Shoppers can now search with a photo instead of words. Retailers say it multiplies conversion. Here is what holds up, what doesn’t, and, new this autumn, how you can finally measure it yourself.
Last verified: 14 November 2026.
Picture someone watching a video. A lamp sits in the background: brass, arched, linen shade. She wants it and has no idea what it’s called. She types “gold curved lamp,” scrolls past forty wrong results, and closes the tab.
That gap, wanting something you can see but can’t name, is the whole business case for visual search. Google describes shoppers using Lens for things like identifying an item that appears in the background of a video.
The pitch is easy to believe. The numbers retailers publish are harder to believe: 2x, 5x, even 8x conversion. I went through the case studies, the platform data, and the little independent research that exists. Some of it is solid. Some is much weaker than the headlines suggest.
The short answer. Visual search lets shoppers search with a photo, and image recognition matches it to products. Google says Lens handles more than 25 billion visual searches a month, with about one in five showing commercial intent. Retailers report conversion lifts from roughly +2% to 8x. But the biggest figures compare people who chose to use the feature against everyone else, which can’t separate cause from intent. The one A/B-tested, whole-site result I found was +2.0% to +11.7% in conversion and +2.3% in revenue per session. Visual search complements keyword search. It does not replace it.
New this autumn: three changes that matter
- You can now measure it. On 24 September 2026, Google added a multimodal filter to the Search Console performance report. It separates web searches where an image was part of the query (Lens, Circle to Search, image uploads, Chrome’s “Search this image”) from typed searches. Details in section 7.
- Pinterest is selling ads against it. At its 17 September event, Pinterest announced visual search ads in beta and said it handles more than 80 billion searches a month, most of them visual.
- Old statistics are still circulating as if they were new. Pinterest’s famous “600 million visual searches a month” is a years-old number. More on that below.
1. The scale is real. Here’s which numbers to trust.
The best-sourced anchor is Google’s own. In October 2024, it said Lens is used for nearly 20 billion visual searches every month, and 20 percent of Lens searches are shopping-related. Do the arithmetic and that’s roughly 4 billion shopping-related visual searches a month (my calculation from Google’s two figures, not a number Google published).
Google’s marketing pages now cite a bigger figure. A Think with Google article says Lens sees 25 billion searches a month and that one in five has commercial intent. A more recent Think with Google piece repeats the 25 billion figure and adds that more than one in six AI Mode queries is entirely non-text. Be careful with dates: several blogs attribute the 25 billion figure to “June 2026” or “October 2025,” and those claims conflict with each other. The Google pages I could open don’t show a clean first-publication date, so I’d cite it as “Google says” without pinning a month.
| Figure | Source | Confidence |
|---|---|---|
| Lens: nearly 20B searches/month, 20% shopping-related | Google, Oct 2024 | High (first-party, dated) |
| Lens: 25B+ searches/month, 1 in 5 commercial intent | Think with Google | Medium-high (first-party marketing page; publication date unclear) |
| ~4B shopping-related Lens searches/month | My arithmetic on Google’s 2024 figures | Medium (derived) |
| Amazon: visual searches up 70% year over year | Retail TouchPoints, citing Amazon (Oct 2024) | Medium (secondhand; growth rate only, no volume) |
| Pinterest: 80B+ searches/month, “most of them visual” | Social Media Today, 17 Sep 2026 | Medium (company claim; see caveat below) |
| Pinterest Lens: 600M visual searches/month | Reported by VentureBeat about a year after Lens launched in 2017 | Stale. Don’t use as a current figure. |
| “Google visual searches top 50B/month” | One crypto-news site relaying a social post | Unverified. Not used. |
A caveat on Pinterest’s 80 billion. Pinterest is a platform where results are images, so “visual” there describes what the results look like. That’s not the same thing as a shopper uploading a photo. Treat the 80 billion as evidence that image-led discovery is huge on Pinterest, and don’t line it up against Lens as a like-for-like camera-search count. (That’s my reading of how the term is used, not something Pinterest spells out.)
2. Two kinds of numbers hiding in one pile
When I lined up the retailer results, they fell into two very different groups.
Group A: “users of the feature convert at X times the rate.” These are the dramatic ones.
Group B: “we ran the test across the whole site.” These are small and far more trustworthy.
Here is how the headline conversion figures compare. All are vendor-reported, and the bar length is the reported multiple:
Reported conversion multiple (vendor case studies)
Chow Sang Sang █████████████████████████████████████████ 8.2x
Coleman Furniture ████████████████████████████████████ 7.1x
Signet Jewelers UK ██████████████████████████████████ 6.8x
CITY Furniture ██████████████████████████ 5.27x
Snitch ████████████████████ 4x
Venca ███████████████████ 3.79x
Rinascimento █████████████ 2.68x
Urban Decay █████████████ 2.5x
Boohoo (Similar) ██████████ 2.0x
Boohoo (Camera) █████████ 1.85x
Baycrew's █████████ 1.76x
-------------------------------------------------------------------
PrettyLittleThing █████▌ 1.02–1.12x
(whole-site A/B test)Look at the last line. It’s the only result where the published summary says the effect was measured across the whole website in an A/B test, and it’s the smallest by far.
3. The case files
Every row below is vendor-reported. I read the published case-study summaries (linked). The full PDFs sit behind a document viewer I couldn’t open, so the “what’s compared” column reflects only what each summary lets me say.
| Retailer | Vendor | Headline result | What the summary says is compared |
|---|---|---|---|
| Boohoo | Syte | Camera Search users: +85% conversion. Shop Similar: 2x conversion, +12% AOV | Feature users; baseline not stated |
| Signet Jewelers UK | Syte | 6.8x conversion, +17.7% AOV | Shoppers who used the Discovery Button |
| CITY Furniture | Syte | 5.27x conversion, +26.3% AOV | Not stated |
| Chow Sang Sang | Syte | 8.2x conversion, +13.4% AOV | Shoppers using the tool |
| Coleman Furniture | Syte | 7.1x conversion, +29% AOV | Not stated |
| Venca | Syte | 3.79x conversion, 3.9x time on site | Not stated |
| Baycrew’s | Syte | 1.76x conversion, +6.13% AOV | Not stated |
| Rinascimento | Syte | 2.68x conversion, +17% AOV (per title) | Not stated |
| PrettyLittleThing | Syte | Total-site conversion +2.0% to +11.7% by region; +2.3% revenue per session | A/B tested, whole site |
| Stuarts London | Salesfire | Influenced 8% of orders; overall conversion uplift of 8.19% | “Influenced” is attribution, not a test |
| Moss | Salesfire | 16.7% conversion via Visually Similar Search; +3.7% AOV | Feature-level rate |
| Urban Decay | Edgecase | +150% conversion, +16% AOV, 5x products viewed | Feature users |
| AJIO | ViSenze | Visual-search users convert at roughly 1 in 5 | Not stated |
| Snitch | ViSenze | 4x conversion, 7x click-through rate | Not stated |
| Myntra | ViSenze | +35% year-on-year growth in image-search adoption | Adoption, not revenue |
Two patterns stand out. First, every case is fashion, jewelry, furniture, or beauty, the categories where describing a product in words is awkward. Second, Myntra’s headline is about how many people use the tool, not what it earned. That’s an honest result. It just isn’t a revenue claim.
4. Why 6.8x probably isn’t 6.8x
Imagine two shoppers. One types “dress” and wanders. The other points her camera at a specific dress she saw on someone and uploads it.
Who buys more often? The second one, almost regardless of the tool. She walked in with a target.
That’s the core problem with most “feature users convert X times higher” numbers. If the comparison is people who chose visual search against everyone else, you’ve measured who picks the feature, not what the feature does. The same trap affects “influenced X% of orders,” which counts orders that touched the feature, not orders it created.
To be fair: the summaries I read don’t spell out how every comparison was built, and some of these results may come from cleaner designs. The point isn’t that vendors are lying. It’s that a number without a stated baseline can’t be interpreted, and a smart buyer should insist on one.
What PrettyLittleThing teaches
The best-documented result is PrettyLittleThing, a UK fashion retailer aimed at women aged 16 to 24. Per the published summary, Syte built a camera-search funnel that used social-media imagery, and the A/B-tested whole-site conversion lift ranged from +2.0% to +11.7% depending on region, with revenue per session up 2.3%.
Notice the gap. Conversion in the best region rose almost twelve percent, but revenue per session rose a little over two. A conversion rate can climb while the money per visit barely moves, which is why you should always ask which metric a case study is reporting. Also note the audience: a young, mobile-first, social-driven shopper is close to the ideal user for this feature. Your customers may differ.
5. What independent research says
Independent evidence is thin, but it exists.
A field experiment on a large Chinese e-commerce platform. Economists Lu Fang, Yanyou Chen, Chiara Farronato, Zhe Yuan and Yitong Wang studied a search tool that suggested refined queries using images and text. They found no immediate effect on orders or spending, but a long-run change in search and purchasing behavior. In the fuller abstract, customers with the tool eventually increased orders and expenditures, especially for non-top-selling products. The gain came from more effective searching, not more time on the platform, and customers appeared to learn to search better on their own.
Two cautions: this tested visual and text refinement of a typed query, not a camera upload, and it covers one platform in one market. But it points to the likely mechanism. The value isn’t a conversion switch. It’s helping people find things they’d otherwise miss, especially in the long tail of a catalog.
A lab study on wine. Heejin Kim, Mijin Kwon and Song Oh Yoon ran three controlled experiments (N = 321) in the Journal of Research in Interactive Marketing, comparing visual search through the Vivino app with text search. They report that visual search raised purchase intention and product valuation, driven by ease of use, novelty and enjoyment (DOI as listed on the aggregator: 10.1108/jrim-04-2025-0193). It’s small, single-category, and measured intention and taste perception, not real store revenue. Suggestive, not decisive.
Notice what the independent work does not contain: a 6.8x.
6. Does visual search replace keyword search?
No, and the evidence points the other way. The NBER authors open by noting that text search remains the predominant tool for product discovery online. Google’s own framing in its recent marketing material is of a broader “intelligent search box” that takes text, voice, and images, not one that drops the keyboard.
Where visual search does the most work:
- Shoppers who can see it but can’t say it. Style, pattern, shape, decor.
- Inspiration-led categories. Fashion accessories were the most-shopped Lens category in Google’s 2024 reporting.
- Long-tail catalogs. The NBER finding that gains concentrated in non-top-sellers is the best independent hint.
- Mobile. Cameras live on phones.
Where it likely adds little (my judgment, not a measured result):
- Shoppers who already know the model number or brand.
- Replenishment and commodity items.
- Small catalogs a visitor can browse in a few clicks.
7. New: you can finally measure it
Until this autumn, the standard advice was “visual search traffic is hard to attribute.” That has partly changed.
On 24 September 2026, Google announced web multimodal reporting in Search Console. Coverage of the launch describes it like this:
- Where: Performance > Search results > Search type > Web > Multimodal.
- What it counts: web results shown when an image was part of the search, including Google Lens, Circle to Search on Android, image uploads to Google Search, and Chrome’s right-click image search. Typed searches are reported as “Web: text-based.”
- What you don’t get: the image itself or any query text. Per a Google product lead quoted by PPC Land, you can see which of your pages appeared and how often.
- Rollout: announced as global and gradual, so it may not be in your property yet. Early coverage noted the Search Analytics API had no multimodal value at launch, so check whether that has changed before building a dashboard on it.
This is the first time a merchant can see, in Google’s own data, which product pages surface when shoppers search with a camera. It reports on visibility in Google’s results. It still doesn’t prove that a Lens search caused a sale. For causation, you need an experiment (section 9).
8. If you sell products: get found in Lens first
Most advice about visual search says “install a widget on your site.” That ignores a bigger question: can shoppers find you when they point Google Lens at a product?
You can’t control Lens, and Google doesn’t publish a ranking formula for it. What’s reasonable to do:
- Test your own products first. Take your ten best sellers, photograph or screenshot each, and run it through Lens on a phone. Note whether your page shows up, whether the price and availability are right, and who outranks you. It’s free and takes an hour.
- Follow Google’s image guidance. The image best-practices page is the primary source (reportedly updated in English on 2 March 2026). It lists supported formats including WebP and AVIF, and explains how to signal a preferred image with schema.org markup or an
og:imagetag. - Keep structured data and feeds accurate. Google’s Product structured data documentation separates product snippets (pages where people can’t buy directly) from merchant listings (pages where they can), and notes that Google may combine page markup with Merchant Center feed data. A wrong price in either place can undermine the match. Google’s 2022 merchant-listings announcement mentions Lens in this context.
- Shoot multiple angles plus one in-context image. Practitioners commonly recommend several views per product. That’s good practice for matching, but I found no controlled evidence on the ideal number, so treat it as craft advice.
- Watch the new filter. Check Search Console’s multimodal view monthly and see which pages appear.
9. If you’re buying a visual search tool
Ask every vendor these before you sign:
- What was the comparison group? Feature users versus all visitors, or a randomized control?
- Was it an A/B test across the whole site? What happened to total revenue per visitor?
- Which metric? Conversion, revenue per session and AOV can move in different directions (see PrettyLittleThing).
- How long did it run, and across which seasons?
- What was the catalog size and category? Their win may not transfer to yours.
- What happened to returns and search latency?
- Can I run my own test before a long contract?
A test design you can run yourself
- Randomize before exposure. Split visitors into control and treatment before they see the feature.
- Measure everyone. Compare revenue per visitor and conversion for all visitors in each group, not only those who tapped the camera icon. This “intent-to-treat” approach is what removes self-selection.
- Pick your primary metric in advance, ideally revenue per visitor, and don’t switch after you see the data.
- Run it through at least a couple of normal buying cycles, avoiding a single promotion spike.
- Add guardrails: page speed, bounce rate, return rate.
- Report a range, not a headline. Expect single digits to low double digits on conversion, as the one documented A/B result suggests, and check revenue per visitor separately.
Sources and confidence
| Claim | Source | Confidence |
|---|---|---|
| Lens ~20B/month, 20% shopping-related | Google, Oct 2024 | High |
| Lens 25B+/month, 1 in 5 commercial intent | Think with Google | Medium-high (date unclear) |
| Search Console multimodal filter, 24 Sep 2026 | Google Search Central Blog; details via PPC Land | High that it exists; medium on fine detail (read via secondary coverage) |
| Pinterest 80B+ searches, visual search ads | Social Media Today, 17 Sep 2026 | Medium |
| Amazon +70% visual searches | Retail TouchPoints, 2024 | Medium |
| Retailer conversion results | Vendor case-study summaries (linked in section 3) | Medium-low (vendor-reported; baselines mostly unstated; full PDFs not opened) |
| PrettyLittleThing +2.0 to +11.7%, +2.3% revenue per session | Vendor summary | Medium (states A/B test) |
| NBER field experiment | NBER WP 32099 | High for what it studied (a different tool) |
| Wine lab study | Synapse listing | Medium (journal named; I read the abstract only) |
| Google image and product-data docs | Images, Product | High that they exist; I read them via search excerpts, not the full pages |
| Pinterest Lens “600M”, Google “50B”, other uncited figures | Stale or unverified | Not used |
FAQ
What is visual search in e-commerce? It lets shoppers search with an image, by uploading a photo, using a camera, or tapping an item inside an image, and matches it to similar or identical products.
How big is visual search? Google said in October 2024 that Lens handled nearly 20 billion searches a month, about 20% shopping-related. Its marketing pages now cite 25 billion-plus.
Does visual search increase conversion? Retailers report increases from about +2% to 8x. The largest figures usually compare feature users to everyone else. The one A/B-tested, whole-site result I found was +2.0% to +11.7% conversion and +2.3% revenue per session.
Will it replace keyword search? Not on current evidence. Text search remains the main discovery tool; visual search fills the gap when shoppers lack the words.
Which stores benefit most? Fashion, jewelry, furniture, beauty and decor, where products are visual and hard to describe.
How do I track Google Lens traffic? Since 24 September 2026, Search Console’s performance report can filter web results by “multimodal” searches. It shows which pages appeared, not the image or query. Availability is rolling out.
How do I optimize for Google Lens? Use clean, multi-angle product images, accurate structured data and Merchant Center feeds, and fast mobile pages. Then test your own top products in Lens.
The bottom line
Visual search is real, large and growing, and as of this autumn it’s measurable in a way it wasn’t before. It solves a genuine problem: wanting what you can see and can’t name. But the 5x and 8x conversion stories are mostly about who uses the feature, not what the feature does. The best-supported expectation is a modest whole-site lift, concentrated in visual categories and long-tail products, that you should confirm with your own randomized test before paying for it.
Data checked 14 November 2026. Vendor figures are as published by the vendors; independent sources are linked. Platform statistics change often, so re-check them before quoting.