Customers search with a photo,
not a guess at the right keyword

When a customer has a picture, not a keyword

A customer sees a product they want somewhere: in a photo a friend sent, on social media, in a store they walked past. They try to find something similar in your catalogue. A text search box needs the right keywords to work. Most of the time the customer does not have those keywords. They have a picture in their head and a photo on their phone. Text search cannot bridge that gap, so the customer either gives up or scrolls through categories hoping to spot it.

The SEO and keyword gap compounds it, for products that are hard to describe precisely: a specific pattern, a particular silhouette, a color with no standard name. Even a motivated customer struggles to type the right search term.

Then there is the lost signal. Every one of these failed text searches is real demand the catalogue had an answer for, but the search experience could not connect the two. That gap rarely gets measured, because there is no easy way to see what a customer was trying to find with an image.

How the agent matches a photo to a SKU

It builds a visual index of your catalogue from existing product photos, encoding what each product looks like, not only its text description. A photo upload option then appears on your storefront or app search. When a customer uploads a photo, it matches against the index and returns ranked results by visual similarity. A confidence score keeps results honest about how close a match actually is.

As new SKUs are added through your normal product feed, the index updates automatically. Visual search stays current without a manual re-index step every time the catalogue changes. Where no strong visual match exists, the experience falls back to text search, rather than showing a confident-looking but wrong result.

It typically connects to your product feed or PIM, the source of truth for the catalogue. The photo upload option then appears on your storefront or app’s search interface.

What stays with your merchandising team

Deciding how prominently to surface visual search in the shopping experience is a call your merchandising team makes. So is interpreting what the search analytics reveal about unmet demand. The agent returns ranked matches by appearance. It does not decide pricing, placement, or which gaps in the catalogue are worth filling.

Safeguards built in

Similarity thresholds start conservative, so a weak match falls back to text search rather than being shown as a confident result. They get tuned against your actual catalogue before launch, not a generic benchmark. Search analytics are logged in aggregate, so your team can see demand patterns without the pipeline retaining customer-uploaded photos longer than the search itself requires.

Before it runs unattended, we test it against a sample of your own catalogue so your team sees exactly what it would have matched. Every build ships with a short written runbook, so your team can pause it, adjust a threshold, or roll it back without waiting on us. The running-cost estimate below is a starting budget you set, with an alert before it is crossed.

Price and timeline

Running cost is usually $10 to $80 a month in model usage depending on volume, with a budget cap set before launch.

Related

Pair this with product tagging for real estate listings if your catalogue spans listings rather than SKUs. Product photo quality check keeps the indexed photos themselves consistent. For the product data side of search relevance, see product descriptions at scale. The full package breakdown is on the AI agents service page and the e-commerce service page. For a real build of a catalogue-heavy storefront, see the Telegram marketplace engine case study.

Ready to let customers search with a photo instead of a guess? Get in touch and we will test it against a sample of your catalogue in the first call.

Tired of doing this by hand? We can take the whole routine off your team, not only this step: Routine takeover, from $400 →