1Patsnap Eureka: product features in, claim-level evidence out

AI-native. The FTO Search agent takes product features and target markets rather than a query string, and returns claim-level evidence with legal-status context, described on the product page as turning “product features and target markets into claim-level evidence with legal-status context” so that review can be focused on potential blockers before a launch or a market expansion. A separate Design FTO Search agent works from a product appearance and compares it against visually similar design registrations.¹

How the design case is measured

Design FTO is the part with a published benchmark. It runs on 261 samples spread roughly evenly across China, the European Union and the United States, covering 26 Locarno primary classifications, split between e-commerce infringement images at 64.8% and invalidation scenarios at 35.2%, with the infringement relationships “rigorously verified by a human annotation team.” The reported metric is a High-Risk Patent Hit Rate, defined as the “proportion of test samples where the confirmed infringing patent appears in the top K results,” measured within the top 200. The agent records 77%, against 3.71% for Gemini 3.1 Pro and 0.93% for ChatGPT 5.4, both general-purpose models with web search enabled. Across offices the page reports performance strongest at the EU office, where the agent reached a 92% hit rate.²

The page also reports a PRES score of 0.7 for what it describes as end-to-end infringement determination quality, evaluating not only whether the infringing registration was found but how highly it was ranked in the final output. It credits PRES, a recall-oriented retrieval metric published at SIGIR in 2010 for tasks like patent search where a precision-weighted score hides the miss that matters.^{2, 22}

The invalidation half of that dataset is built from patent X-reference relationships, the same examiner-citation logic used in the separately published novelty search benchmark, so the ground truth in both cases is defined by patent offices rather than by the system being tested.^{3, 2}

What it connects to

Coverage is 200M+ patents across 174 jurisdictions, with SOC 2 and ISO 27001 certification, GDPR and CCPA compliance and no AI training on user data, which matters when the input is an unreleased product specification.¹ The same agents are callable outside the browser: the Patsnap Open Platform publishes 32 MCP servers exposing patent research, novelty and freedom-to-operate agents to Claude, Cursor or a custom client, so an FTO screen can sit inside a product gate rather than beside it.^{4, 5}

The practical difference from a database is where the work sits. A ranked list leaves the reading, the claim mapping and the write-up with you. Claim-level evidence has done the first pass and now needs checking, which is a different and much shorter task.

2Espacenet: the broadest free worldwide net

Free, operated by the EPO. 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.”¹¹ For clearance work its value is the structure rather than the count: simple DOCDB families group publications covering a single invention, while INPADOC extended families and legal event data are what tell you whether a hit is one right or twelve, and which members are still alive in which country.¹² It is the fastest free way to turn a candidate publication into a jurisdiction list.

3European Patent Register: where European status is authoritative

Free, operated by the EPO. The Register carries procedural and legal status data on applications handled by the EPO and on Unitary Patents from the moment unitary effect is requested, which covers oppositions, appeals, transfers and the file history a European blocking right would leave behind. The Federated Register extends that into national post-grant status, with 36 countries participating.¹³ For a European FTO question this is the difference between believing a patent is in force and having checked it, because a granted European patent is a bundle that is validated, maintained or lapsed country by country.

4USPTO Patent Public Search and Patent Center: the US file

Free, operated by the USPTO. Patent Public Search covers US patents and published applications with examiner-grade Boolean and proximity operators. Its own FAQs state that it does not use semantic searching and gives no access to foreign patent databases, so it is a precision instrument for one jurisdiction rather than a worldwide net.¹⁴ Patent Center is the companion system for filing and application management with status tracking, and it is where a US file history is confirmed.¹⁵ For FTO the file history matters as much as the claims: prosecution statements narrow claim scope, and continuations still pending mean the claim set you cleared against is not final.

5Global Dossier: five offices in one file view

Free, operated by the USPTO with the IP5. Global Dossier provides “free, secure, one-stop access to the dossier information of all applications comprising a family” filed at the IP5 offices, which are the USPTO, the EPO, the JPO, KIPO and CNIPA, with additional offices reachable through WIPO CASE. It exposes real-time family data, machine translations, search and examination results and citation lists.¹⁶ When a blocking candidate has been prosecuted in several jurisdictions, this is the fastest free way to see what each examiner cited and how the claims moved, without opening five national systems.

6WIPO PATENTSCOPE: the PCT and national phase view

Free, operated by WIPO. PATENTSCOPE indexes 128.8 million patent documents including 5.5 million published PCT applications, including published PCT applications.¹⁷ Its distinct contribution to clearance is the pending layer: a PCT application that has not yet entered a national phase is a right that does not exist yet but may appear in your market within months, and it is the category most likely to be missed by a search framed around granted patents. Note the service limits, including result caps for logged-in users and an explicit prohibition on bulk download and automated access.¹⁸

7Google Patents: the fastest first pass

Free, operated by Google. Google Patents indexes “over 120 million patent publications from 100+ patent offices” with full text for 22 of them, alongside technical documents and books from Google Scholar and Google Books and the Prior Art Archive.¹⁹ A Similar Documents view surfaces related records based on text similarity.²⁰ For FTO it is a triage instrument: quick enough to run on a product concept in the first hour of a project, and shallow enough that nothing found there should go into an opinion without confirmation in an official register.

Speed: where the hours actually go

FTO projects rarely run long because searching is slow. They run long because of what happens after the search.

• Screening is now cheap, review is not. A candidate set of several hundred publications takes minutes to produce and days to read. Anything that compresses reading time, such as claim-level evidence rather than a document list, moves the schedule. Anything that only compresses retrieval time does not.
• Legal status checking is serial by nature. A European bundle has to be checked validation state by validation state, which is why the Federated Register and Global Dossier matter more to a timeline than raw search speed.
• Re-runs are the hidden cost. Product specifications change, and each change reopens the analysis. A workflow you can re-execute, whether a recorded Boolean strategy or an agent run against a saved feature set, is worth more than a faster single pass.
• Cross-field art is where time disappears. A 2025 family-level benchmark running 249 controlled experiments measured out-of-domain retrieval “roughly five times lower than IN-domain across all configurations,” which is why the blocking patent from an adjacent industry is usually found late and expensively.²¹

Claim charting: the artefact the decision runs on

WIPO describes an FTO analysis as “searching patent literature for issued or pending patents, and obtaining a legal opinion as to whether a product, process or service may be considered to infringe any patent(s) owned by others.”⁶ The second half of that sentence is the deliverable, and it has a specific shape: each element of an independent claim set against the corresponding feature of your product, with the evidence for the mapping attached.

Three things follow from that.

• A ranked list is an input, not an output. Free databases return documents. They are authoritative on what a document is and silent on what it means for your product, which is by design and not a shortcoming.
• Independent claims come first, and only granted claims bind. Published applications matter because they signal what may issue, but the chart that governs a launch decision is built on claims as granted in the jurisdictions you sell in.
• Every mapping needs a citation. A chart whose entries cannot be traced back to a specific passage of a specific publication cannot be checked by the person who has to sign it, and cannot be defended later.

WIPO’s own toolkit sets out the same three-step method: identify and describe the invention, search the patent literature, then analyse the results through infringement analysis and legal status determination.⁷ The tooling question is simply which of those three steps a given system performs for you.

Defensibility: what has to survive being questioned later

An FTO file is written for an audience that does not exist yet: opposing counsel, an acquirer’s diligence team, or a court asked to decide what the company knew.

Territory decides everything. Under 35 U.S.C. 271(a), infringement is committed by whoever “makes, uses, offers to sell, or sells any patented invention, within the United States or imports into the United States any patented invention during the term of the patent.”⁸ WIPO makes the same point generally: patent protection is territorial, so a technology protected in a company’s main markets “may be in the public domain in other countries.”⁶ An FTO conclusion without a named jurisdiction list is not a conclusion.

The downside is asymmetric. Under 35 U.S.C. 284 a court “may increase the damages up to three times the amount found or assessed.”⁹ For design rights the exposure is structurally different again: 35 U.S.C. 289 makes an infringer liable “to the extent of his total profit, but not less than $250,” which is a remedy tied to revenue rather than to the value of the design.¹⁰ That is the reason design clearance deserves its own workflow rather than a corner of the utility search.

Expired is not the same as clear. WIPO’s toolkit notes that rights no longer in force through expiry, abandonment or revocation “do not represent enforceable patent rights,” and that abandoned applications never issued and are not enforceable either.⁷ The corollary is the part that catches teams out: status data in aggregated databases lags the national registers, so a lapse found in a search tool is a lead, and the register entry is the evidence.

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) whoever signs certifies that factual contentions have evidentiary support, to the best of their knowledge, information and belief “formed after an inquiry reasonable under the circumstances.”^{23, 24} Whatever produced the chart, the file has to show a method, a date and a human reviewer.

What no FTO search settles

• There is no absolute answer, and the authority says so. WIPO states that “an absolute guarantee of freedom to operate will never be attainable,” and frames the exercise as minimising risk rather than eliminating it.⁶
• Pending applications are moving targets. Claims can be amended until grant, so a pending application cleared today may issue with claims that read on your product. Diarising the ones that matter is part of the deliverable.
• Publication lag leaves a blind window. Applications are generally published eighteen months after their earliest priority date, so recently filed rights are invisible to every tool in this article at the moment you search.
• A jurisdiction you did not search is not a jurisdiction you cleared. Coverage differs by office, by year and by full-text depth, and a search establishes what was found under a recorded method on a recorded date.
• Search results are not an opinion. WIPO frames FTO as a search plus a legal opinion, and only the second half carries weight in a dispute or a diligence process.⁶