CSV Data Extractor

What This Prompt Does

This prompt extracts unstructured data from raw inputs (such as email chains, customer invoices, or text logs) and compiles it into a clean, RFC 4180 compliant CSV table. It handles text normalization, escapes commas, and structures values under requested column headers.

System Prompt

You are a data processing and validation engineer. Extract structured information from the provided raw text and format it as a valid, comma-separated CSV block.
Adhere to these rules:
1. Output ONLY the raw CSV text block, using double quotes to escape commas or quotes in values.
2. Do not include markdown code block syntax inside the CSV output if requested.
3. Validate that every row has the exact same column count as the header.
4. Replace missing values with null or empty fields consistently.

User Prompt Template

Extract structured data from the following raw text:
{RAW_TEXT}

Columns to generate: {TARGET_FIELDS}
(e.g. "Name, Email, SignupDate, MonthlyCost")

Ensure all values are escaped correctly.

Example Output

"Name","Email","SignupDate","MonthlyCost"
"Sarah Connor","sarah@sky.net","2026-05-12",150.00
"John Connor","john@sky.net","2026-06-01",0.00

When to Use This

This prompt helps when you need to pull structured fields out of messy, inconsistently formatted CSV data — exported reports with irregular column naming, merged cells, or embedded free-text fields that need parsing into clean, structured columns before further analysis.

Tips for Best Results

• Include a representative sample of the actual messy input, including its inconsistencies, rather than a clean idealized example — the model needs to see the real formatting problems it will need to handle.
• Specify the exact target schema (column names, types, how missing values should be represented) explicitly rather than leaving the output format to be inferred, since ambiguity here is the most common source of inconsistent results across multiple runs.
• For large files, process data in chunks and validate a sample of the output against the source before running the full extraction, since an error in the extraction logic that goes unnoticed early can propagate through an entire large dataset.

Keeping a small regression set of known tricky input rows lets you quickly re-verify extraction quality whenever you tweak the prompt or switch to a different underlying model.

This is a cheap safeguard against silent quality regressions that would otherwise only surface once bad extracted data has already propagated downstream.

It also makes it far easier to catch a subtle prompt regression before it ships, rather than discovering it after a large batch has already been processed incorrectly.