availableService (one node for ACL rehabilitation), practitioner Person nodes with AHPRA credentials, opening hours, and geo coordinates — so the engine can match the situation to the service.
2. Passage-level answers on the service page. A sentence that reads “Our Richmond and Carlton clinics provide bulk-billed ACL reconstruction rehabilitation under an Enhanced Primary Care plan” is quotable. A paragraph of brochure copy is not.
3. Entity strength — a single Organization graph linking both locations, the practitioners, and their professional registrations, so Google treats “the clinic” as a recognised entity rather than two unconnected addresses.

The clinic that ships this gets cited when the box answers a knee photo. The clinic that only chases keyword rankings watches the appointment get booked inside an answer it was never quoted in.

How we approach this at SEOTopSecret

SEOTopSecret is built for exactly this shift — the move from ranking pages to being the cited source. For the Melbourne clinic above, the work runs through one workspace rather than five disconnected tools.

Our Schema Generator ships the MedicalClinic, Organization, and Person graphs that match the visible page and validate against both Schema.org and Google’s Rich Results Test — the entity and structured-data layer the AI Search Box reads first. The Content Brief Engine builds the service and condition pages around citable, passage-level answers instead of brochure copy, so the box has a quotable sentence to lift. And our AI Visibility Tracking shows whether the clinic is actually being cited — across ChatGPT, Gemini, Claude, Perplexity, and Google AI Overviews — by query, over time, so the scoreboard finally matches the search box people are using.

If you want the deeper mechanics, our guide to how to optimise for AI Overviews and our breakdown of SEO for ChatGPT cover the citation work engine by engine, and the schema markup guide lists every JSON-LD type we deploy on production sites. You can see how the tiers fit on the pricing page.

The search box changed for the first time in a generation. The businesses that adjust the scoreboard now — from rankings to citations — will be the ones the answer names. The rest will be the answer’s sources without ever knowing it, or invisible inside it.