Interactive explainer

Patent Classification Explorer: How CPC Codes Widen a Prior-Art Search

Every patent is filed into a vast hierarchical taxonomy — the CPC. Walk the tree below and you’ll see something surprising: the prior art for one self-balancing scooter is scattered across different sections entirely — the vehicle, its balance controller, and its electric drive.

Explore the classification tree
BPerforming Operations; Transporting
B62Land vehicles for travelling otherwise than on rails
B62KCycles; cycle frames; steering devices
B62K 11/00Motorcycles, engine-assisted cycles
B62K 11/007Single-wheel or self-balancing cycles
B62K 3/00Bicycles
B62MRider propulsion; transmissions
B60Vehicles in general
B64Aircraft; aviation
GPhysics
G05Controlling; regulating
G05DSystems for controlling non-electric variables
G05D 1/08Control of attitude, i.e. control of roll, pitch or yaw
G06Computing; calculating
Selected classification
B › B62 › B62K › B62K 11/00 › B62K 11/007
B62K 11/007
Single-wheel or self-balancing cycles

The vehicle itself: where a self-balancing scooter is classified.

2,100
classified here
2,100
in this whole branch

Searching at a classification level sweeps in every document in its branch — related art that shares the concept, whatever words it uses. That is the strength keyword search lacks.

Classification is a map of ideas, not words

The Cooperative Patent Classification (CPC) sorts every patent into a tree of more than 250,000 categories — from broad sections (A–H, Y) down through classes, subclasses, and groups. Crucially, documents are filed by what they are, not by the words they happen to use. That makes classification a powerful complement to text search: find the right node and you inherit every document in that branch, however it’s worded — including the foreign-language and differently-phrased references a keyword query never reaches.

The catch: one invention, many branches

Expand the tree and select B62K 11/007 — the self-balancing cycle itself. Now notice G05D 1/08, sitting in a completely different section (Physics): that’s the attitude-control system that makes the thing balance. And B60L covers the electric drive. The prior art that could invalidate a self-balancing-scooter patent lives in all three places at once. Toggle “Show keyword hits” and the problem snaps into focus: the literal phrase lands on a single node, while the genuinely relevant art is spread across the tree.

Why this is hard by hand. Classification only helps if you already know which branches to look in — and the most dangerous reference is often in a branch you didn’t think to check. The skill is knowing that a “vehicle” invention also lives under “control systems.”

How AI uses classification

A concept-based search doesn’t make you guess the branches. It understands that a self-balancing transporter is a vehicle and a control system and an electric drive, and pulls relevant art from each — then uses classification as a signal to sharpen, not as the only net. See How AI Prior-Art Search Actually Works for the retrieval mechanism, and The Vocabulary Gap for why text alone leaves so much uncovered.

A note on this demo: the tree is a small, simplified slice of the real CPC with illustrative document counts. The actual scheme is far larger and is re-balanced regularly, but the structure — and the “one invention, many branches” problem — is real.

Don’t guess the branches

PatentScan reasons across classifications automatically — finding the vehicle, the controller, and the drive in one search, so the reference in the branch you forgot doesn’t slip away.