Vyzkoušet zdarma
ÚvodPlatformaCeníkOtázkyNápovědaKontakt Kivea FlexiKivea PohodaKivea účetnictvíKivea fakturace Vyzkoušet zdarma

Automatic Product Categorization

Manually assigning hundreds or thousands of products to categories is time-consuming and error-prone. The AI categorization workflow lets you automatically assign each product to the most appropriate category using a combination of semantic search and a language model.

[try:/products|Open Products to Start|shopping-cart]

How It Works

card: Semantic pre-filtering
icon: search
For each product, the system retrieves the top 10 categories that are most semantically similar to the product's input fields (e.g. name and description). This narrows down the candidates efficiently.

card: LLM decision
icon: bot
The language model reviews the candidate categories alongside your prompt context and selects the single best-fitting category for each product.

Step-by-Step Guide

1. Open the product grid and select the products you want to categorize. You can select individual rows or use **Select All** to process your entire catalog.
2. Click **Mass Update** → **AI Edit**.
3. In the AI Edit dialog, choose **Generate New Relations** as the operation type.
4. Under **Input Fields**, select the fields that describe each product — typically **Name** and **Description**. You can include additional fields (e.g. brand, attributes) for better accuracy.
5. Under **Target Relation**, select **Product Categories**.
6. Set the **Default Category** — this will be used as a fallback if no suitable category is found.
7. Enter your **prompt**. Describe your catalog context so the AI can make informed decisions (see tips below).
8. Click **Run** to start the categorization.

Writing an Effective Prompt

The prompt gives the AI important context about your business. Without it, the model has to guess the domain from the product data alone.

**What to include in your prompt:**
- The type of e-shop or catalog (e.g. "This is a B2B electronics distributor with categories for networking, computing, and peripherals.")
- Any conventions or rules the AI should follow (e.g. "Always prefer the most specific subcategory over a general parent category.")
- What to do if no category fits (e.g. "If nothing matches, assign to 'Other'.")

Example prompt:

This is an online pet supply store. Our product categories include food, toys, grooming, health & care, and accessories — each with subcategories by animal type (dog, cat, bird, fish). Assign each product to the most specific matching category. If in doubt between two similar categories, prefer the one for the specific animal mentioned in the product name.

Tips for Best Results

**Choose input fields carefully.** The semantic search uses the fields you selected to find candidate categories. Including a detailed description alongside the name gives significantly better results than using the name alone.

**Run on a small batch first.** Before processing your entire catalog, select 20–50 products to verify the AI's choices match your expectations. Adjust the prompt if needed, then run on the full set.

**Review the results.** Categorization is a suggestion — always spot-check the output and correct any misassignments before publishing. You can filter by category to quickly scan the results.
  • [navigate:/help/en/user-flows/ai-batch-operations|AI-Powered Batch Operations|layers]
  • [navigate:/help/en/user-flows/semantic-search|Semantic Search|search]