Product recommendation settings
Ecommerce → AI configuration decides how the bot turns a question into a shortlist of products. Get this wrong and a good catalog stays invisible.
Search strategy
| Option | How it works | Use when |
|---|---|---|
| Product Search Only | Searches product records directly | Your titles and descriptions are rich and specific |
| Product Search with Category Fallback | Tries products first, then falls back to categories | Most shops — it degrades gracefully |
| Category Search Only | Matches the question to a category, then shows its products | Large catalogs of similar items, or thin product descriptions |
Start on Category Fallback. Move to Product Search Only if your descriptions are strong and category results feel too broad. Only use Category Only if your categories are clean and complete — with empty categories it returns nothing at all.
Product Title Search Finetuning
Describe how your product titles are structured, so the AI knows which part is signal:
Brand - Product name - VariationIf your titles read GreenCo — Monstera Deliciosa — 60 cm, saying so stops the AI treating the size as part of the product name. Shops with a consistent title format usually see the biggest gain from this one field.
Category Search Finetuning
Extra instructions appended to the category search prompt. Use it for the vocabulary gaps between how you name categories and how customers ask:
Customers say “office plants” for the Houseplants category. Treat “gifts” as a request across all categories rather than as a category.
Enhanced Product Detail Prompt
Extra instructions for how recommendations are written up. This shapes presentation, not selection:
Always mention light and watering needs. Say if a product is on sale. Do not claim a delivery date.
Keep it short. Long instructions here make replies wordy and slow.
Brand awareness
Turn on Enable brand awareness when you sell branded goods and customers ask by brand. It tunes product search to the brands in your catalog and adds a Brands page where you choose which brands stay available.
How to tune without guessing
- Collect ten real questions from Chat conversations that returned poor or no products
- Change one setting
- Ask the same ten questions on the Demo page
- Keep the change only if more of them improved
- Check Chats with Products % and Conversion Rate on the Dashboard a week later
When settings are not the problem
If most questions return nothing, the fix is usually upstream:
- No products imported for the visitor’s language
- Descriptions that only list specifications
- Missing
product_typevalues, leaving categories empty
Related articles
Connect a product feed
Import your catalog from a Google Shopping feed, a CSV URL, or a CSV upload.
Categories and brands
Curate which parts of your catalog the bot may suggest, and set follow-up questions.
Product feed field reference
Exact column and element names Mikabot reads from CSV and Google Shopping feeds.
Contact us
Our support team is ready to give you fast, personal help when you need it.