Documentation
From installation to the first processed category, in a quarter of an hour.
Requirements
- WordPress 6.2 or later, WooCommerce 7.0 or later, PHP 7.4 or later.
- An active Fichewise license.
- With the managed service: no API key, only credits.
- With your own keys: a key for writing, Anthropic (recommended) or OpenAI, and a key for web search, Tavily (recommended), Serper.dev, SerpAPI or Brave Search.
- Optional: the
pdftotexttool on the server, only useful if you use OpenAI with PDFs (with Anthropic, PDFs are read directly).
Installation
- After purchase, download the archive from the confirmation email or from your Freemius account. In the WordPress admin, open Plugins > Add New > Upload Plugin, choose the archive, install, then activate.
- Enter your license key on activation, then choose the API access mode in the settings: your own keys or the managed service.
- A Fichewise menu appears in the sidebar. The log table is created on activation.
- The plugin declares compatibility with WooCommerce High-Performance Order Storage (HPOS).
API keys
This step only applies to the own-keys mode. In Fichewise > Settings, choose the AI provider and the search engine, paste the keys, then click Test next to each one.
| Service | Where to create the key | Note |
|---|---|---|
| Anthropic | console.anthropic.com, API Keys | Default model: Claude Sonnet, a good balance of quality and cost. |
| OpenAI | platform.openai.com, API keys | PDFs go through pdftotext. |
| Tavily | tavily.com | 1,000 free searches per month, page content included. |
| Serper.dev | serper.dev | Real Google SERP with "People also ask" and Shopping. Recommended for competitor analysis. |
To keep the keys out of the database, define them in wp-config.php:
define( 'WEA_ANTHROPIC_API_KEY', 'sk-ant-...' );
define( 'WEA_TAVILY_API_KEY', 'tvly-...' );Store profile
Open Fichewise > Store profile and click Analyze the store. The analysis reads a dozen of your most polished descriptions and produces an editorial profile: positioning, audience, tone of voice, style rules, vocabulary, expressions to avoid. It takes less than a minute and only runs again when you ask for it.
To impose your own tone, fill in "Imposed tone of voice" and "Additional instructions" in the settings. They take precedence over the analysis.
First product page
- Open a product that has an EAN or a known brand.
- In the Fichewise side panel, click Enrich now.
- Allow one to three minutes: research, source reading, competitor analysis, writing, proofreading. You are then redirected to the review screen.
On shared hosting, a run longer than two minutes may be interrupted by the server. In that case use Queue: processing moves to the background, step by step.
Review screen
The review screen shows each proposed field next to the current content, with an HTML preview and character counters for the title tag and the meta description. On the right, the sources that were read with their score, the conflicts to settle and the summary of the competitor analysis.
- Choose, on a conflict, keeps one value and marks it as validated by you.
- Regenerate the copy runs the writing again with the same data, for example after settling a conflict.
- Start over redoes the research, the competitor analysis and the writing.
- Publish writes the page to the product. A backup is made first, and Restore original puts it back.
Supplier documents
Below the editor of each product, the Supplier documents panel accepts:
- files from the media library: PDFs, images, text files;
- URLs of online documents, one per line;
- seller notes: origin, particularities, what must not be said.
On the edit page of a category, you can attach a supplier catalog: it serves all products in the category and its subcategories, and the AI locates the line for the right product. The main image and two gallery images are analyzed automatically.
Bulk processing
From the Dashboard, choose a scope (products without a description, a category, products in error, the whole catalog), a limit, then Queue. Processing runs in the background with Action Scheduler, which already ships with WooCommerce. The Products list updates on its own and finished pages land in Review.
The "Fichewise: enrich" bulk action is also available in the native WooCommerce product list.
Products that cannot be found
When no source and no document can be used, Fichewise switches to a fallback mode instead of failing: the page is written from the store data alone, the competitor analysis covers the category, no specification is added, the score is capped at 55 and a banner states what should be provided. Add a document or a label photo, then run again.
SEO and structured data
The title tag, meta description and focus keyword are written into the fields of Rank Math, Yoast SEO, All in One SEO or SEOPress, whichever is active. Without an SEO plugin, Fichewise outputs these tags itself.
On published products, the FAQ generates a FAQPage block and the WooCommerce Product schema is completed: brand, GTIN, manufacturer part number and additional properties.
Shopping, Amazon, languages
In the settings, enable the Google Shopping and Amazon channels: each product then receives a title, a description and attributes that follow the channel's standards. Additional languages produce translated versions of the product page. Everything exports as CSV or JSON from Fichewise > Exports.
Catalog audit
Open Fichewise > Dashboard, Audit tab. The store is read once, with no AI call and no credit, and the result is cached for ten minutes. Three lists:
- Missing fields, per category: description, short description, SEO meta, GTIN, and the attributes declared by the category schema.
- Values to harmonize: spellings of one attribute that name the same value, such as "75 cl" and "750 ml", grouped together. The most frequent spelling is proposed as the reference; you can pick another or type one, tick the groups to correct, then apply. Only ticked groups are rewritten, and the chosen value then counts as your store data.
- Likely duplicates: products that share a GTIN or a SKU, or whose titles are near-identical.
Each list exports to CSV from its header. The scan stops at 3,000 products; beyond 1,500, title similarity is only compared within a category.
Translating a product page
Requirements: WPML or Polylang active, at least two languages declared, and the Products post type enabled as translatable in the plugin's settings. Otherwise the button does not appear and the screen says why.
- Click Translate in the Fichewise Products list, on the review screen or in the product panel, and choose the target languages.
- One draft per language appears in the review. Numbers, units, GTIN, SKU, brands and Store Profile terms are copied as they are; a figure from the source that is missing from the translation marks the draft "to check", with the list of what is missing. Review, and correct if needed.
- On approval, the translated product is created or updated through WPML or Polylang and linked to the source product. Nothing is written before that.
The translation takes the publication status of the source product at the moment it is written, draft, pending or published, and does not change afterwards: you publish each language in WPML or Polylang. With the managed service, a translation costs one credit per language, counted on approval; an abandoned draft returns its credit, and a draft neither approved nor abandoned within 24 hours is counted as consumed.
WP-CLI and REST API
wp wea profile # analyze the store
wp wea enrich --ids=12,15 --sync # enrich immediately
wp wea enrich --category=red-wines # queue a category
wp wea enrich --missing --limit=50 # products without a description
wp wea publish # publish everything awaiting approval
wp wea rollback --ids=12 # restore
wp wea export google_shopping --file=gs.csv
wp action-scheduler run --group=fichewiseThe REST API is exposed under /wp-json/wea/v1/ for users with the manage_woocommerce capability: POST /enrich, GET /status/{id}, GET /dossier/{id}, POST /publish/{id}, POST /rollback/{id}, GET /stats, GET /export/{channel}.
Useful settings
- Trusted domains: manufacturer or wholesaler sites that receive maximum reliability.
- Excluded domains: social networks by default; add your direct competitors if you do not want them used as sources.
- Minimum attribute confidence: below this threshold, the attribute is not asserted in the copy (0.6 by default).
- Replaced fields: title, descriptions, attributes, SEO meta, tags, alt text, GTIN, each one enabled separately.
A question not covered here? Write to us.