What Is Multimodal Search? Google Lens, Circle to Search and AI Mode Explained

You've probably used it without the label. Google's Think with Google site says Lens handles more than 25 billion queries a month, based on internal data from April 2025. Circle to Search reached more than 300 million Android devices by July 2025. And Google now reports this traffic to site owners as its own multimodal search type in Search Console.

This is the searcher's explainer: what the term means, how it works, the main tools as of September 2026 and which one to pick. If you run a website and want to know what these tools do with your images and charts, read our guide to multimodal GEO and how AI search "sees" visuals.

What Multimodal Search Means

The word "modal" refers to a mode of input, such as text, image, audio or video. Classic web search took one mode: typed words. Multimodal search takes several, and it can use them together in the same query.

The everyday meaning

For most people, multimodal search means pointing a camera or sharing a screenshot and asking about it. Snap a plant and ask if it's safe for cats. Circle a jacket in a video and ask where to buy it. Upload a photo of a bookshelf and ask for similar books. The picture supplies what's hard to describe, and the words supply what you want to know.

The developer meaning

Engineers use the term for search systems they build. Microsoft's Azure AI Search docs define multimodal search as the ability to "ingest, understand, and retrieve information across multiple content types, including text, images, video, and audio." Their example is a company search tool that answers a question from a diagram inside a PDF. Same idea, but inside an organization's own documents, not the public web.

How Multimodal Search Works

Every product handles it differently, but Google has described its version in the most detail. Here's the usual sequence.

1. You give it an image, and maybe words. A photo, a screenshot, a live camera view or a circled part of your screen.
2. The system works out what's in the image. For years, Lens has identified objects by matching against a database of images indexed from the web. Since February 2025, it also adds AI Overviews for more unusual images.
3. It splits the image into parts and searches each one. Google calls this "visual search fan-out". In Circle to Search, Google says the model picks the most important parts of an image to crop, runs several searches at once and cross-references the results.
4. It answers. Results can include an AI Overview, search results for objects in the image, similar images and websites that use the same image.
5. You follow up. Type or say "show me this in black" or "is it cheaper anywhere?" and the search continues from the same picture.

A Google search director described the split of labor in a March 2026 interview. The Gemini model looks at the image and decides which tools to use. The visual search backend supplies web results from its index. Then the model reads those results and writes one answer with links.

Multimodal Search Tools as of September 2026

The table lists the main tools and how each vendor describes access, checked on 30 September 2026. For AI Mode's full set of inputs, see our guide to using AI Mode in Google. For voice and camera conversations, see AI-powered voice search.

Google Lens and Circle to Search

Lens is the base layer. In the Google app or Chrome, tap the Lens icon to take or upload a photo, or touch and hold any image on a website and choose to search it. Tap "Ask about this image" to add words. On Android you can press and hold the shutter to ask by voice, in English.

Circle to Search is Lens for your whole screen on Android. Google's help page lists select phones, including Pixel 6 and later. Since February 2026, you can circle several things at once, starting with the Galaxy S26 series and Pixel 10. It can also translate the whole screen, and it can check whether a photo is an original or has been modified. On iPhone, the nearest match is Lens in Chrome or the Google app, which lets you select and search what's on your screen.

Introducing a new way to search | Circle to Search

ChatGPT, Microsoft and Apple

OpenAI says every ChatGPT model accepts images, on free and paid plans, as PNG, JPEG or non-animated GIF files up to 20 MB each. It suits open questions like "what's wrong with this spreadsheet chart?" but it's a chat, not a results page.

Microsoft offers two tools. Bing Visual Search takes a dragged, pasted or uploaded image and returns similar images, products and pages that include it. Copilot Vision watches a screen or camera feed you choose to share and talks you through it, but it needs a Microsoft 365 Personal, Family or Premium plan for consumers.

On iPhone, Visual Intelligence searches for similar images online, looks up businesses and identifies plants and animals. With the ChatGPT extension on, it can also ask ChatGPT about what the camera sees. Our post on what Apple Intelligence does covers the rest of Apple's features.

Which Multimodal Search Tool to Use

Most people only need two or three of these. We call this table the Multimodal Search Picker: start with what you want to do, not with the app.

Shopping is a big share of multimodal search. Google said in 2024 that 20% of Lens searches were shopping-related, and it has since built outfit and room searches that find every item in a picture at once.

What Gets Sent When You Search With an Image

Most multimodal search runs on the company's servers, so your image or screen usually leaves your device. The vendors describe what happens next in different ways.

• Chrome's Lens: a screenshot of the page and page data go to Google. Google says it keeps them only for processing and doesn't use them for training "as a result of this."
• Copilot Vision: Microsoft says session data may be kept for up to 47 hours and isn't used to train its models.
• Bing Visual Search: Microsoft says photos you provide "may be used to improve Bing image processing services."
• Circle to Search: Google notes it may not work in some apps, such as banking and medical apps.
• Apple Visual Intelligence: image searches are subject to your search provider's terms and privacy policies.

If a screenshot holds personal details, crop them out before you search.

What Multimodal Search Means for Site Owners

When someone searches with an image, Google can send them to your page from the visual results. Since 24 September 2026, Search Console's Performance report shows that traffic under the "Web: multimodal" search type, covering Lens, Circle to Search, image uploads and Chrome image search.

To appear there, your images need the basics: real <img> elements, useful alt text, captions and nearby text. Our multimodal SEO guide walks through images, video and audio. For whether AI answers read the text inside your charts, and how to design charts that survive a screenshot, see our full guide to multimodal GEO.

Rankbox helps with the written side: its Citation-Ready Writer researches the live web and writes source-backed articles that give your images clear text around them. It doesn't track visual search traffic or AI citations. The Business plan is $49.50 a month with a 7-day trial (pricing).