Definition
Visual search is a way of searching in which the query is an image: a photo, a screenshot, a live camera view or a region of the screen. The system works out what the image shows and returns matching products, places, facts or similar images, often together with a typed or spoken follow-up question such as 'where can I buy this in green'.
Google runs the largest visual search products. Its 2025 year-end search trends report put Google Lens at more than 25 billion queries a month and said one in five Lens searches has commercial intent. Circle to Search, which lets Android users circle or tap anything on screen, was available on more than 580 million devices by February 2026. That month Google upgraded it with Gemini 3 so it can pick out several objects in one image, search for each in parallel and return results for the whole outfit or room at once. Search Live adds a live camera feed to a spoken conversation with AI Mode. Amazon offers photo search inside Alexa for Shopping, the assistant formerly called Rufus, and Amazon said customers used its Lens camera search 45% more in 2025 than in 2024. General chat assistants also accept images: OpenAI's GPT-5.6 models and Google's Gemini 4 Argon both take image input, so a user can upload a photo to ChatGPT or Gemini and ask what it is or where to buy it.
Under the hood, current visual search runs on multimodal AI. A vision model encodes the image into embeddings, those are matched against indexed images and products, and a language model describes the match and answers the question. Google describes the Circle to Search step of cropping an image into parts and running several searches at once as a visual form of query fan-out. The results then come from ordinary indexes: the Shopping Graph for products, Google Maps for places and the web index for pages. A visual query can only land on a page or product that is already indexed with images and text that confirm what the picture shows.
Visual search differs from reverse image search. Reverse image search looks for copies or near-copies of a specific image across the web. Visual search tries to understand the subject (this chair, this plant species, this error message) and find related things, even when no copy of the photo exists online.
For brands, visual search has been hard to measure until recently. On September 24, 2026, Google added a 'Web: multimodal' search type to the Search Console Performance report. It counts web results reached through Lens, Circle to Search, image uploads to Google Search and Chrome's 'Search this image' option. The report shows pages, clicks and impressions but no queries, because the query is a picture. That gives ecommerce, travel and local businesses a first page-level view of which URLs visual searches reach, in Google Search Console alongside their regular data. What to change on those pages is covered under image search optimization, and the practice of preparing content for answers that mix text, images and video is multimodal search optimization.
Examples of Visual Search
- A shopper on a Pixel 10 circles an outfit in a social media post. Circle to Search identifies the jacket, boots and bag separately and shows shopping results for each item.
- A homeowner points Search Live at a cracked dishwasher rack wheel and asks for the replacement part. AI Mode answers out loud and links to two parts retailers that list the exact model number.
- An outdoor retailer opens the new multimodal filter in Search Console and sees that tent product pages get impressions from image searches while its category pages get almost none, so it adds clear product photos and model names to category pages.
- A user uploads a photo of a houseplant with brown leaf tips to Gemini and asks what is wrong. Gemini names the species, suggests overwatering as a likely cause and cites a university extension page.
