How do you generate thousands of product descriptions from a spreadsheet?

The workflow at a glance

To generate thousands of product descriptions from a spreadsheet, export the catalog as CSV, map its columns to prompt variables, run the prompt across every row, review flagged exceptions, and export the finished descriptions for import into your shop, PIM, or ERP.

1. Prepare and clean the spreadsheet.
2. Map source columns to prompt variables.
3. Run a test, then process the full batch.
4. Review samples and flagged exceptions.
5. Export and import the completed CSV.

Step 1: Prepare the spreadsheet

Use one row per product or variant. Include a stable unique identifier, such as an SKU or product ID, so each generated description remains connected to its source record.

Keep consistent columns for useful attributes such as product name, brand, category, material, size, color, key features, and target keywords. Remove duplicate and empty rows, use one header row, and avoid merged cells.

Save the file as CSV. A single header row, one record per line, and consistent quoting make spreadsheet-to-product-descriptions workflows easier to import and export predictably.

Step 2: Map prompt variables

Map each required source column to a variable referenced by the prompt. For example, a prompt might use {{ product_name }}, {{ brand }}, and {{ material }} to create a description grounded in that row’s data.

Only map fields that contribute useful context. Write generated text to a new output column so the source data stays untouched. Test the prompt on five to ten varied rows, including records with sparse or unusual attributes, before processing the whole file.

Step 3: Run the batch

Run the reusable prompt or pipeline across selected rows or the full CSV. The system processes each row with its own values, removing the need to copy product details into a chat interface one at a time.

A CSV-based pipeline tool such as Bulkbase can map columns to variables, apply reusable prompts, and export the results. Parallel processing suits faster turnaround, while non-urgent runs can use the OpenAI Batch API. Bulkbase offers this through Eco mode, which has a longer processing window and 50% lower token costs.

Step 4: Review the output

Spot-check a sample from across the file rather than reviewing only the first few rows. Include different categories, attribute combinations, and records with missing values.

Use validators or filters to flag empty output, malformed formatting, length violations, omitted required attributes, or banned claims. Correct individual cells or rerun only the affected rows. Human review should still confirm tone and factual accuracy before publication.

Step 5: Export and import

Export the completed descriptions as a structured CSV while preserving the unique identifier and row alignment. Import that file through the CSV importer provided by your shop, PIM, or ERP.

Check a small sample of live product pages after the import. Confirm that descriptions reached the correct products and that formatting displays properly. Retain copies of both the source and output files so you can trace changes or roll back an incorrect import.

Common mistakes to avoid

• Skipping the unique identifier column.
• Using merged cells or inconsistent headers.
• Mapping unnecessary columns into the prompt.
• Running the full file before testing a small sample.
• Reviewing only the first rows in the output.
• Sorting or editing files in ways that misalign records.
• Publishing generated descriptions without factual review.

If you want to test this workflow with your own catalog, book a practical Bulkbase demo. Booking a demo is the first step to start or activate a free trial.