1Patsnap Eureka: disclosure in, feature comparison out

AI-native. The Novelty Search agent takes a technical disclosure rather than a query string. It summarises the technical solution, extracts the distinguishing features, derives search elements, then runs semantic, Boolean, classification and citation-tracking strategies in parallel alongside paper and web routes, screens the results, and returns a comparison report mapping every technical feature against each close reference, with references tagged for the role they play and each mapping linked back to the passage it came from.¹ Every intermediate step is visible and editable before it runs, so a searcher can correct a feature extraction or add a classification code rather than accepting a black-box result.

How it is measured

Patsnap publishes a benchmark methodology, PatentBench, in which ground truth is the set of X references cited by examiners at different patent offices, deduplicated and normalised by patent family, over 340 cross-jurisdiction samples. Two metrics are defined: X Hit Rate, a correct answer appearing in the top 1, 3 or 5 results, and X Recall Rate, the share of correct answers found within the top 100. On the round tested in July 2026 the agent records an 85% X Hit Rate and a 37% X Recall Rate.² The sample size, ground-truth definition and metric definitions are published alongside the result, which is what allows the number to be interrogated rather than accepted.

What it connects to

Coverage is 200M+ patents across 174 jurisdictions, with SOC 2, ISO 27001, GDPR and CCPA compliance and no AI training on user data, which matters when the input is an unfiled invention.¹ The same agents are published as MCP servers, so a novelty search can be triggered from Claude, Cursor or an internal assistant rather than only from a browser, with a free Starter tier of 10,000 credits for 90 days.^{3, 4} New users start with 10,000 free credits.

The practical difference from a database is where the work sits. A ranked list leaves the reading, the mapping and the write-up with you. A feature comparison has done that and now needs checking, which is a materially smaller job and one an attorney can supervise rather than perform.

2Espacenet: the broadest free worldwide collection

Legacy. The European Patent Office announced in February 2024 that Espacenet offers “more than 150 million patent documents” from “more than 100 patent authorities around the world.”⁵ For novelty work its value is less the raw count than the structure around it: simple and INPADOC extended family views plus legal event data, which is what lets you check whether a reference is what you think it is.⁶ Its one AI element is narrow and labelled as such, a CPC text categoriser that suggests classification symbols from a natural language description. Exports are capped at the first 500 results.⁷

3WIPO PATENTSCOPE: the multilingual baseline

Legacy, with an AI assistant added in 2026. PATENTSCOPE indexes “128.8 million patent documents including 5.5 million published international patent applications (PCT)” across 123 national and regional collections, at no cost.⁸ Its distinguishing feature for novelty work is Cross Lingual Expansion across 14 languages, which addresses the failure mode that quietly defeats most keyword searches when the closest art is not in English.⁹ In July 2026 WIPO added an AI assistant that turns natural language into structured queries, hosted on WIPO premises.¹⁰ Result caps sit at 10,000 for logged-in users and bulk download is prohibited.¹¹

4USPTO Patent Public Search: examiner-grade operators, US only

Legacy. The USPTO’s free tool covers US documents only, with full text of most patents from 1971, pre-grant publications from March 2001 and OCR text back to 1836, searchable with the graded proximity operators examiners use.^{12, 13} Two sentences from its own FAQ define its role in a novelty search: it “does not use semantic searching,” and it “does not have access to foreign patent databases.”¹² Worth knowing alongside that: examiners themselves are required to run an AI similarity search in the USPTO’s internal system on plant and utility applications.^{14, 15}

5Google Patents: the fastest route to non-patent literature

Legacy. Google Patents indexes “over 120 million patent publications from 100+ patent offices around the world” alongside technical documents and books from Google Scholar and Google Books and material from the Prior Art Archive, which makes it the quickest free way to reach a disclosure that was never a patent.¹⁶ Two caveats matter for novelty work: full text is indexed for 22 offices rather than all of them, and the “Similar Documents” list is described in Google’s own help as “based on text similarity” rather than as a semantic feature.¹⁷

6J-PlatPat: the route into Japanese classification

Legacy. J-PlatPat is described by the Japan Patent Office as “an official digital library for patents, utility models, designs and trademarks.”¹⁸ Its reason to exist in a novelty workflow is the classification route: alongside number and keyword search it exposes the FI and F-term systems through its classification search, and F-terms in particular index Japanese documents along facets that IPC and CPC do not capture. When the technology is one where Japanese filings dominate, that route finds art the other nine tools on this page will not.

Where each generation wins

• Screening volume: AI-native. When a search costs an analyst-day, you only search what you already believe in. When the marginal cost approaches zero, screening every disclosure becomes routine rather than a budget decision.
• Terminology you do not know: AI-native, plus PATENTSCOPE cross-lingual expansion. This is the failure mode keyword search cannot detect, because a null result looks the same as an absence.
• Verifying what a reference actually is: the official databases. Family structure and legal events are what tell you whether a hit is one document or twelve, and whether it is still alive.
• Reproducibility for the file: official database operators, or an AI platform that records its queries. An exact Boolean string can be re-run in five years; a prompt cannot.
• Non-patent literature: Google Patents for a fast free pass, an AI-native platform for paper and web routes inside the workflow.
• Jurisdiction-specific depth: the official databases. J-PlatPat for FI and F-term, USPTO Patent Public Search for US proximity operators, Espacenet for worldwide family and legal status.
• Producing the deliverable: AI-native. A feature comparison table or claim chart is the artefact the decision runs on, and no database produces one.

What no novelty search settles

• No search proves a negative. Coverage differs by office, by year, by full-text depth and by language, which is why WIPO publishes an office-by-office coverage table. A search establishes what was found under a documented scope, not what exists.
• Cross-field art is the hard case. A 2025 family-level benchmark running 249 controlled experiments found out-of-domain retrieval performance “roughly five times lower than IN-domain across all configurations,” and answered “none” when asked which method closes the gap.¹⁹ That is where the reference that kills a claim usually lives.
• Novelty is not patentability. Novelty is assessed one reference at a time. Inventive step is a separate analysis under a named jurisdictional framework, and no tool performs it for you.
• The certification stays with the person who signs. The USPTO states that “simply relying on the accuracy of an AI tool is not a reasonable inquiry,” and under 37 CFR 11.18(b) the signer certifies that contentions have evidentiary support after a reasonable inquiry.^{20, 21}
• Published accuracy figures are self-reported. Where a vendor publishes a percentage, ask for the sample size, the ground-truth definition and the metric behind it.