Find similar patents with AI

Start from one patent and get the closest patents in the field, ranked 1 to 10 for similarity, with claim-overlap charts and the differences spelled out.

A similar patents search starts from one patent and finds the documents closest to it, ranked by how similar they are. One Transform looks at the source patent's citations, classifications, inventors, vocabulary and family, scores each candidate from 1 to 10 for closeness, and for the closest candidates compares their claims with the source patent's claims element by element. Patent attorneys, analysts and R&D teams use it to see who works near an invention and to find the art a deeper analysis should start from.

What you get

  • Ranked matches grouped as highly similar (7 and above), moderately similar (5 to 6.9) and loosely related (under 5), each with its assignee, dates and a breakdown of why it ranked where it did: technical similarity, citation proximity, classification overlap, timing and commercial signals.
  • Claim-overlap findings for the closest candidates: an element-by-element chart against each independent claim of the source patent, with a note on whether the candidate could anticipate the claim and how it bears on obviousness.
  • A differences write-up for top candidates: the distinguishing features, what they achieve technically and where a design-around could lie.
  • Deeper analysis of the closest matches, covering why they are similar, how they differ, and how they relate to the source patent through citations, family or examiner art.
  • A similarity report synthesising the ranked results: the closest art, confidence in each match, family and assignee context, and the sub-themes the matches fall into.
  • Charts and exports: an Insights tab with sub-theme clusters and the score spread, and Excel, PDF and Word exports with the same content.

How it works

  1. Source analysis. The source patent is loaded with its claims, classifications, citations, inventors and family.
  2. Discovery anchored on the source. Candidates come from documents close in meaning, patents that cite or are cited by the source, patents frequently cited alongside it, the same classifications, the source's own vocabulary, the same inventors, and recent US grants in the same classes from USPTO data. The source patent's own US continuations, divisionals and continuations-in-part are added from USPTO continuity data, because they are by definition among the closest documents.
  3. Deduplication. The same patent under different number formats or kind codes counts once, and the source patent is never listed as its own match.
  4. Scoring. Each candidate is scored for closeness to the source. Classification overlap is computed from the shared classification codes, citation proximity from the actual citation links and co-citations, and commercial signals from citation counts and family size; technical similarity and timing are model judgments on the text.
  5. Depth on the closest matches. The top candidates get the claim-overlap charts, the differences write-up and the deeper analysis, and the report is written last.

When to use it

  • Before a validity or prior art project, to find the closest documents to start from.
  • When a new patent issues in your space, to see what sits next to it and who holds it.
  • When valuing or licensing a patent, to understand how crowded its neighbourhood is.
  • When drafting continuation claims, to see how close the nearest third-party art comes to each independent claim.

Similar patents works from one patent outward. To map a whole technology field from a description, use a technology landscape. To build an invalidity case against the source patent, run a prior art search, which filters by priority date and maps references against the claims.

Limits

  • A similar patents search needs a source patent. It does not start from a plain description; the technology landscape and white space searches do.
  • Matches are not filtered by the source patent's priority date, so later patents appear alongside earlier ones. That is the point of a similarity search, but it means not every match is prior art.
  • Claim-overlap notes on anticipation and obviousness are research pointers, not conclusions.
  • Similarity scores are relative to the source patent. A 9 means close to this patent, not important in absolute terms.
  • The results support professional review and are not legal advice.

Frequently asked questions

How is similarity measured?

Each candidate is scored from 1 to 10 for closeness to the source patent. The score combines technical similarity read from the text with signals computed from the patent records: shared classification codes, direct citations and co-citations, and citation and family counts. Each match shows that breakdown.

Why do continuations of my own patent appear?

Continuations, divisionals and continuations-in-part share the source patent's specification, so they are usually the closest documents there are. They are included so you can see what the family itself already covers; patent families explained covers how they relate.

Can I narrow the results?

Yes. You can set keywords, classifications, jurisdictions, a date range, patent status and a minimum similarity score, include or leave out published applications, and exclude assignees.

A similar patents search ranks closeness to a patent, whatever the dates. A prior art search keeps only material published before the priority date and maps each reference against the claim elements to build an invalidity case. Both are included in every plan; see pricing.

Try it on one of your patents

The fastest way to evaluate One Transform is to run a search on a patent you already know well.

7-day trial with 5 searches · No credit card · Work email required