Search Term Classifier
Classify Google Ads search terms by intent using a hybrid rule-based + ML approach. Designed to handle 100k+ terms efficiently with minimal LLM cost.
Process 100k+ terms
Classify massive search term reports in seconds, not hours
Minimal LLM cost
500 LLM calls can classify 60,000+ terms via pattern propagation
Actionable insights
Know exactly where your budget is going by intent category
Find wasted spend
Quickly identify negative keywords draining your budget
Try the Classifier
Watch the 10-step pipeline classify search terms in real-time.
Classification Pipeline
Click "Run Classification" to see the pipeline in action
Watch as 18 search terms are classified through 10 pipeline steps
How It Works
The 10-Step Pipeline
The classifier prioritizes fast, free methods before expensive LLM calls. Most terms are classified without ever touching an AI model.
Flags non-Latin terms in Latin-dominant accounts
Marks bottom 5% of terms by impressions
Identifies your brand terms from configured strings
Detects competitor and navigational searches
Reuses previous classifications (60-80% hit rate)
Classifies by intent keywords (buy, price, how to, etc.)
Catches brand typos using Levenshtein distance
Sends top N terms to AI for classification
Learns patterns and applies to remaining terms
Assigns medium_intent to unmatched terms
Intent Categories
Search terms are classified into 8 actionable categories based on user intent and conversion likelihood.
Your brand terms
e.g., your brand name
Competitor brand searches
e.g., competitor name
Purchase signals
e.g., buy, price, near me, discount
Product browsing
e.g., product category searches
Research queries
e.g., how to, what is, guide
Won't convert
e.g., jobs, reddit, DIY, free
Bottom 5% impressions
e.g., insufficient data
Different script
e.g., non-Latin characters
The ML Propagation Magic
The key insight: you don't need to LLM-classify every term. By classifying 500-1,000 high-impression terms with an LLM, the system learns patterns it can apply to the remaining 60,000+ terms:
- N-gram patterns: "office chair" → medium_intent, "buy desk" → high_intent
- Word patterns: Terms containing "cheap" are 85% high_intent
- Cosine similarity: "ergonomic gaming chair" is similar to "gaming chair" (already classified)
Typical results: 500 LLM calls → 97% coverage of remaining terms via propagation.
LLM Cost Estimates
Using Gemini 2.5 Flash for LLM classification keeps costs extremely low:
| LLM Terms | Model | Est. Cost | Est. Time |
|---|---|---|---|
| 500 | Gemini 2.5 Flash | ~$0.01 | ~15s |
| 1,000 | Gemini 2.5 Flash | ~$0.02 | ~25s |
| 2,000 | Gemini 2.5 Flash | ~$0.04 | ~45s |
Want to classify your own search terms?
Join the Ads to AI community to get access to the full Search Term Classifier, along with other Google Ads automation tools, scripts, and AI-powered analysis.
