How complete is the patent corpus coverage?

Deep in software categories (G06F, G06N, G06Q, G06T, H04L, H04W) and growing weekly from the USPTO grant feed. Not yet exhaustive of every software grant and no other categories today. Use it to find the closest candidates fast, not to certify nothing else exists.

Why does a description work better than keywords?

The search is meaning-based (semantic embeddings). Describe the mechanisms: what the parts are, how they connect, what happens to data.

Is my search description stored?

No. It is embedded, searched, and discarded. Only public patent text lives in the corpus.

What does the match score mean?

Cosine similarity between your description and a claim limitation, 0 to 1. It ranks candidates. It is not a legal opinion on infringement.

Find patents similar to your work

Describe what you are building. PatentDistance searches every software-patent claim limitation it has captured and shows the closest ones so you can screen your design before you take it to a patent attorney.

Loading corpus size…

How to describe your design for the best matches

Name the parts and how they connect, not the marketing goal. "Collapsible multi-layer pouch that empties as fluid drains, with a self-sealing fill port" beats "a medical fluid product". For software, name the architecture: cache, sync protocol, rate limiter, model pipeline. Two or three sentences is enough.

Software patents only today: data processing (G06F), AI (G06N), business methods (G06Q), graphics (G06T), networking (H04L/H04W). Every patent lookup in the app feeds this corpus. Submitted designs are never stored.

What this search is

This is a semantic reverse search over 2.6 million patent claim limitations, built for one job: before you ship a product, find out which granted patents describe something similar. Traditional keyword search fails at this because patent language rarely matches engineering language. A claim about "a sealing collar that radially compresses against a rotating surface" describes a shaft seal, but nobody writes "shaft seal" in the claim. So instead of matching words, this search matches meaning.

Every claim limitation in the corpus is embedded as a vector, a numeric representation of what the text means. Your description is embedded the same way, and the search returns the limitations whose meaning sits closest to yours. The match score (0 to 1) is the similarity between the two. A 0.6 match is a strong signal that a patent is doing something structurally like your work. A 0.3 match is noise you can ignore.

The corpus is seeded from the USPTO’s own bulk full-text grant files, parsed weekly as new patents grant. Scope is deliberately focused: data processing (G06F), AI (G06N), business methods (G06Q), graphics (G06T), and networking (H04L/H04W), the CPC classes where software lives. Coverage of those classes is growing week by week and is not complete yet, so treat this as a strong first pass, not an exhaustive search. A patent attorney running a formal clearance search will always search wider than any tool can.

Everything here is public data. The corpus holds claim text from granted patents only. What you type into the search box is used to answer your query and is never stored.

How to use it with the rest of the site

The search is step one of a three-step workflow. Find candidates here, then use Screen this on any result to run a full clearance check on that patent against your design, or use Run batch screen to check the top 8 results in one pass and get a ranked verdict table. Search finds candidates; the check scores them properly, limitation by limitation, with the same engine that powers the clearance checker.

Questions about the corpus

How complete is the coverage?: Deep in software categories and growing weekly from the USPTO grant feed. It is not yet exhaustive of every software grant, and it holds no other categories today. Use it to find the closest candidates fast, not to certify that nothing else exists.
Why does a description work better than keywords?: The search is meaning-based. Describe the mechanisms: what the parts are, how they connect, what happens to data. That is what the embeddings score.
Is my description stored?: No. It is embedded, searched, and discarded. Only public patent text lives in the corpus.
What does the score mean?: Cosine similarity between your description and a claim limitation, 0 to 1. It ranks candidates. It is not a legal opinion on infringement.