How to Choose Keywords for a Patent Search

Why Most Patent Searches Fail on the First Query

Patent searches usually fail in the same way: someone types the name of their product, gets almost nothing back, and concludes the invention is original. It is a false negative, and it is expensive. Patent attorneys draft deliberately broad, functional language, so a product you call a bike bottle holder may be indexed as a "device for securing a cylindrical vessel to a tubular frame member." Choosing patent search keywords is the skill that separates a search producing false confidence from one that actually shows you what exists.

The method below is what we use to build the term list behind a prior art search before starting engineering work on a new product. It takes an afternoon and it is the highest-return afternoon in the entire project.

Start From Function, Not From the Product

Write one sentence describing your invention with no brand names, no marketing words, and no shape description: what it does, what it acts on, and by what means. Then split that sentence into three columns:

• Action — the verb. Secure, dispense, filter, sense, damp, fold, align, heat.
• Object — what is acted on. Bottle, cable, powder, limb, air stream, seed.
• Means — the technical mechanism. Clamp, magnet, spring, screw thread, vacuum, elastomeric sleeve.

Your query is the product of the three columns, not one flat list. Four to six terms per column gives reasonable coverage for most mechanical and electromechanical inventions; two terms per column almost guarantees a shallow search, and ten begins to produce noise you will not read.

A useful trick: describe the invention twice — once as you would to a ten-year-old, once in the driest technical register you can manage. Plain language finds consumer-facing filings; technical language finds corporate ones. You want both.

Build Synonym Groups Like a Drafter Would

Patent databases run in English, and translation is where most searches break. For every concept, prepare variants along four axes:

• Generic nouns. Device, apparatus, assembly, mechanism, system, unit, arrangement, means. Drafters cycle through these, and each pulls a different set of documents.
• Spelling. US and British forms both appear: color and colour, tire and tyre, aluminum and aluminium, meter and metre.
• Truncation. An asterisk covers word families — fasten* catches fastener, fastening and fastened; seal* catches sealing and sealant. Use it on every stem where the ending varies.
• Obsolete terms. Older filings use vocabulary nobody uses today. What is now a "wearable device" appears in older documents as a "body-worn appliance," and a drone is a "remotely piloted vehicle" or "unmanned aerial vehicle." Old documents are still prior art.

Add adjacent industries deliberately. Medical device language, automotive language and packaging language describe the same physical mechanism with completely different words, and cross-industry hits are where the surprises live.

Mine Good Documents for Vocabulary

One well-chosen patent document is worth more than an hour of brainstorming. Once a search returns something genuinely close, harvest it: the title, the abstract wording, the terms used in the independent claim, the alternative terms listed in the description, and the classification codes on the front page. Working through how to read a patent pays for itself here, because the description usually spells out synonyms explicitly — drafters list alternatives on purpose to broaden coverage.

Operators, Fields and Classification

With the lists built, assemble real queries. OR inside each synonym group, AND between groups, quotation marks for exact phrases, and NOT only sparingly — it is the fastest way to exclude the document you needed. Restrict fields when noise gets high: searching title and abstract only is tight, searching full text is broad, and searching claims only finds documents that actually claim your feature rather than mentioning it in passing. Some databases support proximity operators requiring two terms within a set number of words, which is far more precise than a plain AND. The field code syntax differs by database — our walkthrough of Espacenet patent search shows one complete implementation.

Then step outside language entirely. A patent classification search finds documents by the technology an examiner assigned them to, catching everything your synonym lists missed. Keywords and classification are not alternatives; the combination is the search.

Work in Iterations, and Log Them

Treat the process as a loop rather than a single query: search broad, skim results, harvest new vocabulary and codes, search again more precisely. Three or four iterations is enough for most projects. Semantic and AI patent search tools can shortcut vocabulary discovery by matching meaning rather than words, but they do not replace a structured term list — they are best used as one more iteration that suggests terms you had not considered.

Keep a log table with the exact query string, the database, the hit count, and the documents pulled from it. Without a log you repeat searches you already ran and never notice the ones you never ran.

How to Tell Your List Is Good

Two signals suggest you have reached the center of the field. First, results start including documents from industries you never considered — a sign your language is now functional rather than product-specific. Second, the same applicants and inventors keep reappearing across different queries, which means you have found the people working on this problem rather than a random slice of documents. If neither happens, your terms are still too close to your own product name.

Conversely, if a query returns tens of thousands of hits, one of your groups is too generic — usually the means column, where "device" and "system" swallow everything. Tighten that column, not the others.

One boundary worth stating: a keyword list finds documents, it does not interpret them. Whether a document actually blocks you is a legal question, and Projects House is an engineering firm, not a law firm — this article is educational only, and patentability or infringement conclusions belong with a registered patent attorney.

Bring Us the Findings

Search results are engineering input, not just legal input. A crowded field means differentiation has to be designed in, and a blocking document is a design constraint we would rather know on day one than after tooling. Once you have a list of documents that worry you, send them to us through the contact form along with your concept, and we will tell you what it means for the design.