Google's secret weapon to merge AR and AI

Ever since the generative AI frenzy took off, critics have been quick to write Google's obituary, especially after its rushed Bard debut stumbled out of the gate.

Yet writing off Mountain View ignores a massive structural advantage: Google sits on a 20-plus-year knowledge graph that might be the most comprehensive AI training dataset on Earth.

While rivals focus on conversational chat boxes, Google is quietly merging computer vision, augmented reality, and large models into something far more tangible.

Search what you see and say

Describing an object you encounter in real life using text alone is often frustrating. Google's solution is Multisearch, which it rolled out earlier this month to all mobile devices where Google Lens is supported.

Multisearch lets users snap a photo of an item and immediately refine the query with text, such as pointing at a shirt and typing "show me the same shirt in green."

The feature runs on Google's Multitask Unified Model, an AI engine capable of understanding connections across text, photos, and videos indexed over decades.

Google is also developing a "search your screen" capability, effectively moving Lens from your outward-facing camera to anything currently displaying on your phone display.

Mapping the physical world

The real moat, however, comes from combining visual queries with local commerce through Multisearch Near Me.

Consider spotting an appealing dish on Instagram. Users can scan the image with Lens, then trigger a local query to find nearby restaurants serving similar food.

Upstarts like OpenAI have impressive neural networks, but they lack the physical infrastructure data. Google has spent years amassing local merchant listings via Google Business Profiles and product catalogs through Google Shopping.

Paired with its Live View AR navigation, which uses computer vision to localize devices outdoors, Google is essentially building a spatial knowledge graph for physical environments just as it did for the web.

Why it matters

For consumers, Multisearch delivers immediate everyday utility. Instead of wrestling with text prompts to describe a pattern or dish, combining a camera snapshot with a quick filter makes visual discovery genuinely useful, and it is ready to use today on Lens-enabled phones.

For the tech industry, this shifts the battleground away from generic chatbots toward real-world monetization. Language models cannot recommend what to buy or eat without deep merchant data, local business relationships, and accurate proximity mapping.

In our view, tethering multimodal AI to AR computer vision is Google's smartest defensive play yet. It sidesteps the text-only copycat race and leans directly into assets competitors cannot easily replicate.

The catch is consumer habit. Getting people to routinely point their cameras or search their screens rather than typing into a search bar takes friction removal, and any breakdown in local inventory accuracy could sour trust quickly.