Copy written from the spec sheet
Because generation is anchored to extracted product data, the descriptions are written from the measurements, materials, capacities, and features in the product data you review. The aim is persuasive merchandising copy that does not drift from the facts — and the review step is where you confirm it.
Review the extracted product fields and any available source evidence before approving the listing. Generated listing sections remain editable drafts and do not link each claim to a source location.
Rewrite, restructure, and match your voice
Need a punchier title or a different tone? Rewrite individual sections with AI, adjust the bullet order, or edit the text directly. The listing structure — title, descriptions, bullets, specifications — stays intact while you refine the wording.
This makes it practical to produce consistent copy across hundreds of products and still tailor the hero listings that matter most.
One description, many languages
Sell internationally by translating listings into more than 100 languages while preserving structure and formatting. The same reviewed description becomes localized draft copy for each market you enter, without re-writing from scratch.
What goes into a generated listing
Each generated listing has the same six sections, so copy stays consistent across a catalog and every section can be edited, rewritten, or translated on its own. Generation works from the product’s extracted data plus the settings you choose: target platform, language, and optionally a saved brand voice or listing template.
Price, shipping, discounts, and availability are left out unless they appear in the product data — those belong to your store settings, not the supplier document.
Listing sections and where their content comes from| Section | What it contains | Where it comes from in the product data |
|---|
| Title | Plain-text product title in the listing language; brand names, model numbers, and codes kept as-is | Identity fields (brand, product name, model) and key attributes |
|---|
| Short description | One to three plain-text sentences summarizing the product | Product type, intended use, and headline attributes |
|---|
| Long description | Formatted body copy: paragraphs, sub-headings, lists, and tables where useful | Extracted features, materials, measurements, contents, and usage information |
|---|
| Bullet points | Typically four to seven plain-text selling points | Features and specifications in the extracted data |
|---|
| Specifications | Attribute and value table | Measurements, materials, ratings, standards, and other extracted attributes |
|---|
| GPSR information block | Manufacturer details, responsible-person details where the manufacturer is not established in the EU, and warnings that apply | Parties, addresses, and safety fields from the documents; gaps shown as [REQUIRED: …] placeholders |
|---|
Tone and length controls
Every section has an AI rewrite with four presets — shorter, longer, more professional, more casual — and a free-text box for a custom instruction such as “lead with the battery life” or “write for trade buyers”. A rewrite is shown as a proposal; nothing is saved until you accept it.
Rewrites are instructed to change style, structure, and emphasis only, keeping facts, figures, units, codes, regulatory wording, and [REQUIRED: …] placeholders. Treat the result like any other edit and read it before accepting: saving a change invalidates the listing’s review, so it has to be reviewed again before publishing or export.
Rewrite and voice controls| Control | What it does | Typical use |
|---|
| Shorter | Condenses a section while keeping its facts | Short descriptions, mobile-first pages, tight bullet space |
|---|
| Longer | Adds detail and explanation from the same content | Hero products and categories where buyers compare specs |
|---|
| More professional | Shifts to a formal, precise register | B2B catalogs, technical and industrial products |
|---|
| More casual | Shifts to a conversational register | Consumer brands, lifestyle and gift products |
|---|
| Custom instruction | Applies your own written instruction to one section | Reordering emphasis, audience changes, house style rules |
|---|
| Brand voice | Applies a saved tone, style guidelines, and example phrases at generation | Keeping a large catalog in one consistent voice |
|---|
Formatting for each channel
You choose the target platform when you generate a listing, and the copy is shaped for it. The facts stay the same; what changes is how the sections are written and where they land in the store.
- Shopify: generation aims for a concise title and writes the long description as product body HTML with short paragraphs, sub-headings, lists, and a two-column specification table. Shopify has no separate short-description field, so the short description opens the product description. You choose which sections to include and can add section headings, translated for common listing languages. Products are created as drafts.
- WooCommerce: the short description goes into WooCommerce’s Short description field and the long description into the main Description. Specifications are written as a simple attribute and value table. Products are created as simple draft products.
- eBay: titles aim for search-led wording within eBay’s 80-character limit and specifications are written as explicit key-value pairs. New users export this eBay-targeted copy as generic CSV or JSON — not an eBay import schema. Existing connected accounts can reconnect and continue direct publishing.
- Any channel: export the reviewed listing as JSON, CSV, PDF, or DOCX to use in another system.
Review workflow and [REQUIRED: …] placeholders
The generator is built not to fill gaps. When the listing needs a piece of information the product data does not provide — a manufacturer address, a warning the category usually carries, a net quantity — it inserts a visible placeholder such as [REQUIRED: manufacturer postal address] instead of a plausible-sounding value. Rewrites and translations keep these placeholders in place.
A placeholder is a gap, not a pass. Resolve each one by adding the missing data from the supplier or by editing the section, then review the listing. Doc2Shelf records the reviewed content, and publishing or export is blocked until the current version has been reviewed.
- Check the extracted product fields first, starting with low-confidence extractions and the page-and-quote evidence for their values.
- Read each listing section against the reviewed fields and your source documents; generated sections do not carry a per-section source map.
- Resolve every [REQUIRED: …] placeholder with real data from the supplier or a deliberate edit.
- Confirm warnings and regulatory wording are complete and present where they apply to the product.
- Mark the listing as reviewed, then publish it as a draft or export it. Any later edit sends it back for review.
Descriptions in more than 100 languages
You can generate a listing directly in a target language or translate a finished listing into more than 100 languages. In both cases the title is localized too: brand and trademark names, model numbers, and product codes stay unchanged, while generic product-type and descriptive words are translated.
Translation works section by section and keeps formatting, numbers, units, and [REQUIRED: …] placeholders intact. Units are not converted automatically, so decide per market whether to add imperial or other units during review. Translated listings are review-ready drafts: a native speaker who knows the category should check wording and search terms before publishing.
What the generator will not do
These are the rules generation is built around. A language model can still make mistakes, which is why the review gate exists — but the design goal is copy that says less rather than copy that says something the documents do not support.
- Invent specifications, dimensions, capacities, or materials that are not in the product data.
- Add certifications, standards, test results, or markings the documents do not show.
- Fill a missing value with a “typical” or averaged one — it inserts a [REQUIRED: …] placeholder instead.
- Add disease-treatment or cure claims for supplements, or efficacy and “clinically proven” claims for cosmetics, that are not in the source data.
- Paraphrase hazard statements and safety warnings — regulatory wording is meant to be kept verbatim.
- Decide whether the product is compliant or which legal requirements apply. Doc2Shelf does not assess conformity, certify products, or replace qualified review.