Auto-Generate Product Descriptions for WooCommerce with n8n

If you sell 200+ products and each one still says "Lorem ipsum" where the description should be, you already know the problem. Writing fresh copy by hand does not scale. Hiring a copywriter for bulk work is expensive and slow. And pasting product specs into ChatGPT one at a time is just hand-work wearing a different hat.

This guide shows you how to build a single n8n workflow that fixes all three problems at once: it reads every WooCommerce product whose description is empty, hands the specs to an AI node that generates polished, SEO-aware copy, and writes the result straight back to WooCommerce via the REST API. Schedule it to run nightly and it keeps up with new stock automatically.

---

What the Workflow Does, End-to-End

1. Fetch products from WooCommerce that have no description.
2. Loop over each product.
3. Build a prompt from the product's name, SKU, categories, and attributes.
4. Call the AI/LLM node to generate the description.
5. Update the product in WooCommerce with the new text.
6. Log results (optional Slack or email notification when the batch finishes).

Total node count: around 7. Build time: 20–30 minutes.

---

Step-by-Step Build

1. Schedule Trigger

Add a Schedule Trigger node. A daily run at 03:00 server time is a safe default — your store is quiet and the workflow has time to process a large catalogue without hitting API rate limits.

Trigger At: 03:00
Timezone: UTC (or your store's timezone)

For a one-off bulk backfill, swap this for a Manual Trigger to run it immediately.

2. HTTP Request — Get Products Without Descriptions

Use an HTTP Request node to call the WooCommerce REST API.

Method:  GET
URL:     https://yourstore.com/wp-json/wc/v3/products
Auth:    Basic  (WooCommerce consumer key / secret)
Query parameters:
  per_page  100
  page      1       ← you will handle pagination in the loop
  status    publish

Tip: The WooCommerce API returns description as an HTML string. An empty description comes back as "". Filter inside n8n with an IF node immediately after this call: {{ $json.description === "" }} — only items that match proceed.

If you have more than 100 products you need to paginate. Add a Set node before the HTTP Request to hold the current page counter, and a Loop Over Items / IF combination that increments page and loops back until the response returns fewer than 100 items.

3. Split In Batches

Wire the filtered products into a Split In Batches node (batch size: 5–10). This prevents you from hammering the LLM endpoint or the WooCommerce API with hundreds of simultaneous requests.

4. Set — Build the Prompt

Add a Set node to construct the prompt you will pass to the AI node. Reference n8n's expression editor to pull fields from the current item:

prompt:
"You are a product copywriter for an online store.
Write a 2–3 sentence product description for the following item.
Make it SEO-friendly, avoid superlatives, and end with a benefit for the buyer.

Product name: {{ $json.name }}
SKU: {{ $json.sku }}
Categories: {{ $json.categories.map(c => c.name).join(', ') }}
Attributes: {{ $json.attributes.map(a => a.name + ': ' + a.options.join('/') ).join(' | ') }}

Return only the description text, no headings, no quotes."

Adjust the tone instruction to match your brand (formal, casual, technical, lifestyle).

5. AI / LLM Node — Generate the Description

Add the AI / Chat Model node (sometimes labelled "Basic LLM Chain" or "AI Agent" depending on your n8n version).

• Model: pick any model from the dropdown. For bulk product copy, a mid-tier model (e.g. google/gemini-flash-1.5 or meta-llama/llama-3.1-8b-instruct) gives a good balance of quality and speed. You can switch models without touching anything else in the workflow.
• Prompt: {{ $json.prompt }} (the value set in step 4).
• System message: you can move the role instruction ("You are a product copywriter…") here and keep the prompt itself shorter.
• Max tokens: 300 is enough for 2–3 sentences; set a ceiling so you do not burn credits on runaway output.

No API key setup. On AgentRoost the node is already wired to your included credits.

6. HTTP Request — Update the Product in WooCommerce

Add a second HTTP Request node:

Method:  PUT
URL:     https://yourstore.com/wp-json/wc/v3/products/{{ $json.id }}
Auth:    Basic  (same WooCommerce credentials)
Body (JSON):
{
  "description": "{{ $('AI / Chat Model').item.json.text }}"
}

Reference the product id from the original item (use Merge or sticky references if the AI node drops upstream context — pin the item via a Set node right after step 3 to carry id through the chain).

7. Notification (Optional)

After the batch loop finishes, a single Slack or Send Email node summarising how many products were updated costs one extra node and keeps the workflow auditable.

---

Prompt Tuning Tips

Goal	Adjustment
SEO-first copy	Add "Include the product name and primary category in the first sentence."
Formal/technical	Replace "copywriter" with "technical writer" in the system message
Shorter output	"Write exactly 1 sentence." or reduce max tokens to 100
Structured HTML	Ask for <p> tags; WooCommerce renders HTML in the description field
Avoid repetition across variants	Pass the parent product name and variant SKU separately

Run the workflow on 5–10 test products first (add an IF node that matches only a test category), read the output, then tweak the prompt before processing the full catalogue.

---

How to Run This on AgentRoost

Here is the actual path from zero to a running workflow:

1. Sign up at agentroost.app (email/password, or one click with Google, Microsoft, or Discord).
2. Create a workspace — choose the n8n framework, give it a name.
3. In about two minutes your private n8n editor opens at https://<your-id>.agentroost.app. It is your instance; you own the data and the workflows.
4. Build the workflow above. When you drop the AI/LLM node, the credentials dropdown already shows a working connection — AI credits are included in your plan, so there is nothing to configure.
5. Add WooCommerce credentials (Basic Auth with your consumer key/secret) in n8n's Credentials panel. That is the only external secret you need.
6. Activate the workflow. The Schedule Trigger handles everything from there.

Pricing starts at $19.99/mo all-in — compute, LLM credits, SSL, public subdomain, everything. No separate OpenAI bill. No "bring your own API key" friction. 14-day money-back guarantee if it does not fit.

Compare plans · See the n8n workspace

---

Pitfalls to Avoid

Rate limits on the WooCommerce side. The REST API is synchronous PHP. Batch size 5 with a 1-second delay between batches (use a Wait node set to 1s inside the loop) keeps you well under typical shared-host limits.

HTML vs. plain text. description accepts HTML. If you ask the AI for plain text and WooCommerce renders it fine — but if you later want <ul> feature lists, change the prompt; you do not need to rebuild the workflow.

Overwriting manually-written descriptions. Add a second filter after step 2: {{ $json.description !== "" }} routes to a "skip" branch. This way the workflow only touches blank descriptions.

Context loss between the AI node and the update node. n8n's AI node does not automatically forward the upstream item's fields. Use a Merge node (mode: Combine, by position) to zip the AI output back together with the original product data before the PUT request.