IPC section G reference

IPC section G: scope, purpose and verified edition

IPC section G is titled "Physics". The current reference edition is 2026.01. Its coverage includes measurement, optics, control, computing and information processing, but that high-level heading is not an exhaustive subject-matter definition or a patent-search result. The relevant choice is which feature of a claimed invention belongs in the section, then which class, subclass and subgroup actually describes that feature. This page explains how to use the official classification as a research route, not how to obtain a legal opinion. WIPO, Guide to the IPC (2026)

Why this section has a distinctive search boundary

Section G is more than a container for software patents. It covers measurement, optical phenomena, computation, regulation and other physical information-handling functions. A technology may be called 'artificial intelligence' because it uses a model, but the claim might actually improve a measuring instrument, image formation process or device-control mechanism. Avoid selecting G06N merely because machine learning appears in the specification. The important question is whether the claimed computational arrangement, training technique or task-specific technical interaction drives the inventive contribution.

Official class families and candidate research routes

The following class-level routes are examples for investigating measurement, optics, control, computing and information processing. They are deliberately class-level, not fabricated subgroup allocations. An individual patent may belong to one, several or none of these candidates after examination of its actual technical disclosure. Each route must be checked against the official IPC scheme and the relevant subclass notes before it becomes a filter in a recorded research protocol.

An illustrative subordinate candidate is G06N, concerning Computing arrangements based on specific computational models. Verify its current title, inclusions and exclusions in the scheme instead of using the abbreviated label here as a complete definition. A class table provides an orientation map but cannot resolve conflicts between function, application, manufacturing route and detailed apparatus claims. When a class differs in IPC and CPC at lower levels, keep both pathways documented rather than asserting a direct correspondence. WIPO, Guide to the IPC (2026)

Worked research example: separating physical features from application terms

Consider a computer-vision system that detects defects in manufactured surfaces. One claim may concern an optical illumination and camera configuration; another may concern a trained model's inference architecture; another may concern the mechanical inspection line. An evidence map should distinguish image acquisition from feature extraction, classification and actuation. Searchers can then compare G01/G02/G06 pathways and relevant B-section machinery rather than relying on one 'industrial AI' filter. The eventual exact symbols require official review of claim substance, not merely an application label.

In an IPC-oriented search, write each claim feature as a neutral technical statement, then use the 2026.01 hierarchy to test possible subclasses. Record a broad initial search and a narrower group-based search. Review a sample that appears in the broad set but not the narrow one; this can reveal an omitted class or terminology change. Preserve the source IPC symbols printed on retrieved patents instead of replacing them with your modern working classification.

Boundary tests and neighboring classifications

A data-processing method is not automatically classified as a communication protocol because it exchanges messages. Conversely, data sent across an innovative network may be principally a transmission invention rather than a new computational model. In medical AI the underlying patient measurement or intervention can also be relevant to Section A. Section G research should explicitly test whether the signal-processing component is an invention in its own right or an ordinary component within a specialist instrument, with multiple paths kept where justified.

A useful boundary matrix for this section compares: Algorithm architecture versus measurement apparatus; optics versus image analysis; control method versus communications protocol. For each pair, record the feature said to distinguish the invention, the candidate class from the relevant scheme, the exclusion or limiting note, and the documents found using the competing route. The objective is to explain why a search set differs, not to create an exhaustive list of all inventions in the industry. A cross-section candidate should be retained whenever it captures a technically plausible part of the claim.

IPC hierarchy and revision controls

The IPC's Section G hierarchy provides a common international language for physics and information-related invention search. For a cross-border AI corpus, the searcher should preserve source IPC symbols, identify when subclass titles have changed and avoid interpreting newer CPC-specific groups as timeless IPC equivalents. WIPO's guide explains that the class title is only a broad indication; exact subgroup meaning depends on the complete chain of parent entries and notes. The claim's operative technique must lead the class selection. WIPO, Guide to the IPC (2026)

For IPC section G, compare present search results with the historical classification of documents from earlier years. WIPO provides past schemes, the current 2026.01 publication and revision support. Annotate any changes in class or subgroup boundaries before reporting a trend. The 2026 guide explains that subgroup wording can depend on parent titles and indentation; reading a lone excerpt can therefore materially misstate the coverage of a patent-search filter. WIPO guide: layout and symbols.

Reproducible evidence packet for this section

A reproducible G-section packet includes algorithm steps, data transformations, apparatus components, sensor characteristics and an explicit description of the claimed technical effect. For AI patents, separate the input data, model architecture, training procedure, inference output and downstream control action. A diagram showing that flow can expose differences that a broad keyword search obscures. Record whether the claim focuses on a general computational model or a specific measurement application, and then compare the candidate class definitions.

1. Identify the patent or technical question and record the inventive feature being investigated.
2. Capture the current 2026.01 IPC section, class and candidate subclass from official source pages.
3. Record included and excluded subject matter, limiting references, and any related sections.
4. Run a functional-language search separately from the classification-based query.
5. Compare result sets at patent-publication and family level; retain false-negative and false-positive examples.
6. Date the source consultation and disclose uncertainty or incomplete classification mappings.

How to assess completeness without making unsupported conclusions

Do not assume a patent is evidence that a claimed AI model works in production. Classification indicates the subject matter assigned by an office, not an independent validation of performance, novelty or patent eligibility. Review claims, figures and examples before preparing metrics such as patent counts or technical maturity. For trend studies, an explicit decision is required on whether to count all documents with computer-related classification or only those directed to a specified computer model, test or measurement function.

For physics, quality of a classification-based research result should be measured through documented decisions: was the relevant technical function described, were adjacent classes evaluated, could another reviewer reconstruct the symbol list, and did the method inspect material documents rather than rely on bibliographic codes alone? The scheme itself does not provide a ranking of patent strength. Evidence gaps should be shown in the result, particularly when published sources are incomplete, data platforms use different classification vintages or the technical field crosses section boundaries.

Primary authorities and related comparison

• WIPO official Guide to the IPC (2026)
• WIPO authentic IPC publication
• WIPO edition 2026.01 notice

This research page does not assign definitive symbols to a particular invention. For the corresponding CPC section, check its independent scheme, definitions and version before translating search terms. Consult an appropriate professional for claim-specific legal decisions; a public classification guide explains evidence routes but does not determine legal status, novelty, inventiveness or freedom to operate.