When customers send a photo instead of a question: answering “do you have this?” from your catalogue
When a customer sends a product photo in a direct message, AdaptiveReply’s assistant reads the image, looks for the product in your own catalogue and answers with its name and price, then confirms with one question. It works from exact to similar, and when it cannot verify a match it asks instead of guessing.
By the AdaptiveReply teamPublished 7 min read
Key takeaways
- A photo is a question: “do you have this, and how much is it?”
- The assistant matches the photo against your own catalogue in a fixed order: product code, written product name, then visual similarity.
- When a match is not verified, the assistant quotes no name or price; it asks for a clearer photo or a product code.
Why do photo messages stall a team?
A customer sends a screenshot of a scarf with one word: “price?”. Nobody typed a product name, so there is nothing to search for. Someone on the team has to recognise the scarf, scroll the catalogue to find which of forty similar ones it is, check the price and write back. During the day that takes minutes. At night it waits until morning, and by then the customer has asked two other shops.
Photo messages are common wherever products are visual: fashion, home, cosmetics, jewellery, spare parts. Customers screenshot a post, a story, an ad or a friend’s purchase, because showing is easier than describing. For the team, each one is a small lookup job that only the person who knows the catalogue by heart can do quickly.
How does the assistant read a customer’s photo?
AdaptiveReply treats an incoming image as a lookup against the catalogue connected to the assistant, not as a picture to describe. Visual product matching is the process of finding which catalogue item a customer’s photo shows. The assistant works through the possibilities in a fixed order, from the most certain to the least:
- Is it a product at all? Receipts, invoices and shipping labels are recognised as documents and are not matched to products.
- Is there a product code on the image? A code that exists in your catalogue gives an exact match, with no guessing.
- Is there a product name written on it? Many screenshots carry one. The assistant searches the catalogue for that name.
- No text at all? The assistant compares the photo with your product images and keeps only the candidates it can verify as likely matches.
What does the customer see?
The customer sees a normal, confident reply: the product by name with its price, and one question to confirm. The assistant does not send the customer off to compare links, and it does not pass its own uncertainty on as hedging.
| What the assistant found | What it replies |
|---|---|
| A product code that is in the catalogue | It answers from that exact product: name, price and whether it is available. |
| One likely match | It names the product with its price and asks one confirmation question, mentioning a distinguishing detail such as colour or pattern. |
| Several likely matches | It lists them by name and price and asks which one the customer means. |
| A written name that fits several items | It shows the options from the catalogue and asks which one. |
| The customer says “no, not that one” | It asks for a clearer photo or a product code. |
Once the product is confirmed, the conversation continues like any other: variants, availability, the purchase link and, if the customer wants to buy, a recorded request for your team to confirm. Availability means “available” or “not on sale”; the assistant does not quote stock counts. If it sends a product photo back, the image comes from your catalogue and nowhere else. The rest of that journey is covered in the Instagram DM guide.
What happens when the photo is not a product?
When the image is a payment receipt, an invoice or a shipping label, the assistant does not try to match it to a product. It asks politely what the customer needs and offers to pass the document to your team.
This matters more than it sounds. Text on a document can look like a product name, and an assistant that guesses would answer a bank receipt with a sweater. A document usually means a transaction is in progress, which is work for a person. The conversation can then be handed to your team as a case with a summary, as described in the handoff guide.
What happens when nothing matches?
When the assistant cannot verify a match, it does not name a product, quote a price or share a link. It asks for a closer photo or a product code, and looks again when the customer sends one.
There are three usual reasons for a miss:
- The product is not in your catalogue, for example an item from another shop or one you no longer sell.
- The photo is dark, cropped or shows the product from far away.
- The catalogue item has no image, so there is nothing to compare the photo with.
What do you need for photo answers to work?
You need a catalogue with product images, connected to the assistant on the Sources tab. Nothing has to be tagged or trained by hand. A short checklist:
- Connect your store (Shopify, WooCommerce, ikas or Ticimax) so products, prices and images sync, or import your catalogue from an Excel or CSV file.
- Check that every product has at least one clear image; a product without one can only be found by code or by name.
- Keep product codes in the catalogue if customers see them on labels or packaging.
- Send five real customer screenshots in the test chat and read what comes back.
- Add a rule for media the assistant cannot read: “WHEN the customer sends a video, DO hand off to the team”.
One limit to state plainly: the assistant reads images only. The contents of videos, voice messages and files are not read, so decide by rule what should happen to them. How the catalogue connection works is explained in the store catalogue guide, and typical scenarios are shown in the use cases section.
Frequently asked questions
Can an AI assistant recognise a product from a screenshot?
Yes. AdaptiveReply’s assistant first reads any text on the screenshot. A product code found in your catalogue gives an exact match, and a written product name triggers a catalogue search. Only when there is no text does it compare the picture with your product images and propose the closest items.
Is the match always exact?
No. A match by product code is exact. A match by visual similarity is a proposal: the assistant names the closest catalogue item with its price and asks one confirmation question, or lists several candidates and asks which one is meant. If the customer says no, it asks for a clearer photo or a code.
Does the assistant understand videos or voice messages?
No. AdaptiveReply’s assistant reads images only; the contents of videos, voice messages and files are not read. You can add a rule so that when a customer sends a video or a voice message, the conversation is handed to your team instead of being left unanswered.
Do I need to tag or label product images?
No. There is no tagging or training step. The products need images in the catalogue, which arrive automatically when you connect Shopify, WooCommerce, ikas or Ticimax, or which you add to the items of a catalogue you manage yourself. Products without an image can still be found by code or by name.
This guide is part of a series. Start with the overview: One inbox for Instagram, WhatsApp, Facebook, email and web chat: why scattered channels lose customers