Extract product attributes and specs from supplier PDFs

Confidence-scored, traceable extraction

Review extracted values alongside the supplier document. The extraction’s confidence score and the page-and-quote evidence help you decide what to check first; they are not a guarantee that a specification is correct.

Check model numbers, units, and which product a specification belongs to before reusing the data. If the source is missing a value or gives conflicting values, resolve it with the supplier instead of treating an AI inference as a confirmed fact.

Extract product specifications and attributes from supplier PDFs

Doc2Shelf extracts product attributes from datasheets: model numbers, dimensions, materials, capacities, ingredients, and other product-specific fields. For example, an illustrative datasheet stating model AX-100, stainless steel, and capacity 1.5 L provides three separate values to verify: model, material, and capacity with its unit. This example is not a measured extraction result.

A product data sheet describes a product and its technical specifications. To extract product specs from a PDF reliably, check table headings and footnotes: one datasheet may describe several models with different dimensions or capacities. Keep each value associated with the correct model before preparing a listing or declaration draft.

From PDF to product listings, declaration drafts, and DPP-oriented data

Extraction is the foundation for everything else. The same reviewed structured data can support Shopify and WooCommerce listings, reviewable eBay-targeted draft copy, EU/US declaration drafts, applicable GPSR content, and separate DPP-oriented structured JSON for downstream implementation.

Export the raw structured data as JSON or CSV if you want to use it in your own systems, or use it to prepare listings and declaration drafts for review.

Built for scale

Upload up to 500 PDFs at once, within your plan’s document allowance, and let the background pipeline extract them in parallel. Reused extractions and consistent structure keep large catalogs coherent, and the review gate keeps quality high before anything downstream is published.

Which supplier documents work best

Doc2Shelf extracts from the PDF files you upload; it does not crawl supplier portals or fetch documents for you. Digital and scanned PDFs can both be read, though poor scans lower the extraction confidence, and a document that covers several products produces several product records.

What you can expect to get out of a file depends on what kind of document it is:

What gets extracted and where it is used

Each extraction returns product records organized into the same field groups, whatever the supplier’s layout. The groups map directly onto the content Doc2Shelf prepares next, so a value checked once in review is reused everywhere it is needed.

Alongside the structured fields, the extraction keeps its evidence: a list of facts with the page number and a short verbatim quote, every data table reproduced row by row, facts that fit no standard field, and a list of required information the document did not contain. Information that is still missing when listings or declaration drafts are generated appears as a [REQUIRED: …] placeholder instead of an invented value.

Confidence scores, evidence, and the review gate

Every extraction carries a confidence score and a completeness score, with notes on caveats such as illegible areas, ambiguous values, or scan quality. Use them to decide where to look first. A high score does not make a value correct, and a low score usually points to a document problem — a poor scan, an unclear table — that may be worth fixing at the source.

Nothing extracted is published or exported unreviewed. When you mark content as reviewed, Doc2Shelf records a content hash; any later edit changes the hash and invalidates the review, so publishing and export stay blocked until someone checks the changed version. A practical review order:

• Start with the documents that have the lowest confidence or completeness.
• Check identifiers and safety-relevant values against the quoted evidence first.
• Confirm that every value from a multi-model table sits on the right product.
• Resolve flagged missing data with the supplier rather than filling it in.
• Mark the content as reviewed, and review again after any edit.

Batch extraction and output formats

Upload up to 500 PDFs in one batch, within your plan’s document allowance, and extraction runs in the background. Built-in templates focus the extraction on a document type — Food label, SDS / chemical, Electronics spec sheet, or General product — and you can choose focus areas, add custom instructions, switch on an audit mode that captures every fact and table row with a quote, and, depending on your plan, save your settings as a template for the next supplier.

Values are kept in the document’s language, with units exactly as printed, so nothing is silently converted between the source and your review. Once reviewed, the data leaves Doc2Shelf in the format each workflow needs:

How it differs from generic PDF parsers and OCR tools

Generic PDF-to-spreadsheet tools and OCR engines solve part of the problem: they get text and tables out of a file. Turning that text into product records — and those records into listings and compliance drafts — is the part they leave to you. The comparison below is about the type of tool, not any specific product.

A generic tool is the simpler choice when you only need one clean table copied into a spreadsheet, or when you process large volumes of one fixed form. Doc2Shelf is built for the opposite case: many suppliers, varied layouts, and data that has to end up in listings and declaration drafts. The guide to extracting product data from PDFs compares all the methods in more detail.

Data quality and limits

AI extraction can miss a value, misread one — especially in poor scans, dense tables, or values shown only in drawings — or attach a value to the wrong model in a multi-product table. That is why the review gate exists and why the extraction keeps evidence you can check. Know these boundaries before you plan a catalog migration:

• It extracts only what the document states; it does not look up missing values elsewhere or estimate them.
• Units and values are kept as printed. Converting units and normalizing value lists for your channels is part of your review; the guide to product attributes covers how.
• A SKU is captured only when the document explicitly labels one, never derived from a file name or model number.
• Standards, markings, and certificate references are recorded as stated. Doc2Shelf does not test products, assess conformity, or verify certificates.
• Text stays in the document’s language; listing translation is a separate step that produces review-ready drafts.
• Uploads are PDF files; other formats need to be saved as PDF first.