What is visual search?

Visual search flips the starting point of a query. In a classic search, the user types words and Google returns pages or images that match that text. In visual search, the user provides an image, a photo taken on a phone, a screenshot or an uploaded file, and Google identifies what that image contains in order to return related information.

The tool that runs this feature on most devices is Google Lens, built into the Google app, into Chrome and into google.com's own image search. According to Google Search Help, uploading or dragging an image into the search box can return AI overviews, search results about the objects in the image, similar images and websites that contain that image or a similar one.

It is worth distinguishing it from classic image search: typing «red sneakers» and seeing a gallery of photos. There, text remains the starting point; in visual search, it is the image. Although Lens is the most widely used implementation, other systems such as Pinterest Lens or Bing Visual Search follow the same principle. This article focuses on Google's implementation, the one relevant for SEO in its search results. It is worth being clear that Google Lens is not a separate search engine, but an additional entry point into the same results page that also appears for a classic text search.

The process starts when the user provides an image: a photo taken with the camera, an uploaded file, a pasted URL or an image selected within another search. Lens sends that image to Google's servers, which analyse it with computer vision algorithms to detect objects, text, colours, shapes and, where applicable, specific brands or models.

That visual analysis is then cross-referenced with Google's image index, the same one that feeds Google Images. This is where the signals that already existed for image SEO come in: alt text, the textual context of the page the image lives on, and structured product data. Google's documentation on image SEO explains that it uses alt text together with computer vision algorithms and the content of the page to understand what the image is about. Visual search adds a layer of direct recognition of the visual content, but it does not replace those signals, it complements them. This overlap between visual recognition and the text index explains why two visually identical images can return different results if the textual context around them is different.

The result the user receives can be an AI-generated overview, a list of pages related to the identified object, a gallery of visually similar images or, in the case of products, purchase links with price and availability if the source page includes that markup. When the user adds text to the image, for example «in blue» or «size 38», Lens combines both signals to refine the result, a feature Google calls multisearch.

For a page to have any chance of appearing in these results, the image must be crawlable, not blocked by robots.txt, and must not be embedded only as a CSS background image, because Google does not index that type of image. It must also live on a page with clear textual context about what it shows. Because this matching happens largely in real time, the result for the same photo can change if Google updates its image index or a page revises its metadata.

Visual search gains weight in categories where describing what you're looking for in words is harder than showing it: fashion (a specific print, a particular cut), furniture and decor (a piece of furniture seen in someone else's photo) and spare parts (a mechanical part with no clear commercial name). In these sectors, a buyer can photograph the object directly instead of trying to describe it in a text search.

There is no public data with a verifiable methodology on what share of searches in Spain today happen through an image rather than text. Any figure circulating about this without citing who measured it and how should be treated as unverified, not as fact.

What is verifiable is that Google has built specific infrastructure for this feature, with Lens built into Search, into Chrome and into Google Photos, and that it actively documents it as an additional discovery path alongside text search. The phenomenon is not exclusive to Spain: Google deploys the same Lens infrastructure in every market where it operates, which suggests a global bet rather than a one-off test, again without publicly verifiable figures on its relative weight by country. For a shop or catalogue with a strong visual component, ignoring this path means relying solely on the buyer hitting on the right words, when they could instead find the product with a photo taken on the street or cropped from another website. For marketplaces and comparison sites, this also means a user can land directly on a seller's page from someone else's photo, without ever knowing the manufacturer's brand name. Investing today in good image quality and clean product data lays the groundwork for both classic image search and visual search, because both rely on the same signals.

• Write a descriptive, specific alt text for every product or content image, avoiding generic text such as «image1.jpg».
• Publish sharp, high-resolution images: Google favours clear photos because they create a better experience and make visual recognition easier.
• Mark up product pages with structured data (Product, Offer) so Lens can show price and availability alongside the identified image.
• Place the image next to relevant text about what it shows: a title, description and context help both the user and the algorithm.
• Use supported, lightweight formats (WebP, JPEG, PNG, AVIF) and make sure the image URL is not blocked by robots.txt.
• Submit an image sitemap if your catalogue is large, so Google discovers photos it might otherwise not crawl.

• Embedding the product only as a CSS background image: Google doesn't index CSS images, so they fall outside visual search.
• Leaving the alt text empty or filling it with a list of keywords unrelated to the actual image.
• Publishing blurry photos, with an intrusive watermark, or cropped in a way that hides the main object.
• Separating the image from any textual context, for example in a carousel with no visible title or description.
• Accidentally blocking the images folder from crawling in robots.txt, a common mistake after a CDN migration.