Image search in shopping apps: 4 practical use cases
From photo to product: practical uses for image search in shopping apps
Four practical ways shoppers use a photo to find products: find this exact item, find something similar for less, identify a replacement part, and match a style. With input tips, confidence handling and when to add a text query.
A camera button in a shopping app looks like one feature, but shoppers use it for different jobs. Someone photographing a friend's lamp wants that lamp. Someone photographing a broken blender jug wants the part that fits. Treat these as the same request and the results will feel random. This guide covers four common uses for image search, how to get better photos from shoppers, and how to handle uncertain matches.
If you want the mechanics of how a photo becomes structured product candidates, read
from a product photo to structured candidates
first. This post is about product design: what shoppers are trying to do and how your app should respond.
What do shoppers use image search for?
Find this exact item
"I saw this lamp at a friend's house." The shopper wants the same product, ideally with a price and somewhere to buy it.
Find something similar for less
"I like these trainers but not the price." The shopper wants look-alikes in a lower price band.
Identify a replacement part
"My blender jug cracked." The shopper needs a part that fits, so compatibility matters more than looks.
Match a style
"Cushions that go with this sofa." The photo sets a colour and mood, not a product type.
Designing each flow
Each job needs a different result layout and a different idea of success.
Job
What success looks like
How to show results
Find this
The same model, confirmed by identifier or clear details
One top candidate with evidence, then "other possible matches"
Similar for less
Visually close items below a price ceiling
A varied list sorted by price, each with a dated price
Replacement part
A part that fits the original product
Ask for the model number before recommending anything
Style match
Items in a different category that suit the photo
A small, varied set with a line on why each fits
You can ask the shopper which job they mean with one tap ("Find this", "Find cheaper", "Find a part", "Match the style"), or infer it from what they type alongside the photo. Asking is usually faster than guessing wrong.
Replacement parts need words as well as pictures
Parts are where image search alone is weakest. Two vacuum filters can look almost identical and fit different models. Use the photo to narrow the category, then ask for the model number on the appliance's label. If the shopper provides an identifier, an exact lookup is more reliable than any visual match. Never tell a shopper a part fits unless the product data states that compatibility.
Helping shoppers take a useful photo
One product per photoAsk the shopper to crop or tap the item they mean when the scene is busy.
Fill the frameThe product should take up most of the image, not a corner of it.
Even lightAvoid strong shadows and colour casts that change how the item looks.
Show the distinctive partLogos, stitching, handles or ports often separate one model from another.
Add the label for partsA second photo of the model plate helps more than a better angle.
Handling confidence honestly
Image search returns candidates, not certainties. Each result should carry a relevance score, and your app should treat a strong visual match and a confirmed match differently.
Confirmed: an identifier or several specific details match. Say "This looks like the same product" and show the evidence.
Likely: strong visual similarity, no identifier. Say "Closest matches" and show two or three options.
Weak: low scores or very mixed results. Ask a follow-up question or suggest a clearer photo instead of guessing.
Empty: no usable results. Say so plainly and offer text search.
Set your thresholds by testing real photos, not by picking a number. Our guide to evaluating product search quality covers building a test set.
When should you combine a photo with a text query?
A photo is good at showing shape, colour and style. Words are better at constraints: size, budget, material, use. Combine them whenever the shopper has a constraint the photo cannot show. "These trainers, but waterproof" or "this jacket in a size 12 under £60" both need the text.
With Simlir, the image search endpoint takes a public HTTPS image URL, a market and an optional text query, and returns structured products with titles, identifiers, specifications, images, dated prices and retailer links. Each image search costs 2 credits; a text search costs 1. If the shopper has typed enough to describe the product without the photo, a text search may be the better first call. See natural language product search for how text queries handle intent.
Code examplejson
POST /api/v1/search/image
{
"image_url": "https://example.com/photos/lamp.jpg",
"market": "gb",
"query": "brass table lamp under £80",
"limit": 5
}
A limit is a ceiling, not a promise. Expect fewer results, or none, and design for that. Full parameters are in the Simlir docs and credit costs are on the pricing page.
Frequently asked questions
Can image search confirm two products are identical?+−
Not on its own. Visual similarity suggests a match; an identifier such as a GTIN or model number confirms it. Show the difference to shoppers.
Should the shopper upload a photo or paste a link?+−
Either can work for the shopper. The API needs a public HTTPS image URL, so uploaded photos must be stored somewhere reachable first.
What should happen when no results come back?+−
Say so plainly, suggest a clearer photo, and offer a text search. Do not fill the gap with unrelated products.
How much does an image search cost compared with a text search?+−
With Simlir, each image search costs 2 credits and a text search costs 1. If the shopper's words already describe the product well, starting with a text search is cheaper.
