Descriptions nobody has to write by hand — and that you still read before they're saved.
One click in the product editor, or a bulk run across the entire product list. With your own prompt templates, placeholders pulled from real product data, and a background queue that doesn't fall over at 20,000 items.
Short description: Steel burr grinder with 40 stepped settings — espresso fine through to French press coarse, with an even grind throughout.
Long description: A grinder for people who weigh their beans. The flat steel burrs hold the grind size steady across the whole dose, so extraction becomes repeatable …
Empty description fields don't just cost rankings — they cost sales. Flip the switch and watch what changes across the same three products.
Figures from a real shop with 2,400 items after the first bulk run.
A button in the product sidebar. Click, eight seconds, preview — then you decide whether the text goes into the field. Nothing is saved without you.
Select products, choose the bulk action, let it run. Filters for "no description", by category or by product type — so you don't rewrite what's already good.
Your tone, your structure, your rules — saved as a reusable template. With placeholders like {{product_name}}, {{categories}} and {{attributes}} filled from the real product data.
Processing runs through WooCommerce's Action Scheduler. Close the window, carry on working, pause the run and resume it later — no PHP timeout.
Optional: newly created products get their description after an adjustable delay, so categories and attributes are finished first. Only empty fields are filled.
Short and long description controlled separately — plus two text fields per category, tag and brand: an intro above the grid and SEO copy below it.
Who, when, which model, how many tokens, what it cost. Every text is traceable in the log and can be restored from there with one click.
Runs with the new WooCommerce order tables. API key encrypted with AES-256, capability checks on every action, a warning before any existing text is overwritten.
Existing copy becomes the fact source: numbers, measurements and details stay, the AI restructures to SEO format, adds from product data and cuts clichés — without carrying over a single sentence verbatim.
The system prompt explicitly forbids invented data, certificates and compatibilities. When data is thin, the AI writes shorter and undershoots the minimum rather than hallucinate.
Enter your own OpenRouter key, pick a model, done. Registering with OpenRouter is free.
Set tone, length and structure once. Placeholders pull the product data in automatically.
One at a time in the editor, or filtered in the list — say, everything without a description in one category.
The queue works in the background. You watch the progress, read the results and approve.
Through OpenRouter a single account reaches the leading providers. Cheap models for the bulk run, strong models for your best sellers — set per template.
Two thousand items from the supplier, two thousand empty description fields. One bulk run overnight, and in the morning you're reading rather than typing.
The same manufacturer text sits in twenty other shops. With unique copy per item the problem disappears, and nobody has to paraphrase anything.
One template per client, one licence per shop. The catalogue is described before the site goes live — without a copy budget in the quote.
For people who maintain products but don't write: set the template up once, and from then on the text is a click rather than a meeting with a copywriter.
New items every season, the same copy backlog every time. With automatic generation on product creation it keeps pace instead of piling up.
Not just products: an intro above the grid and a longer SEO text below it, per category, tag and brand — the pages that usually rank with no copy at all.
Descriptions sends your prompts straight to OpenRouter with your own key. We take no margin on tokens, and nothing passes through our servers.
Register with OpenRouter, paste the key, done. You pay the model providers directly by usage and see every cent in their dashboard — and in ours, per run and per model.
Prefer not to deal with a second provider account? We set the key up, configure the models per template and hand the whole thing over ready to run — with starting credit included.
Either way the plugin is the same. The managed key is convenience, not a feature gate — every function works with your own key from day one.
Every feature in every plan. The only difference is how many products and sites you cover. Model costs go straight to OpenRouter — usually fractions of a cent per text.
Installation, template design in your tone of voice, the first bulk run supervised, and a walkthrough for your team. Booked by the hour, billed transparently.
Two hours together on the prompts: tone, structure, which attributes carry the text, where the rewrite mode fits. You leave with templates that need no further tuning.
We run the whole catalogue: templates, generation, review of a sample, handover. Ideal right after a feed import, when a few thousand fields are empty at once.
You add your own OpenRouter key and pay the model costs directly there, by usage in USD. A product text costs fractions of a cent to a few cents depending on the model — 2,000 descriptions typically come in under ten euros. Before every bulk run the plugin shows a cost estimate using the real model prices; afterwards the actual costs are in the log. A monthly token limit hard-stops generation before it gets expensive. If you'd rather not run a second provider account, the managed key is €199 once, with starting credit included.
Only if you want it to. By default, filled description fields are skipped, and every overwrite triggers a warning first. Content from Elementor, WPBakery, Divi, Beaver Builder, Oxygen, Avada or Brizy is detected and protected. Every generation lands in the log and can be restored from there with one click.
The template sets structure and tone; the content comes from the product data: title, attributes, categories, tags, weight, price. That produces a distinct text per item rather than a stencil with the name swapped out. The system prompt also bans AI clichés like "Discover the" outright. For very similar variants we'd suggest a dedicated template that leans harder on attributes.
Instead of writing from scratch, the existing copy becomes the fact source. Numbers, measurements and specifics stay exactly as they are; the AI restructures to an SEO format, fills gaps from the product data and cuts filler — without carrying over a single sentence verbatim. Useful when supplier text is factually correct but reads like a spec sheet.
Yes. For product categories, tags and brands the plugin generates two texts: an intro above the product grid of 60 to 120 words and a detailed SEO text below of 200 to 350 words with subheadings. The lower field is preset for WoodMart and configurable for any theme. The data basis is the term name, parent category, product count and the term's newest products.
Around eight seconds per product, depending on model and text length. 2,000 products take roughly four hours — in the background, with no need to keep a window open. The queue runs through the Action Scheduler, can be paused and resumed, and survives a server restart.
Yes. The plugin writes into WooCommerce's standard short and long description fields. If a product uses a page builder or shortcodes in the description field instead, the plugin detects that and leaves the content untouched until you explicitly agree.
WordPress 6.x to 7.1 and WooCommerce 8.x to 11.1, HPOS-ready for the new order tables, multisite capable. PHP 8.4 and above only — a deliberate decision for strict types, modern language features and performance, without compatibility ballast for old servers.
The plugin retries up to three times and flags the item in the log if it ultimately fails. The queue carries on — a single failure doesn't abort the whole run. Failed products can be re-queued together afterwards.
A WooCommerce product page without copy is a thin page. Google sees a title, a price and an image, and ranks a competitor who wrote three paragraphs above it. The gap rarely comes from unwillingness — it comes from arithmetic. Eighteen minutes per product is a realistic figure for a properly researched description, and at 2,000 items that is 600 hours of work nobody has.
So most shops do one of two things. They leave the field empty, or they paste the manufacturer's text — which sits word for word in twenty other shops and turns the page into duplicate content. Both outcomes cost rankings, and both cost conversions, because a visitor who can't find the answer to a simple question leaves rather than asks.
FLAVOR Descriptions closes that gap by writing from the data you already maintain. Title, attributes, categories, tags, weight and price go into the prompt; a short description for listings and a long description for the detail page come out. Eight seconds per product instead of eighteen minutes, with the whole catalogue processed in a single background run.
The point is not volume for its own sake. A text built from real attributes says something specific — burr type, settings, material, warranty — and that specificity is what makes it both readable and rankable. Generic filler would perform no better than the empty field it replaced.
Three templates ship with the plugin — Standard, Technical, Emotional — and you can build as many more as you like. A template fixes tone, language, formal or informal address, word limits and structure, then pulls the specifics through placeholders like {{product_name}}, {{attributes}} and {{categories}}.
Assign a template per category and an espresso grinder reads differently from a gift voucher, without anyone rewriting a prompt. The system prompt underneath is the same in every case: no invented data, no fabricated certificates or compatibilities, and no AI clichés. When the product data is thin, the model writes shorter rather than making something up.
Generating 2,000 descriptions in a browser tab is how PHP timeouts happen. Descriptions queues the work through WooCommerce's own Action Scheduler, in batches you size yourself, with a pause between them. Close the window, restart the server, come back tomorrow — the run picks up where it stopped.
Filters decide what gets touched: only products without a short description, only one category, only a product type. Before the run starts you see a cost estimate based on the real model prices; afterwards the actual token count and cost sit in the log, per product and per model. A monthly token cap stops everything cold if a bad prompt would otherwise run up a bill.
Every generation produces a preview first. Filled fields are skipped by default, page-builder content is detected and protected, and each text lands in a log with model, tokens, cost and the previous version — restorable with one click. Automatic generation on product creation waits out an adjustable delay, so categories and attributes are finished before the copy is written.
That is the difference between a tool that helps and a tool you have to clean up after. The AI drafts; you approve. On the technical side the plugin stays out of the way: API key encrypted with AES-256, capability checks on every action, HPOS-ready, PHP 8.4+ with strict types, and no product data routed through our servers — your key talks to OpenRouter directly.
One bulk run overnight, and tomorrow your catalogue reads as though somebody wrote every page by hand.
A WooCommerce AI product description pays off once it covers every item in the shop. FLAVOR Descriptions writes short and long copy from data your store already holds: title, categories, tags, attributes, price, SKU and weight.
Out of those fields comes text that fits the single item, not a template with the product name dropped into it. Each WooCommerce AI product description carries details that exist nowhere else in the catalogue. That detail is what indexing rewards later.
The difference to an online generator sits in the access: the plugin works where your product data lives, not in a browser tab you paste every item into. A WooCommerce AI product description written inside the shop knows variants, stock status and the category tree. Between the first click and finished copy, seconds pass instead of minutes.
One item is quick work. What do you do with 2,000? That is the case we built the bulk mode for, so it runs on the Action Scheduler of WooCommerce, the same queue that already handles orders and emails. Start it from the product list, then close the browser.
While an item moves through the queue, the WooCommerce AI product description takes around 8 seconds. If the run stops or the server restarts, the queue picks up where it left off. Filter beforehand by missing description, by category or by product type, so curated items stay untouched.
What that means in time and money depends on catalogue size and model:
| Catalogue size | Runtime in the background | Typical model cost |
|---|---|---|
| 250 items | around 35 minutes | under 2 euros |
| 1,000 items | around 2 hours | under 5 euros |
| 2,000 items | around 4 hours | under 10 euros |
OpenRouter bills the model cost by usage, 0.002 to 0.02 euros per text. Where an agency would write for weeks, a WooCommerce bulk product description run is done in an afternoon, and single runs of 20,000 items work the same way. What the WooCommerce AI product description delivers still passes your review before it goes live.
A tool catalogue reads differently from a jewellery collection. To keep that difference, you create as many prompt templates as you need, usually one per category, each with its own tone from technical and short to warm and narrative.
Put placeholders such as {title}, {categories}, {attributes} or {price} into the template. During generation the WooCommerce product description generator pulls the real values of the item, which is why measurements, materials and features are named correctly instead of invented.
Before anything is saved, the preview shows what the WooCommerce AI product description has written. Nothing reaches the shop unchecked. Existing copy is only overwritten when you confirm it, and an earlier version comes back out of the log.
Requirements are WordPress 6.x to 7.1, WooCommerce 8.x to 11.1 and PHP 8.4. On that base the WooCommerce AI plugin stays HPOS-ready and multisite capable. Because many shops run page builders, it detects Elementor, WPBakery and shortcodes, so the layout survives the rewrite.
Every WooCommerce AI product description depends on an external API account, so your key sits encrypted in the database and each request checks nonce and user rights. The log records tokens and cost per text, exports to CSV, and a monthly token limit keeps spending predictable. Multilingual shops run through WPML and Polylang with one template per language, so copy is written in the target language rather than translated afterwards. Where our WordPress translation plugin handles the language layer, both tools work side by side.
For as long as WooCommerce has existed, we have been building shops with it. We started in 2010 and crossed the 1,000-project mark along the way, which is why we know the moment when 800 items are imported and the description fields stay empty. That moment produced our WooCommerce AI product description, feature by feature out of client work.
What clients make of it is public on ProvenExpert. Whoever answers in support has worked on the queue code themselves. Every other product in our premium WordPress plugins range is built to the same standard, and the finished copy gets its markup from our WordPress SEO plugin.
Ten models are available through OpenRouter, among them GPT-4o and GPT-4o-mini, Claude 3.5 Sonnet, Claude 3 Opus, Gemini 2.0 Flash, Llama 3.1 and Mistral. Which one your WooCommerce AI content generator uses is a per-template setting, together with temperature and max tokens.
Three tiers cover the WooCommerce AI product description, sorted by catalogue size. Starter costs 59 euros a month for one site and up to 250 products, Professional 129 euros for 3 sites and up to 2,500 products, Agency 299 euros for 10 sites without a product limit. Paying yearly saves 2 months.
Catalogues from roughly 100 items upwards, where manual writing costs more than a licence. Running 20 products, you do not need an AI product description plugin, just an afternoon.