Every patent analysis, whether a freedom-to-operate check, a landscape or a competitor watch, depends on one thing: a good search. If relevant documents are missing, the analysis is wrong. If the results are full of noise, it is unreadable. This guide shows how professionals build and test patent search strategies.
The three building blocks
1. Keywords. Words that describe the invention: its function, components, materials and applications. Keywords are intuitive, but patent language is deliberately broad and unusual. A “bicycle” may be described as a “human-powered two-wheeled vehicle”. Different translations of the same foreign-language patent may use different terms.
2. Classification codes. Patent offices assign every document codes that describe its technical content:
- The International Patent Classification (IPC), maintained by WIPO, has about 70,000 groups and is used by over 100 patent offices.
- The Cooperative Patent Classification (CPC), run jointly by the European and US patent offices, extends the IPC to about 250,000 groups. It is used by many offices, including China’s, for at least part of their documents.
A code such as H01M 10/0562 (solid electrolytes for secondary batteries) captures documents whatever words they use, in any language.
3. Applicants and inventors. Useful for competitor monitoring, but names need cleaning. One company can appear under dozens of spellings and subsidiaries.
A step-by-step method
Step 1: Write down the concept in parts
Split the invention into its essential concepts. For “solid-state lithium batteries for electric vehicles”:
- Concept A: solid electrolyte
- Concept B: lithium battery / secondary cell
- Concept C (optional, narrows scope): vehicle application
Step 2: Collect synonyms for each concept
| Concept | Keywords (with truncation) |
|---|---|
| A | solid electrolyt*, solid-state electrolyt*, ceramic electrolyt*, sulfide electrolyt*, garnet, LLZO |
| B | lithium, Li-ion, secondary batter*, rechargeable cell*, accumulator* |
| C | vehicle*, automotive, EV, car* |
Use the truncation (*) and proximity operators your database supports. Proximity operators such as “within 5 words” are often more precise than simple AND.
Step 3: Find the classification codes
- Search keywords in the IPC/CPC scheme browsers.
- Even better, take 5 to 10 documents you know are relevant and look at which codes they have. The codes that recur are your candidates.
- Read the code definitions to check what is included and excluded.
Step 4: Combine
A robust strategy usually combines codes and keywords:
(CPC = H01M10/0562 OR (A-keywords NEAR B-keywords))
AND publication date ≥ 2010Using codes or keyword combinations catches documents that one approach alone would miss. Add concept C only if you truly need to restrict the scope.
Step 5: Test recall and precision
- Recall: make a “gold set” of known relevant patents, ideally 20 or more, from experts, citations and earlier studies. What share does your search find? If it misses some, examine why and adjust.
- Precision: read a random sample of 50 results. What share is relevant? If precision is low, add exclusions or tighten proximity.
Iterate until both are acceptable for your purpose. Freedom-to-operate searches need very high recall. Trend analyses can accept somewhat lower recall if the noise is low.
Step 6: Group by family and clean
Count patent families, not individual documents. Clean applicant names before ranking anyone.
Step 7: Document everything
Record the databases, dates, full query strings, and the recall and precision checks. Patent searches are often repeated to update an analysis, and an undocumented search cannot be repeated.
Choosing a database
| Database | Cost | Strengths |
|---|---|---|
| Espacenet (EPO) | Free | Very broad worldwide coverage, CPC, family grouping, legal status links |
| PATENTSCOPE (WIPO) | Free | PCT applications, cross-lingual search, many national collections |
| Google Patents | Free | Fast full-text search, machine translations, citations, links to scholarly literature |
| The Lens | Free | Links patents to scholarly works, good for analysis and export |
| National office databases (CNIPA, USPTO, JPO, KIPO…) | Free | Authoritative national data and legal status |
| Commercial platforms | Paid | Cleaned names, analytics, alerts, better legal status data |
Language matters
Over half of the world’s patent applications are now filed in Asian languages. A search built only on English keywords will miss documents whose English abstracts use different terms, or that have no English abstract at all. Classification codes, machine translation and cross-lingual search features help close the gap. For important searches, add native-language keywords.
Semantic and AI search
Many databases now offer semantic search: you paste a description of the invention and the system finds documents with similar meaning. It is excellent for discovering vocabulary and codes you had not thought of, and as a check on recall. Its results are hard to reproduce and explain, however. Professionals use it alongside a documented Boolean strategy, not instead of one.