ECOMMERCE

Extract product attributes from a spec sheet

Use when a supplier sends unstructured specs and you need clean attribute values for the feed.

attribute-extraction-spec-sheet.md

Download .md

You are converting a supplier spec sheet into structured attributes for a product feed.

Target attribute schema (use these field names exactly):
{{ATTRIBUTE_SCHEMA}}

Units required: {{REQUIRED_UNITS}}

Source text:
{{SPEC_SHEET_TEXT}}

Output a table: Field | Value | Unit | Confidence (high, medium, low) | Source phrase quoted from the input.

Rules:
- One row per field in the schema, in schema order, even if empty.
- If a field is not in the source, put "NOT FOUND" in Value and leave the source phrase blank. Do not infer it from the product name or from typical values for this product type.
- Convert units where required and show the original value in the source phrase column so the conversion can be checked.
- Mark Confidence low wherever the source is ambiguous, contains a range, or lists more than one candidate value.
- Where the source gives a range, keep the range. Do not pick the midpoint.
- After the table, list any values in the source that do not map to a schema field, so the schema can be extended.
- ASCII output only, no thousands separators in numeric values.

Fill in before running

Replace each placeholder with your own detail. The more specific you are, the less the model invents.

{{ATTRIBUTE_SCHEMA}}
{{REQUIRED_UNITS}}
{{SPEC_SHEET_TEXT}}

Getting a better result

The source phrase column is the whole point - it lets you spot check 20 rows in a minute instead of re-reading the PDF.
Run one product through first and correct the schema before batching the rest.
Treat every low confidence row as a manual check, not a suggestion.

Questions about this prompt

When is this better than typing the values into the feed?

When supplier specs arrive as prose or a PDF table and the units do not match what your feed requires. If the supplier already sends a structured file, map the columns instead. This earns its place where values are buried in sentences and somebody would otherwise be reading and converting by hand.

What do I need before batching a supplier sheet?

Your exact field names in {{ATTRIBUTE_SCHEMA}}, the units the feed requires, and the source text. Run one product through first and correct the schema before batching, as the tips say, because a field name that is nearly right produces an entire run to redo rather than a single row.

What does the table give me to check against?

One row per schema field in schema order, with value, unit, confidence and a quoted source phrase, then a list of source values that map to no field. The source phrase column is what makes the run checkable: twenty rows spot checked against it beats rereading the whole PDF.

What do people misread in the output?

Treating a low confidence row as a nearly right value. It means the source was ambiguous or offered more than one candidate, so it is a manual check rather than a suggestion. Watch that ranges stay ranges as well. A midpoint reaching the feed is a specification you invented and can be held to.