PatentsView — USPTO Patent Data API

It is the research platform built over the United States Patent and Trademark Office's own grant record, and it does the two things raw patent files never do: it disambiguates the actors and it classifies the technology. Millions of granted patents arrive as rows keyed by patent_id, carrying titles, abstracts, grant dates and patent types, joined through disambiguated inventor, assignee and location tables. Grants reach back to 1976 and pre-grant publications start in 2001, so one table spans the entire modern biotech, software and semiconductor eras.

Around the spine sit the joins that make the corpus analytic rather than merely searchable: classifications across four systems (USPC, CPC, IPC and the WIPO technology fields), backward and forward citations both US and foreign, other references, related applications, applicant and attorney rosters, and full text down to individual claims.

For research and consulting work this is the output side of the innovation economy - where NCSES counts R&D dollars going in, this record shows what came out the other end. Get a sample of this dataset scoped to your technology classes.

A few rows from different corners of the same corpus:

# PatentsView -- USPTO Patent Data -- grant row -- research & consulting services
patent_id        : 10000000
patent_title     : Coherent LADAR using intra-pixel quadrature detection
patent_date      : 2018-06-19
patent_year      : 2018
patent_type      : utility

# an older grant from the same table - the history runs deep
patent_id        : 7861317
patent_date      : 1981-10-06
patent_year      : 1981

# classification join -- the technology address per grant
patent_id        : 9954111
cpc_class_id     : H01L            (semiconductor devices)

# entity join -- who owns it, resolved to a stable identity
patent_id                        : 11872626
assignees.assignee_organization  : DIVERGENT TECHNOLOGIES, INC.

Read them as one document wearing three hats. The first two rows are the bibliographic spine - what was granted and when, from 1981 hardware to a 2018 optics grant, in one continuous series. The classification join adds the technology address: cpc_class_id H01L pins a patent inside the semiconductor-device literature, which beats the word "chip" in a title by any measure you care to run. The entity join carries ownership - and because the organization name travels with its disambiguated form, "Divergent Technologies" stays one company even when the printing changes.

Values arrive exactly as the office publishes them: raw codes and un-normalized names alongside their resolved identifiers. Get a sample of this dataset and read your own slice.

Twelve fields carry the core extract, doing four jobs. patent_id is the join surface everything else hangs off. The bibliographic group - patent_title, patent_abstract, patent_date, patent_year, patent_type - describes the document. patent_num_times_cited_by_us_patents measures its footprint in one column. And the relational group reaches outward: inventors.inventor_name_last and assignees.assignee_organization name the actors, application.filing_date timestamps the idea rather than the grant, cpc_class_id places it technologically, and citation_sequence keeps reference lists faithful to the printed document.

Everything else rides under additional fields on request: pre-grant publications, applicant and attorney rosters, location identifiers, the remaining classification systems, foreign citations, related applications and the full-text surfaces.

<!--TABLE:field_dictionary-->

• Geography: United States grants and pre-grant publications, with worldwide reach in the entity tables - a US-granted patent with a Tokyo assignee and a Munich inventor keeps all three geographies.
• Temporal: the full history of US patents. Grants run back to 1976, pre-grant publications to 2001 - five decades long enough to watch a technology class rise, peak and be absorbed into the next one. Each build is stamped with its data-through date, so trend work knows exactly where the window closes.
• Granularity: record-level entities with cross-entity joins. One row per document on the spine, then one row per classification, per inventor, per assignee, per citation in the joined tables - the fan-out that makes network analysis possible instead of approximate.

Against the wider catalogue - average quality score 7.81 across all 1,744 datasets - this slice scores 6/10, docked mainly for self-documentation that runs thinner than a statistical agency's. The entity resolution, meanwhile, is something no raw patent dump provides at any price.

API, files, or your warehouse. Daily, weekly, or hourly.

The delivery step earns its keep here, because the hard part of patent data was never finding it - it was assembling it. Reconciling inventor spellings into one person, aligning classification versions across decades, and rebuilding citation joins are exactly the projects that eat a quarter and then need defending forever. Delivered extracts arrive with that assembly done: identities resolved, classifications decoded, joins keyed on patent_id. Cadence is yours to set - daily for monitoring, weekly for landscaping refreshes, hourly if a pipeline demands it.

Technology landscaping and white-space mapping. Subgroup-level classification turns a crowded field into a map: count filings by CPC class across five decades and the thin patches announce themselves.

Competitive intelligence on R&D. Filing counts, class mix and claim depth by disambiguated assignee give the earliest public signal of where a rival's engineering budget is going - years before a product roadmap confirms it.

Citation-network analytics. Backward references and forward citing documents make influence traceable, so a foundational patent can be followed into everything built on it, with citation_sequence preserving the order the examiner saw.

Innovation benchmarking and policy studies. University-versus-corporate output splits and national innovation comparisons computed from the official record - quotable in client decks because the citation traces to the patent office itself.

Prospecting for IP-adjacent services. Firms that file in a given class are named, classified buyers for prior-art search, patent-analytics and translation vendors; assignees.assignee_organization is the account list.

Market Researchers & Consultants get the innovation-output layer their engagement models usually fake - technology trends measured from filings rather than sketched from a vendor deck. Competitive Intelligence & Product Teams get rivals' R&D direction as an always-current feed of classified, owned, dated documents. Data Scientists & ML Engineers get stable entity keys and panel structure clean enough to model directly, skipping the entity-resolution project that usually eats the first month. Journalists, Academics & Students get the citable official record behind almost every story about American innovation.

Related pages: market researchers using research & consulting services data · competitive intel & product teams using research & consulting services data · data scientists using research & consulting services data

Three things worth saying plainly. First, names arrive un-normalized by design: the printed organization name travels beside its resolved identity, so analyses should group on the identifier and label with the name - never the reverse.

Second, classification depth varies by era. A recent patent may carry a dozen CPC entries while a 1976 design grant carries one legacy code, so class-count comparisons need era-aware baselines; the multi-system classification tables exist precisely to bridge that gap.

Third, the platform itself is being folded into USPTO's own data infrastructure, and the plumbing is moving with it. Datadory absorbs that churn so your extracts stay continuous - and every build is stamped with its data-through date, so pin snapshots if a report will be re-run. Name the classes, assignees or years you want and the sample arrives cut to them. Get a sample of this dataset