Point your camera at a flower, a jacket, a restaurant menu or a broken washing-machine part, and Google can tell you what it is, where to buy it, or what it says. That is visual search, and it now runs at a scale most publishers still ignore.
Google says its Lens tool handles more than 20 billion visual searches every month, with about one in five carrying shopping intent. This guide explains how Google Lens, Google Images and Ask Photos actually work, how they differ, and what changed in September 2026 when Google made visual search measurable for site owners.
How Lens Turns a Photo Into a Search
Every Lens search starts the same way: with an image instead of words. You point your phone camera at something, upload a photo, or long-press an image in Chrome, and Google treats that picture as the query.
Behind the scenes, Google’s vision systems analyse what is in the frame. They recognise objects, read any text visible in the image, identify landmarks and products, and match the result against images and pages across the web. If the picture contains a product, Google can also match it against its shopping listings, which is why a photo of a shoe can return places to buy that shoe.
You can then refine the result by adding words. Photograph a plant and ask how often to water it, or photograph a jacket and ask for it in another colour. The image sets the subject, and your words set the question. That combination of picture plus question is what Google calls multimodal search.
Lens lives in several places at once. It is built into the Google app on Android and iPhone, into the camera on many Android phones, into Google Photos, and into Chrome on desktop through the right-click “Search this image” option. You do not need a separate app to use it on most phones.
What Lens Can Actually Do Today
Lens is best understood as several tools sharing one camera. Point it at text and it can copy it, translate it, or read it aloud. Point it at a landmark and it can name it and pull up visiting information. Point it at a plant or an animal and it can identify the species.
Shopping is the most commercially important use. Google says roughly a fifth of Lens searches are shopping-related: people photograph products they see in shops, in friends’ homes, or on screen, and Lens finds matching or similar items for sale. For a shopper, the camera replaces the product name they do not know.
Lens also handles homework-style questions. Photograph a maths problem or a passage of text and Lens can work through it or explain it, a feature students use heavily. And it translates in real time: point the camera at a sign or a menu in another language and Lens overlays the translation on the original text.
What Lens cannot do is just as important. It identifies what is visible, not what is hidden. It can name a skin condition’s lookalikes but cannot diagnose you. It can find a product but cannot tell you whether it is good quality. And its answers are only as good as the image: a blurry, dark or cropped photo returns vaguer results than a clear one.
Google Images and Lens: Same Index, Different Question
Google Images and Lens draw on the same web, but they answer different questions. Images is the classic way to search: you type words and get pictures back. Lens reverses the direction: you start with a picture and get words, products and pages back.
That reversal matters for how results are chosen. A typed Images query leans on the words around a picture on its page: the caption, the nearby text, the page title. A Lens query leans on the picture itself: what the image shows, what text appears inside it, and which pages carry matching images.
For publishers, the practical consequence is simple. A page whose images are clearly labelled, well described and placed next to relevant text is easier for Google to connect to a visual query than a page whose images sit unexplained in a gallery. The image needs context, not just a file name.
Google Images still has its own role. It is where people browse for inspiration, compare options and find a picture to use or buy. Lens is where people start from something they already see. The two meet in the middle: a Lens result often leads to an Images-style grid of similar pictures.
Ask Photos: Searching Your Own Library With Gemini
While Lens searches the public web, Ask Photos searches your private photo library. Announced at Google I/O 2024, it is an experimental feature inside Google Photos powered by Gemini models. Instead of scrolling or filtering by date, you ask a question in plain language.
Google describes the process in three steps. Ask Photos first understands your question and forms a plan to find the answer, looking for people, places, dates and concepts such as “themed birthday party”. It then studies the results, using Gemini’s multimodal abilities to understand what is happening in each photo, including reading text inside the image. Finally it crafts an answer and picks the photos and videos to show you.
The classic example Google gives is a parent asking what themes their child’s birthday parties have had over the years. Ask Photos can look across years of party photos, recognise decorations and cakes, and answer with the themes it found, together with the matching pictures.
Ask Photos can also help with tasks beyond search. At the end of a trip, you can ask it to suggest the best pictures for a highlight reel, and it will pick the shots and even draft a caption to share. And you can describe edits in words — brighter sky, different background — and watch Gemini apply them to the photo.
There are limits worth knowing. Ask Photos is still experimental, so Google warns it will not always get things right, and it can remember corrections you give it for next time.
It is not available in every country or language, and on the web it is not offered at all: you need the Google Photos app on a phone. Google also says your Photos data is not used for advertising, and that Ask Photos conversations are not reviewed by people except in rare cases involving abuse or harm.
Circle to Search: Searching From Any Screen
Circle to Search, available on Android, removes even the step of opening an app. You circle, highlight or tap whatever is on your screen — a product in a video, a word in an article, a building in a photo — and Google searches for it without you leaving what you were doing.
It is the same idea as Lens applied to the screen itself. The circled thing becomes the query, and you can add a typed question on top, just as with Lens. A screenshot, a social media post or a paused video frame can all become a search.
For site owners, Circle to Search counts in the same bucket as Lens. When someone circles a product image on your page and lands on a competitor’s listing, that journey started with your image. When they circle something elsewhere and land on you, your image did the attracting.
The September 2026 Change: Visual Search Becomes Measurable
Until September 2026, visual search traffic was effectively invisible to site owners. It arrived mixed into general web traffic in Search Console, with no way to separate a visitor who typed a query from one who pointed a camera.
On September 24, 2026, Google changed that. It introduced a web multimodal search type in Search Console’s performance reports, covering both the Search results report and the Generative AI features report. Site owners can now filter their performance data to see traffic from searches that started with an image.
Google’s announcement names four entry points that feed the new filter: Google Lens, Circle to Search on Android, image uploads to Google Search, and Chrome’s right-click “Search this image” feature. If a visit began with any of those, it can now appear in the multimodal view.
Google also made a point that matters for reading the numbers: this traffic was not previously counted in Search Console’s web totals. It is newly visible volume, not a relabelling of existing traffic. A site that suddenly sees multimodal clicks is not necessarily gaining traffic; it is seeing traffic that was always there but unmeasured.
For a fuller account of Google’s generative AI reporting, including how the Generative AI features report works, see our guide to Search Console’s generative AI reports. The multimodal filter sits inside the same performance reports that guide covers.
What the Multimodal Filter Does Not Tell You
The new filter has one large blind spot: it shows no queries. When the multimodal filter is on, the Queries tab is unavailable, because the query was a picture, not words. You can see which pages surfaced, how often they were shown, and how often they were clicked, but not what image or question led there.
What you can still break down is useful. The Pages view shows which of your pages visual searchers land on. Countries, devices and dates work as usual, so you can see whether visual traffic is mobile-heavy or concentrated in particular markets. And you can export the data for analysis elsewhere.
The filter also only shows data if your site actually receives this kind of traffic. A site with no visual search visits will see an empty view, which is itself information: it tells you your images are not currently pulling people in through cameras and screenshots.
There are still open questions. Google has not said whether image prompts inside AI Mode or live video search sessions are included, and early reports suggest data availability varies by property. Treat the filter as a new window, not a finished map.
How Publishers and Shops Can Show Up in Visual Search
Showing up in visual search is mostly about making your images easy for Google to understand. The work is unglamorous, and it overlaps heavily with ordinary good image practice.
Start with the image itself. Use clear, high-quality photographs rather than tiny or heavily compressed ones, and show the product or subject plainly instead of hiding it in a busy scene. Google can only match what it can see.
Then give the image context. A descriptive file name and alt text that name the subject in plain words help Google connect the picture to the right queries. The text around the image matters too: a product photo sitting next to its name, price and specifications is far easier to match than the same photo floating in an unexplained gallery.
Make sure the image is crawlable. Pictures loaded as ordinary image elements on the page can be indexed; pictures baked in as page backgrounds generally cannot. If Google cannot fetch the image file, it cannot return it for a visual query.
Keep pages mobile-friendly. Most visual searches start on a phone, so a page that loads badly on mobile will lose the visitor even if its image wins the match. Speed and readability are part of visual search performance, not separate from it.
For shops, one detail deserves special attention: the product’s identifying information. A part number, model name or brand written on the page near the photo gives Google something exact to match when a shopper photographs a label. Vague listings match vaguely.
Finally, check your own multimodal data. Open Search Console, apply the multimodal filter, and look at which pages actually receive visual traffic. Those pages are your visual shop window. Improving their images and descriptions is measurable work now, where before it was guesswork.
A Worked Example: The Spare-Part Photo
Consider a technician repairing a washing machine. The failed part carries a worn label with a model number, but the technician does not know the part’s name or where to buy a replacement.
They photograph the label. Lens reads the text on the label, matches the model string to pages across the web, and returns sellers stocking that part. The technician compares prices and orders without ever learning the part’s formal name.
Every step of that journey rewards a specific page feature. The winning page had a clear photo of the part, the model number in its title and nearby text, and a crawlable image. The losing pages had the same part but a vague title, a gallery with no descriptions, or an image Google could not fetch.
The same pattern repeats across categories: a plant identified from a leaf, a dress found from a screenshot, a restaurant dish recognised from a holiday photo. In each case the camera replaced a word the searcher did not have, and the best-described image won.
What Is Still Hard for Visual Search
Visual search is powerful but uneven. It excels at “what is this” and struggles with “is this good”. It can identify a restaurant from a photo of its sign but cannot tell you whether the food is worth the price. Judgment still needs reviews, and reviews still need words.
Similar-looking things confuse it. Two nearly identical products from different brands, or two species of plant with similar leaves, can return mixed results. Adding a word to the image query — a brand, a colour, a place — usually resolves the ambiguity faster than taking another photo.
Privacy is a live question. Pointing a camera at people, number plates or documents and searching the result raises issues that typing a query does not. Google’s tools are built for objects, places and products; using them on people is both less reliable and, in many contexts, inappropriate.
And the private side has its own limits. Ask Photos is experimental, regionally limited and phone-only. It can misread a scene, and its answers deserve the same scepticism as any AI summary. The right response to a wrong Ask Photos answer is a correction, which Google says the feature remembers.
Why Visual Search Matters More Every Year
Three trends point the same way. Cameras keep getting better, so the starting image keeps getting clearer. Models keep getting better at understanding images, so the match keeps getting sharper. And younger users, who grew up photographing first and typing second, keep becoming a larger share of searchers.
Google’s own measurement change is the clearest signal of all. A company does not build a reporting filter for a curiosity. The September 2026 multimodal filter exists because image-led search is now large enough, and commercially important enough, to deserve its own line in the report.
For publishers, the implication is practical rather than dramatic. Nobody needs to rebuild their site for visual search. They need to describe their images properly, keep their pages fast on phones, and check the new filter the way they already check their ordinary search performance. The camera is becoming a search box. Treat it like one.
How to Get Better Results From Lens
Good Lens results start with a good picture. Fill the frame with the subject, shoot in decent light, and keep the camera steady. A close, sharp photo of one object beats a wide, dark photo of a whole room every time.
If the first result is wrong, crop tighter. Lens lets you adjust the area it analyses, and cutting out background clutter often changes the answer completely. A jacket photographed against a plain wall matches better than the same jacket photographed in a crowded shop.
Add words when the picture alone is ambiguous. Photograph a plant and add “is this poisonous to dogs” and you get a different, more useful answer than the plant’s name alone. The words narrow the intent; the picture supplies the subject.
Try another angle if the match fails. A product photographed from the front may match instantly, while the same product photographed from behind may not. Labels, logos and distinctive markings are the strongest visual clues, so include them in the frame when you can.
For text, get close and keep the text level. Lens reads printed text well when it is straight, well-lit and in focus. Handwriting works less reliably, and curved or reflective surfaces — a bottle label, a glossy menu — may need a second attempt from a different angle.
For shopping, photograph the product rather than its picture on a screen. A photo of a screen adds glare and moiré patterns that confuse matching. If you only have a screenshot, crop to the product itself before searching.
Finally, remember that Lens results change. New listings appear, pages get better images, and Google’s models improve. A product that did not match six months ago may match today, so a failed search is worth retrying after time has passed.
Lens, Images and Ask Photos Compared
The three tools look similar but answer different needs, and choosing the right one saves time. The table below is a summary; the sections above explain each in detail.
Lens answers “what is this thing in front of me”. Its input is a camera photo, an uploaded image or a screenshot. Its index is the public web plus shopping listings. Its strength is identification and shopping. Its weakness is judgment: it names things, it does not review them.
Google Images answers “show me pictures of this”. Its input is typed words. Its index is the public web’s pictures. Its strength is browsing and comparison: seeing many options at once. Its weakness is that you must already know the words for what you want.
Ask Photos answers “find this in my own pictures”. Its input is a question in plain language. Its index is your private Google Photos library. Its strength is memory: finding a moment across years of photos without scrolling. Its weakness is availability: experimental, phone-only, and not in every country.
Circle to Search sits alongside Lens. It answers “what is this thing on my screen” without making you leave the app you are in. Its input is a gesture on the screen. For most purposes you can treat it as Lens with a faster front door.
Used together, the four cover the whole journey. You circle a dress in a video, Lens finds it for sale, Images shows you similar styles, and later Ask Photos finds the photo of you wearing it at last summer’s wedding. The camera starts the search, and your library remembers the ending.
Background
Google Lens launched in 2017 and has grown from a camera trick into one of Google’s largest search surfaces, while Google Images has indexed the web’s pictures since 2001. Ask Photos, previewed in 2024, extends the same idea into the private photo library, and the September 2026 Search Console update gave site owners their first dedicated view of the traffic it all produces. You can browse the subjects we cover on our topics index.
Sources: Google Search Central: “web multimodal reporting”; Google Blog: “Ask Photos”.
0 Comments