Patent search for researchers

A patent has to disclose enough for a skilled person to reproduce the invention. That makes patent text unusually specific about apparatus, process parameters, tolerances and failure modes — exactly the detail a journal paper compresses into a methods paragraph or omits entirely.

There is also a large body of applied work that is patented and never published. In instrumentation, materials processing, formulation, device engineering and industrial chemistry, the patent literature is not supplementary; for some questions it is the primary record.

Publication timing helps too. An application publishes eighteen months after its earliest priority date regardless of whether it is granted, so patent disclosure often predates the corresponding paper by years.

Patent language is drafted to be broad, which makes it a poor keyword target. An attorney will write elongate flexible member where you would write cable, and deliberately so. Keyword search fails here more badly than anywhere in the scholarly literature.

Classification codes solve this. CPC and IPC are hierarchical schemes assigned by examiners describing what an invention is, independent of the words used.

1. Find two or three patents you know are on target, by any means — keyword, a company name, a citation in a paper.
2. Read the classification codes on their front pages. Note the ones that recur.
3. Browse the full contents of those subclasses. This is the step that finds what keyword search cannot.
4. Refine: narrow to a more specific group if the volume is too high, or step up one level if it is too low.

Search at least two, and always look at the patent family rather than a single document. A family is the same invention filed in multiple jurisdictions; the description can differ between members, and an English-language family member is often the readable route into a filing you found in another language.

Do not read it front to back. The parts that answer a research question are not in the order the document presents them.

1. Front page: title, abstract, assignee, dates, classification codes and cited references. Thirty seconds, and it decides whether to continue.
2. Background: states the problem and, usefully, what the applicant considered inadequate about prior approaches.
3. Detailed description and examples: the actual disclosure. Worked examples with real parameters are here, and this is what you came for.
4. Figures: process diagrams and apparatus, usually clearer than the prose.
5. Claims: last, and only if you care about legal scope. Claims define what is protected, not what is disclosed — everything in the description is disclosed whether or not it is claimed.

For research purposes the description matters and the claims usually do not. That is the opposite of how a patent attorney reads the same document, and it is why patent-reading advice aimed at legal work is misleading here.

Patent citations come in two kinds, and the difference matters. Applicant citations are supplied by the filer. Examiner citations are added during prosecution by an examiner whose job was to find the closest prior art — which makes them a curated relevance judgement, and a very good one.

Examiner citations frequently point at journal papers, which is one of the more effective ways to cross from the patent literature back into the scholarly literature. Forward citations — later patents citing this one — show where the technology went.

Give the inventors, the title, the patent or publication number with its two-letter country code and kind code, the assignee, and the date. Say whether you are citing a published application or a granted patent — these are different documents with different numbers, and the text can differ.