Patent Quality does two things. First, it assesses whether your invention is patentable — eligibility, novelty, non-obviousness and disclosure — with the reasoning an examiner would apply. Second, it surfaces every similar idea already public: granted patents, published applications, papers, and open-source code, mapped claim by claim.
Claude reads your invention disclosure or draft claims and evaluates them against the statutory requirements of 35 U.S.C., with a written rationale for each score so counsel can check the reasoning, not just the number.
Runs the Alice/Mayo two-step analysis — is the claim directed to an abstract idea, and if so, does it recite an inventive concept or practical application? Scored twice — against USPTO examination practice (2024 AI eligibility examples, Ex parte Desjardins) and against Federal Circuit precedent (Recentive v. Fox), because the two currently diverge for ML claims.
Checks whether any single prior-art reference discloses every element of a claim. Flags your own public disclosures — talks, preprints, blog posts, public repos — and the one-year U.S. grace period they start.
Looks for combinations of references a skilled practitioner would plausibly join, following the Graham factors and KSR. Highlights which claim elements carry the inventive weight — and which are routine.
Tests whether the description supports the full scope of the claims and whether claim terms are definite. Catches functional "black box" ML language that commonly draws written-description rejections.
Flags issues for absolute-novelty jurisdictions and the EPO's technical-character and problem-solution approach, so a U.S. filing doesn't foreclose protection abroad.
Inventors must be natural persons (Thaler v. Vidal), and under the USPTO's November 2025 guidance AI is treated as a tool. Records which human conceived each claimed element — and what the AI contributed.
Patent-only searches miss where AI innovation is actually published. Patent Quality searches patent offices, the research literature and open-source repositories together, then de-duplicates patent families so you review each invention once.
Embedding search finds references that describe the same idea in different words; CPC classification codes and keyword queries catch what embeddings miss.
Each independent claim is split into elements and mapped to the passages in each reference that disclose them — the format attorneys and examiners already use.
Follows forward and backward citations from the strongest hits to find related art a single query would never reach.
Clusters the results by approach and assignee, showing who is active in your area — and the gaps where a narrower claim is still open.
Paste an invention description or draft claims, or send a candidate from Patent Feasibility.
Parallel search across patent offices, literature and code; families de-duplicated.
Claim elements mapped to the closest references, with overlap scores.
§101 / §102 / §103 / §112 scores with written rationale and suggested claim narrowing.
Export a PDF for counsel. Reports are kept in confidence, never published.
Patent Quality is a decision-support tool for inventors and patent counsel. It does not provide legal advice or a legal opinion on patentability, and does not replace review by a registered patent attorney or agent.
We're onboarding design partners now — direct access to the founding team and first access to new modules.