The fastest way to create a product catalog is to describe it. CSV Pilot turns chat messages into structured Hyperzod-compatible CSVs: you type what you want, the AI builds the products, and you review and export. No column names, no template downloads, no spreadsheet formulas. This guide explains the flow in plain terms, with the prompts that produce the best results.
What happens when you type a request
When you send a message like "create 5 pizza products with sizes", the agent does three things:
- Understands the intent — this is a creation request for products with variants.
- Builds the structure — products, prices, categories, and option groups, using the Hyperzod column format.
- Shows the result — the preview grid fills with editable rows and validation results.
The AI writes the data; the tool enforces the structure. You never see a column name unless you want to. This split matters: language models are good at interpreting "10 pizzas with sizes and toppings" and bad at reliably producing a valid CSV encoding — so the model handles the meaning and deterministic code handles the encoding.
Prompts that produce good catalogs
Good prompts are specific. Compare:
- Weak: "make a catalog"
- Good: "Create 10 grocery products for an organic store, categories like Beverages and Snacks, prices between ₹50 and ₹300, each with a size option."
Specificity to include: product types, quantity, price range, categories, variants, add-ons, and any naming style you want. The AI fills gaps with plausible values — and marks them as inferred so you know what to check.
More examples of effective prompts:
- "Create 6 burgers with add-ons: cheese +₹20, bacon +₹40, extra patty +₹80."
- "Create a winter menu with 12 hot drinks, ₹99–₹199, tagged seasonal."
- "Create 15 products for a hardware store, grouped into Tools, Fasteners and Paint."
Why the preview matters
The AI infers details you do not mention. If you say "create 5 pizzas" without prices, it will pick plausible ones — and mark them as inferred. The preview highlights those cells so you know exactly what needs your judgment. This is the difference between generation and automation: CSV Pilot generates a draft, and you approve it.
Fixing the output by chatting
You do not have to click every cell. Follow-up messages work:
- "Make the pizzas ₹100 cheaper."
- "Add a veg tag to all vegetarian items."
- "Remove the Pineapple topping."
- "Rename the category to Artisan Pizzas."
Each follow-up applies as an edit to the current catalog, shown in the same preview. The conversation is the editing interface — describe, review, refine.
When chatting beats typing
Chat generation wins when the alternative is manual entry:
- Building a catalog from scratch (tens or hundreds of products)
- Creating variants and add-ons without knowing the column format
- Drafting a menu to review with a client before committing
- Producing test data for a staging environment
It is not magic — it is a draft-and-review loop. Type a specific request, review the preview, refine by chatting, and export. The loop is the same whether the catalog has five rows or five hundred.