How Keyword Matching Works in Zitcha
This article explains how Zitcha matches shopper searches to sponsored products. It covers dynamic search terms, positive keywords, and negative keywords, how each one works, their priority order, and how they interact with the AI relevancy model and auction ranking.
Overview
Zitcha’s onsite search ad serving balances automation with advertiser control.
At its core, the platform uses an AI-driven relevancy model to determine which products should appear for a shopper’s search. Keywords do not replace this model. Instead, they allow advertisers to fine-tune when products can appear, and when they must not.
There are three mechanisms involved in matching a shopper’s search to a product ad.
| Mechanism | Source | Match Type | Priority |
|---|---|---|---|
| Negative Keywords | Advertiser | Exact phrase match | Highest (blocks ads) |
| Positive Keywords | Advertiser | Exact phrase match with limited typo tolerance | High (adds eligibility) |
| Dynamic Search Terms | System generated | Exact phrase match with limited typo tolerance | Standard |
Dynamic Search Terms
Dynamic search terms are automatically generated phrases that describe a product and how shoppers are likely to search for it.
They are derived from:
- Product titles and descriptions
- Brand, category, size, colour, and style
- Retail and specialist language
- Natural language patterns learned by the system
How matching works
Dynamic search terms require:
- Exact word order
- Full token matches (no stemming)
- Case-insensitive matching
Example
Product: Nike Air Max
Dynamic search terms:
- running shoes
- athletic footwear
- nike trainers
| Shopper Search | Result |
|---|---|
| running shoes | Matches |
| running sneakers | No match |
| shoes running | No match |
Positive Keywords
Positive keywords are advertiser-supplied phrases that expand when a product can appear.
They do not override the AI relevancy model. They add additional search phrases on top of what the system already understands.
How they work
- Phrase match only
- Same word order required
- Limited typo tolerance
- Additive, not restrictive
Example
Product: Premium Leather Wallet
Dynamic search terms:
- leather wallet
- mens accessories
Positive keywords:
- gift for dad
- fathers day present
| Shopper Search | Result |
|---|---|
| leather wallet | Matches (dynamic) |
| gift for dad | Matches (positive) |
| fathers day accessories | Matches (positive + dynamic) |
Negative Keywords
Negative keywords are advertiser-supplied phrases that block a product from appearing.
They have the highest priority in the system.
How they work
- Phrase match only
- Exact word order required
- Singular and plural treated separately
- Always override positive keywords and relevancy
Example
Product: Dyson V15 Vacuum
Negative keywords:
- cheap
- refurbished
| Shopper Search | Result |
|---|---|
| vacuum cleaner | Matches |
| cheap vacuum cleaner | Blocked |
| refurbished dyson | Blocked |
Phrase and N-gram Matching for Negative Keywords
Shopper searches are broken into all possible word groupings (n-grams).
If any grouping matches a negative keyword, the product is excluded.
Example search
dyson vacuum cleanersGenerated phrases:
- dyson vacuum cleaners
- dyson vacuum
- vacuum cleaners
- dyson
- vacuum
- cleaners
Precedence Rules
Evaluation order:
- Positive keywords, dynamic search terms, or system terms
- Negative keyword exclusions
- Return eligible products
Negative keywords always win.
Matching Behaviour Details
Fuzzy matching rules
| Word Length | Typos Allowed |
|---|---|
| ≤ 6 characters | 0 |
| 7–9 characters | 1 |
| 10+ characters | 2 |
Additional rules:
- No prefix matching
- No stemming
Sorting and Ranking
Eligible products enter a CPM-based auction.
- Ranking uses eCPM (bid × quality score)
- Equal bids are randomly rotated
- User-defined sorting takes precedence
Best Practices
- Rely on AI relevancy for broad coverage
- Use positive keywords sparingly
- Include singular and plural forms for negatives
- Use negatives for brand safety
Updated about 1 month ago