Retail and e-commerce

Write product descriptions for a whole catalogue with ChatGPT

Build one reusable prompt, then turn a product sheet into clear descriptions that your team checks against the facts before anything goes live.

Published 25 Oct 2026 · 7 min read
← Back to blog
Tools used
ChatGPTClaudeGeminiGmailGoogle SheetsGoogle DriveGoogle Docs

A small online shop in Bengaluru or Surat can have 300 products with two-line descriptions, or none at all. Writing each one by hand takes a whole week, and the copy often says different things about the same fabric or size. This guide shows how to write one brand voice rule, test it on a few products, then run a batch through ChatGPT, Claude or Gemini in Google Sheets. Every description is checked against the product row before it is published.

What you need

  • A product sheet with one row per product: SKU, product name, material, size, colour, price in ₹, care instructions and two or three key features
  • Three descriptions your brand already likes, copied from your website or from older listings
  • A paid business account in ChatGPT (Business or Team), Claude (Team), or Google Workspace with Gemini turned on by your admin
  • Someone on the team who reads Hindi and English, for the final check

Step 1: Clean the product sheet firstGoogle Sheets

The AI can only write what the sheet says. Fill in the blanks before you start. A missing size or material should be a visible gap in the sheet, not something the AI guesses. Add a column called "Reviewed" and leave it empty for now. Keep the price in one place only, the sheet, so the descriptions never carry an old price.

Step 2: Capture your brand voice as rules

Do not write a long brand guide. Ask the AI to read your three good descriptions and turn them into a short list of rules.

Prompt

Read these three product descriptions that we like. List the tone rules they follow, such as sentence length, whether we use Hindi words, whether we mention care instructions, and what we never say. Give the rules as a numbered list of no more than 8 items. Descriptions: [paste the three descriptions]

Read the list and delete any rule you do not agree with. Keep the final list in a file called "Brand voice". You will paste it into every batch.

Step 3: Test one product

Before running the whole catalogue, test one product. Paste the brand voice rules and one product row.

Prompt

Using the brand voice rules below, write a product description for [product name] in [English / Hinglish] in 60 to 80 words. Use only the facts in this row: [paste the product row]. Do not claim anything the row does not support, such as best quality, bestseller or medical benefits. End with the care instructions from the row. Brand voice rules: [paste rules]

If the first draft adds a claim, add a rule to the brand voice list, such as "No claims that are not in the row", and test again.

Step 4: Run a batch in Sheets or in a chatChatGPTClaudeGeminiGoogle Sheets

You have two routes. Route one is a chat in ChatGPT or Claude: export the product sheet as a CSV, upload it to a project, and paste the batch prompt. If your store runs on Shopify, Matrixify's walkthrough shows the export and re-import steps for this same route. Route two is Gemini in Google Sheets. Google's Workspace update describes a side panel that can help you organise data and generate formulas inside Sheets. Button names and positions change between versions, and some Workspace accounts need an admin to turn Gemini on, so look for it in your own Sheets menu. Work in batches of 10 to 20 rows either way.

Prompt

For each row in the table below, write a description that follows the brand voice rules. Keep each description under 80 words. Start each line with the SKU so I can match it back to the sheet. If a row is missing a size, material or price, write "MISSING: [field]" instead of a description. Brand voice rules: [paste rules]. Table: [paste 10 rows]

Paste the output into a new column in the sheet, next to the SKU column. Do not paste the descriptions back into the product list until they are reviewed.

Step 5: Check every description against its rowGoogle Sheets

This is the step that protects you from wrong claims. Run a second prompt over the output, with the original rows beside it.

Prompt

Check each description against its product row. List any claim the row does not support, any superlative, and any size, material, colour or price that differs from the row. Return a short list by SKU. Say "no issues" for the SKUs that pass. Rows and descriptions: [paste both]

Then read the flagged items yourself. Mark each row "Reviewed" in your sheet only after a person has read it.

Step 6: Get the Hindi or Hinglish copy checked

If you publish in Hindi or Hinglish, ask the AI for a first draft, then have a fluent team member check the wording for natural phrasing. Do not publish machine text in Hindi that nobody has read.

Prompt

Rewrite the English description below in simple Hinglish for a customer in [city], keeping every fact exactly the same and keeping the price out of the text. Description: [paste]

Step 7: Publish in batches and keep a recordGoogle Sheets

Import the reviewed copy into your website or marketplace's bulk upload file, batch by batch. Keep the sheet as the master record, with the SKU, the date reviewed and the name of the reviewer. When a price or size changes, update the sheet, then regenerate only that row.

Tips

  • Start with your best-selling 10 products. Mistakes there cost the most, so they get the most review.
  • Ask for 60 to 80 words. Longer copy is harder to check.
  • Keep product names in the sheet as the source. The AI should not rename a product.
  • Do not invent review quotes or ratings. Ask for them only if you already have real ones, pasted in.
  • Track how many rows were flagged in each batch. A high count means the sheet needs cleaning.

Keep it safe

Your product sheet may include supplier cost prices, so treat it as confidential. Use your company's business account, not a personal login. Do not paste customer names, addresses, phone numbers or order details into any AI tool. Google's help page on Gemini in Workspace says Workspace data is not used to train the underlying models, and OpenAI's help article on ChatGPT Business explains how business workspace data is handled. Read your provider's current terms before you decide what your team may upload.

Sources

More for Retail and e-commerce: see what AI can do for your industry →

Cut costs. Get hours back.

In 15 minutes we show where AI saves your team the most time and money.

Get my savings plan