What Google Lens treats as shoppable: 752 categories, ranked

When you point the camera with Lens, an object detector runs on the phone itself. It knows 703 kinds of objects. Each one is a Knowledge Graph entity, named by a code called a MID: /m/01940j, for instance, is the backpack. Lens puts a chip on the objects it keeps, and tapping the chip runs a search on that object.

The Lens table ranks these 703 objects, plus 49 general categories (Product, Vehicle, Mammal…) that the detector never returns. The start of the file, as found in the app (“gleam” is the chip):

# Lens Proactive Shopping use case annotation description:
#   ALLOWLIST: Class labels of mobile RAID which should trigger gleam on those objects.
#   NEUTRAL: Class labels of mobile RAID which should not trigger gleam on those objects.
#   DENYLIST: Class labels of mobile RAID which should not trigger any intent.
# class label mid id, class label display name, lens prime annotation
/m/0584n8,Briefcase,ALLOWLIST
/m/0c_jw,Furniture,ALLOWLIST
/m/0c06p,Candle,ALLOWLIST

9:41Google Lens

1

Armchair

2

Floor lamp

3

4

Armchair: Shopping results

Products, merchants and prices picked by the server

1 · Detection 2 · MID and score 3 · Verdict 4 · Chip 5 · Server

1 Armchair Shopping chip

/m/01mzpv Chair 0.94 ALLOWLIST

/m/0c_jw Furniture 0.71 ALLOWLIST

/m/0fqt361 Stool 0.22 below 0.35

2 Floor lamp Shopping chip

/m/0dtln Lamp 0.89 ALLOWLIST

/m/019sc6 Lighting 0.64 ALLOWLIST

3 Dog No chip

/m/0bt9lr Dog 0.95 NEUTRAL

/m/0jbk Animal 0.61 NEUTRAL

4 Books No chip

/m/0bt_c3 Book 0.81 NEUTRAL

Lens ignores a label scoring under 0.35. Tapping a chip runs the search; the server picks the products. In 2026 the table was not used in any of the cases we observed (see section 5).

One banned label is enough

Lens often gives the same object several labels and ignores those whose score is too low. A single remaining label on the DENYLIST is enough to drop the whole object. A kitchen knife, for instance, carries the label “Kitchen knife”, which is neutral; if Lens also gives it “Knife” with a high enough score, the object is gone.

Families are our own grouping: Google only provides a flat list. Hover a bar for examples; the full detail is in the table of all 752 MIDs.

What Lens considers shoppable

• Four families hold 261 of the 364 shoppable objects: fashion (84), furniture and decor (66), kitchen (64) and sport (47). These are things to equip a home or oneself.
• All 20 musical instruments are shoppable, and no other family is.
• Food (82 objects) and animals and plants (115) have no shoppable object at all. Lens recognises them, but an apple or a dog calls for an information search rather than a purchase.
• Only 9 electronic objects out of 49 are shoppable: Smart TV, television, smartwatch, activity tracker, microphone, binoculars, tripod, jukebox and infrared thermometer. Phones, laptops, headphones, consoles and cameras are neutral.
• Vehicles, people, places and signs have no shoppable object: the Lens table gives no Shopping chip to a car or a motorbike.

The Lens Lite list keeps only 68 objects, against 364 shoppable objects in the Lens table. Of those 68, 40 are fashion items; furniture and decor come next with 9 objects. Lens Lite also has a minimum score of 0.3 and a filter that spots the text of an online shop.

The list does not quite match the Lens table: 11 of its objects are neutral there, including beer, the tin can, cream and headphones.

The shop-text filter

\$ \d+|\$\d+|(?i:shipping)|\b(?i:shop)\b|(?i:sale)|(?i:add to cart)|USD|(?i:in stock)|(?i:low stock)|(?i:out of stock)|\%(?i:off)|\% (?i:off)

The shop-text filter catches a dollar price, the words shipping, shop, sale, add to cart, in stock, low stock, out of stock, the string USD and discounts (% off). Together with the object list, it most likely serves to recognise an online shop page or a screenshot of one.

The filter only knows English and dollars: “Ajouter au panier”, “auf Lager” or a price in euros go unnoticed.

The app only reads the Lens table when no other setting gives it a list of objects. In every case we observed, a list was provided, so the table was not used. On the phone, two filters run over the detected objects, with the settings in the table below.

The first filter therefore does no sorting by category. A fruit, a dog or a smartphone are kept as soon as their score exceeds 0.35. The lists only come into play in the second filter. An object that fails it stays on the list and only loses its flag. The first chip shown comes from the server, which puts it on the object it judges most salient in the image, when it picks one.

The 2026 list drops kitchenware and adds animals and vehicles

The 2026 list names 431 objects that get no chip, 18 of which deny the object a chip even if it carries other labels: weapons, sex toys, the syringe, the crutch and the lifejacket. It holds no allowed objects. Any object missing from the list can get a chip, which leaves 272 of the 703 objects the detector knows.

These 272 objects only partly overlap with the Lens table’s shoppable objects. The 2026 list takes the chip away from 205 of them: almost all of the kitchen (except the espresso machine, kitchen appliances and the refrigerator), every instrument except the piano, all of the stationery and almost all appliances.

It gives one, on the other hand, to 135 objects that are neutral in the table: 107 animals and plants, vehicles (car, motorbike, boat, train…), a few fruits (grape, peach, pumpkin…) and three toys (doll, teddy bear, toy vehicle).

Table: the Lens table’s verdict. 2026: status in the 2026 list (eligible, neutral or blocked; “not detected” = general category the detector never returns). Lite: one of the 68 Lens Lite objects. Labels are Google’s originals.