Machine-readable product data for AI-ready chat shops

A practical guide for Bangladeshi chat sellers on turning scattered product posts into clean, labelled data, so customers and AI assistants get fast, correct answers on size, color, and stock.

O
OrderPilot
6 min readNovember 16, 2026
A flat lay of a folded panjabi, black sneakers, and three lipsticks, each with a small blank tag, beside a smartphone on a wo

Machine-readable product data means each product detail sits in its own labelled field instead of a caption or a photo album. When a customer asks if an item comes in their size, a clear record gives a fast, correct answer, whether you or an AI is replying. Here is how to set it up.

Why chat sellers end up with messy product data

Most sellers start with whatever is quickest: a Facebook album with captions, or a notebook beside a few photos. That works while you are the only one replying, because you remember that "the red one" means the red polo in size L.

Two things go wrong as the shop grows. Details live inside free text, so one line mixes the name, size, color, and price. And the same detail gets written several ways, such as "Blue", "blue", and "Blue / M". A person reads past that. Software cannot reliably tell that those mean the same thing.

What messy product data costs

Customers write short, messy messages, often late at night, and a message that waits is usually a lost sale. If the answer means scrolling an album or flipping through a notebook, replies slow down. When an order does arrive, details like size, color, and quantity get lost or confused, and then you retype everything by hand once the chat is over.

Fixing the order side is a separate job, and how to stop losing orders in WhatsApp and Messenger covers it: give every order one fixed place to live the moment a customer says yes. Clean product data does the other half. It makes the customer's question answerable in the first place.

Why clean product data matters more now

Shoppers are starting to ask AI tools for help, and those tools work best on clear fields. A field reading "fabric: cotton" can be read directly. A caption has to be interpreted, and interpretation is where wrong answers come from.

Shopify reported that AI-driven traffic to its stores grew eight-fold year over year in the first quarter of 2026, and that orders from AI-powered searches rose nearly thirteen-fold. That is Shopify's global data, not Bangladesh's. The direction still applies to you: whoever reads your product information next, a customer's AI tool or your own AI assistant, needs it written down in a structured way.

You do not need technical website markup or an expensive product information system to start. For a chat seller, machine-readable simply means a tidy catalogue with labelled options and one record per variant.

Machine-readable product data, step by step

Each step below turns one more detail from a sentence into a labelled field.

Give each product one clear name

Use a plain name such as "Cotton Panjabi" and move the marketing lines and emoji into the description. The name is what you and the AI use to find the product, so keep it short and identical every time.

Split options into their own fields

Do not bury choices inside the name. Create separate options, such as Size and Color for a panjabi, or Fabric and Color for a sharee, and give each a fixed list of values. Pick one spelling and keep it: "ml" everywhere, never "ML" on one product and "milliliter" on another. Consistent values let a search or an AI match a customer's words to the right item.

Create one variant for every combination

A phone case in the same color for two different phone models is two different things to a customer, because one may be in stock and the other sold out. Give each combination its own row with its own SKU, price, stock count, and availability.

Keep stock and price current. An out-of-date stock count produces a wrong answer delivered with full confidence, which is worse than no answer.

Attach a photo to each variant

When a customer asks to see the coral lipstick, the picture that goes out should be the coral one. Tie each photo to the variant it shows, not to the product as a whole. Otherwise someone has to guess which of ten images matches the request.

Automate your catalogue instead of retyping

Once the structure is clear, put it in one spreadsheet with one row per variant and columns for name, each option, SKU, price, and stock. If your products already live on Shopify or WooCommerce, export that file and start from it instead of typing everything again.

Before importing, count the empty cells in each column. Blanks are where you, or an AI, will have to guess. Fill them first, then import the file in one go.

Retyping is where small mistakes creep in, and the same applies to orders. How OrderPilot eliminates repetitive manual workflow friction explains why manual entry in chat commerce causes delays and shipping errors.

A worked example: black sneakers, size 42

A customer writes: "bhaiya, black sneaker 42 ache? 1 jora lagbe".

With clean data, the record behind that message is one variant: Sneaker, color Black, size 42, with its own SKU, price, stock count, and photo. You check one row, say yes or no, send the photo if asked, and note a quantity of 1.

With the old setup, you scroll to find the album, locate the right post, and hope the comments say whether that size has sold out. The question is the same in both cases. The structure of your data decides how long the answer takes and whether it is right.

How to tell it's working

Three signals are easy to check:

  • You can answer a stock question from one place without scrolling.
  • Orders arrive with size, color, and quantity already clear, so you stop sending follow-up questions like "kon size?".
  • A photo request takes one step because the right image is already attached to the variant.

There is a longer-term benefit too. Once product and order details are saved cleanly, you can see what each buyer bought, which makes it easier to follow up with a reason to come back. Repeat customers from chat orders: 6 steps to start shows how.

OrderPilot's AI answers customers from your catalog, shares the exact variant photo, and can import your products from a CSV or XLSX file, including Shopify and WooCommerce exports.

When this doesn't apply

If you sell one or two items with no options, such as a single product in one size, a full variant setup is more than you need. A clear name, a price, and a photo are enough. The same goes for made-to-order work where each item is priced per request. In that case, capture the details of each order in the order notes instead.