Question 1

What is a patent invalidity search?

Accepted Answer

A patent invalidity search is a targeted prior-art investigation aimed at finding a reference, or a minimal combination of references, that discloses every limitation of a specific issued patent claim before that claim's critical date (generally the priority/filing date, though pre-AIA patents can turn on the invention date). Unlike a general prior-art or freedom-to-operate search, an invalidity search is claim-driven: it starts from an already-granted, already-construed (or construable) claim and works backward to find art the examiner missed, rather than starting from a technology area and working forward to see what exists. The deliverable is typically a ranked list of candidate references plus, for the strongest candidates, a full element-by-element invalidity claim chart mapping each limitation of the challenged claim against the specific disclosure of that reference.Invalidity searches are used in several distinct postures: as a defensive tool when a company is accused of infringement and wants to challenge the asserted patent's validity; offensively, before filing an IPR/PGR petition at the PTAB or asserting invalidity in district court; and diligence-stage, when a company is evaluating whether to license, acquire, or design around a patent and wants to understand how vulnerable it is. The posture materially affects search depth and reference selection; an IPR petition, for instance, can only rely on patents and printed publications (35 U.S.C. §311(b)), so a strong system-prior-use or on-sale-bar reference found during a search may be extremely valuable for district court litigation but entirely unusable in an IPR.

Question 2

How do you perform a patent invalidity search?

Accepted Answer

An invalidity search begins with claim construction: the target claim must be broken into its constituent limitations, using the patent's own numbering or a logical breakdown of its clauses, and each ambiguous term resolved against intrinsic evidence (the claim language itself, the specification, and the prosecution history) before any searching starts, because searching against an unconstrued or misread claim wastes effort finding references that don't actually match what the claim covers once properly interpreted. From the construed claim, the analyst extracts a set of core technical concepts not just vocabulary, but the underlying structural or functional ideas each limitation represents and builds parallel search strategies for at least three source categories: patent literature, searched by classification codes as well as keywords; non-patent literature, including academic papers, standards documents, product manuals, and trade publications; and, where litigation posture allows, non-documentary evidence of prior public use, sale, or knowledge.Because a single reference that shows the invention was not novel can invalidate a claim on its own, while a combination of references can invalidate it by showing the claimed invention was an obvious combination of known elements, the search strategy typically runs two parallel tracks: hunting for one reference disclosing everything (the strongest possible outcome), and, failing that, identifying two or three references that collectively disclose everything along with a plausible motivation a skilled person would have had to combine them. The search is iterative; an early found reference often surfaces new vocabulary, inventors, assignees, or citation chains that seed the next round of searching, so a single search pass is rarely sufficient; a thorough invalidity search typically runs three to five iterative rounds, each informed by what the prior round found, before the analyst concludes that no better reference exists.

Question 3

What is the difference between a prior-art search and an invalidity search?

Accepted Answer

A prior-art search is the broader category: any search for material that predates a given date, undertaken for any of several purposes: patentability assessment before filing a new application, freedom-to-operate clearance before launching a product, landscape/competitive-intelligence mapping of a technology space, or invalidity. An invalidity search is a specific, narrower type of prior-art search whose defining feature is that it is claim-driven rather than concept-driven: it targets the exact limitations of one or more issued claims, with the specific goal of defeating those claims' validity, rather than surveying a general technology area.The practical difference shows up in how the search is scoped and documented. A general prior-art or landscape search (as in a taxonomy-building exercise) is organized around technical categories and tends to be inclusive, capturing many references to characterize a field. An invalidity search is organized around claim limitations and is comparative and adversarial in character; every candidate reference is immediately tested against the claim element-by-element, and references that don't map to a specific limitation are typically discarded rather than retained for landscape context. A reference an invalidity searcher would discard as irrelevant (because it lacks one dispositive limitation) might be exactly the kind of reference a landscape analyst keeps, because it still illustrates the state of the art in the category.

Question 4

How do you find prior art for a patent?

Accepted Answer

Prior art discovery draws on several distinct source pools, and a competent search touches all of them rather than relying on any single database. Patent literature is searched through classification systems the Cooperative Patent Classification (CPC) and International Patent Classification (IPC) codes, which group patents by underlying technical concept rather than by the specific words used to describe them in addition to ordinary keyword search, using tools like USPTO Patent Public Search, Espacenet, Google Patents, and commercial platforms (e.g., Derwent, PatSnap, Innography) that support citation-graph navigation and semantic/concept search. Non-patent literature includes academic databases (IEEE Xplore, ACM Digital Library, PubMed, arXiv for recent computer-science and AI work), conference proceedings, standards-body archives, trade publications, and product documentation predating the critical date. Public-use and on-sale evidence product launch dates, press releases, the Wayback Machine for archived web pages, user manuals with copyright dates, and FCC filing dates establish prior public availability even absent a formal publication.A frequently underused source is the file history and citations of related patents: the patent being challenged, its family members, and closely related patents in the same CPC subclass often cite (or were cited against, during prosecution) references the examiner already considered relevant, reviewing what the examiner rejected the application over during prosecution, and what the applicant successfully distinguished it from, both often surfaces close prior art quickly and reveals exactly what an examiner (and by extension a future defendant) will consider the closest known art. Inventor and assignee history is another efficient lever: searching a named inventor's full patent and publication history often surfaces earlier, less-polished versions of the same idea provisional filings, earlier continuations, academic papers predating the patent filing that disclose the same concept before it was refined into the claimed invention.

Question 5

How far back should a prior-art search go?

Accepted Answer

The formal legal boundary is the claim's critical date, generally its earliest effective priority date, which may be earlier than the issue date or even the filing date of the patent itself if the application claims priority to an earlier provisional or parent application (a chain that must be verified, not assumed, since a continuation-in-part can have different effective dates for different claims depending on which limitations were newly added). A search should therefore begin by confirming the actual priority chain from the patent's face and file history, not just accepting the listed filing date, since incorrectly assuming a later priority date can cause a searcher to prematurely accept a reference as prior art that the patentee can actually swear behind (for older, pre-AIA patents) or, more commonly today, to miss that a slightly earlier reference just barely misses the true critical date and is therefore useless.Practically, how far before the critical date to search is a judgment call driven by the technology's maturation pace rather than a fixed number of years. In fast-moving fields (software, AI, consumer electronics), the most valuable references are typically clustered in the one-to-five years immediately preceding the critical date, since that is when directly comparable implementations are most likely to exist; searching decades back in such fields mostly returns either foundational/generic art (useful for obviousness combinations establishing a well-known base component) or nothing relevant. In slower-moving mechanical or chemical fields, highly relevant art can be decades old, and a search artificially bounded to a recent window risks missing the single best anticipatory reference. A disciplined search states its date range explicitly and justifies it by the technology's pace, and extends the window if the highest-value modern references all turn out to cite a much older common ancestor.

Question 6

How do you search for prior art when patent terminology is different?

Accepted Answer

Patent drafting deliberately uses idiosyncratic, often deliberately broad or unusual vocabulary: a 'fastening means' instead of 'clip,' a 'photon-responsive element' instead of 'photodiode' both because patent attorneys draft to maximize claim scope and because independently-drafted patents in the same technical space are written by different attorneys with different vocabulary preferences, so keyword search alone systematically misses relevant art that describes the identical concept in different words. The core countermeasure is to search by classification code (CPC/IPC) rather than, or in addition to, keywords, because classification is assigned based on the underlying technical concept the invention represents, independent of the specific words used to describe it. Two patents claiming the same photodetector arrangement in completely different vocabulary will very likely share a CPC subclass even if they share no meaningful keywords.The second countermeasure is systematic vocabulary expansion before searching: for each core concept in the claim, the analyst should generate a synonym set spanning formal/technical terms, colloquial/product terms, older/legacy terminology (a concept that is now called one thing may have been called something else ten or twenty years ago, especially in fast-evolving fields), and terms used by non-US patent offices, which often reflect translation choices distinct from US drafting conventions. Citation-graph navigation following the web of references that patents cite and are cited by also sidesteps the vocabulary problem entirely, since it follows examiner- and applicant-identified relevance links rather than textual matching. Increasingly, feature-level semantic retrieval embedding-based search that matches the underlying technical concept a claim limitation describes rather than its literal wording is used specifically to close this terminology gap. Keyword similarity is an increasingly inadequate proxy for genuine technical-concept overlap, and modern search workflows should treat concept-level and passage-level retrieval as a required complement to keyword search, not an optional enhancement.

Question 7

How do you find non-patent literature for patent invalidity?

Accepted Answer

Non-patent literature (NPL) is frequently the most damaging category of invalidity art precisely because examiners search patent databases far more thoroughly than they search academic and industry literature, so NPL references are statistically more likely to have been missed during prosecution. The core academic sources are field-specific: IEEE Xplore and ACM Digital Library for computer science, electrical engineering, and telecommunications; PubMed and relevant clinical/biomedical databases for medical device and pharmaceutical patents; arXiv and its subject-specific mirrors for recent, fast-moving computer-science and AI research, which frequently predates formal peer-reviewed publication by many months and can be decisive for establishing an early critical date; and Google Scholar as a broad cross-disciplinary search layer that also surfaces citation counts useful for identifying influential, well-known references (which matter for obviousness arguments about what a person of ordinary skill would have known).Beyond academic literature, product-facing NPL is often the strongest evidence for consumer-technology patents specifically because it documents a product that was actually offered to the public, which supports on-sale-bar and public-use arguments (§ 102(a)(1)) in addition to printed-publication arguments: user manuals and service documentation with verifiable publication or copyright dates, trade show materials and press coverage (dated conference proceedings, trade publication archives like EE Times or IEEE Spectrum), standards-body working documents and meeting minutes (which are often publicly archived with precise dates well before the formal standard's publication), and open-source software repositories with commit-history timestamps that can establish precise implementation dates for software claims. A search that stops at academic databases and never checks product documentation, trade press, or open-source history is systematically underweighting the source category most likely to produce a decisive reference in consumer-electronics and software patent disputes.

Question 8

How do you find prior art hidden behind different terminology?

Accepted Answer

Finding prior art disguised by different terminology is fundamentally a retrieval-methodology problem, not a search-term-generation problem, and the answer increasingly involves moving beyond exact or fuzzy keyword matching toward retrieval methods that operate on technical meaning. Classification-based search remains the first and most reliable lever, since CPC/IPC assignment is concept-based rather than lexical. Citation-graph traversal (forward and backward citations, and citations of citations one or two hops out) is the second lever, since it follows relationships identified by examiners and applicants who already recognized the conceptual link regardless of vocabulary differences. The third and increasingly important lever is embedding-based semantic search: representing claim limitations and candidate reference passages as vectors in a technical-concept space and retrieving by proximity in that space rather than by literal term overlap, which is specifically designed to catch the case where two documents describe the same underlying feature using entirely non-overlapping vocabulary.A practical worked example: a claim reciting a 'ring-shaped photodiode surrounding a central light-emitting element' might be described in a different reference as an 'annular detector concentric with the emitter,' in yet another as a 'donut-configuration PD encircling the LED die,' and in a much older reference simply as a 'circular detector array positioned around the source' without the word 'ring' or 'annular' appearing at all. A keyword search for 'ring photodiode' finds none of the latter three. Classification search (both references likely share a CPC subclass for optical proximity/reflectance sensor structures) and semantic search (all four phrasings encode the same spatial relationship: a detector element positioned circumferentially around an emitter element, even though no two of them share more than one content word) both surface the full set. Because feature-level retrieval of this kind is still noisier and more expensive to run exhaustively than keyword search, the most effective current workflow uses keyword and classification search to build an initial candidate pool quickly, then applies semantic/passage-level retrieval specifically to the highest-value claim limitations to catch terminology-obscured art that the first pass would have missed.

Question 9

How do you determine whether a reference anticipates a patent claim?

Accepted Answer

Anticipation under 35 U.S.C. §102 requires that a single prior-art reference disclose every limitation of the claim, arranged as claimed. This mirrors the all-elements requirement that governs infringement analysis, applied in reverse: instead of asking whether an accused product contains every claimed element, anticipation asks whether a single earlier reference already disclosed every claimed element. Critically, disclosure can be either express (the reference states the limitation outright) or inherent (the reference necessarily, though not expressly, discloses the limitation, meaning the limitation is unavoidably present whenever the reference's express teaching is practiced, not merely probably or typically present). Inherency is a frequently misapplied doctrine: a reference does not inherently disclose a limitation merely because it would be a natural or common way to implement the reference's teaching; the limitation must be necessarily present, and courts and the PTAB apply this standard strictly.A rigorous anticipation analysis builds a full claim chart against the single candidate reference and checks, row by row, whether that reference alone closes out the limitation, marking each row as express, inherent, or absent. If even one limitation cannot be closed by that single reference, either expressly or inherently, the reference does not anticipate, and the analysis shifts to whether it can instead serve as a base reference in an obviousness combination with one or more secondary references supplying the missing limitation. A common error is 'picking and choosing' features from different embodiments within the same reference that are not disclosed as existing together (e.g., combining a feature from the reference's Figure 2 embodiment with a feature from its Figure 5 embodiment, where the reference never actually teaches those features occurring in the same device) for a single-reference anticipation argument, the disclosed elements generally must be shown arranged or combined as claimed within that one reference, not merely present somewhere within its broader disclosure.

Question 10

What is the difference between anticipation and obviousness?

Accepted Answer

Anticipation (§ 102) requires a single prior-art reference to disclose every claim limitation; obviousness (§ 103) allows the combination of two or more references (or a single reference plus the general knowledge of a person of ordinary skill in the art) to collectively cover every limitation, provided there was a reason a person of ordinary skill would have combined them to arrive at the claimed invention. Anticipation is a strictly binary, reference-by-reference test with no room for hindsight or motivation analysis; either the single reference discloses everything, or it doesn't. Obviousness is a more flexible, multi-factor legal conclusion informed by underlying factual findings (the Graham v. John Deere factors: scope and content of the prior art, differences between the prior art and the claims, the level of ordinary skill in the art, and any objective indicia of non-obviousness such as commercial success, long-felt but unsolved need, failure of others, or unexpected results).Practically, this means anticipation is the stronger and legally cleaner invalidity theory when available; a single, clean anticipatory reference is far harder for a patentee to rebut than an obviousness combination, because obviousness combinations are vulnerable to hindsight-bias challenges (the patentee's counsel will argue the combination is only 'obvious' in retrospect, after seeing the claimed invention, and that nothing in the references themselves would have prompted a real engineer to combine them) and to evidence of objective indicia that cuts in the patentee's favor. This is why an invalidity search always tries anticipation first, exhausting single-reference candidates before falling back to obviousness combinations, and why an obviousness case is built with particular attention to articulating a genuine, reference-grounded motivation a skilled person would have had to combine the references, rather than an after-the-fact rationalization built only with hindsight.

Question 11

How do you analyze a patent claim under 35 USC §102?

Accepted Answer

A § 102 (novelty/anticipation) analysis is claim-by-claim and reference-by-reference: for each asserted or challenged claim, the analyst selects a single candidate reference and builds a full limitation-by-limitation chart testing whether that one reference discloses every element, arranged substantially as claimed, either expressly (the reference states the limitation outright) or inherently (the limitation is necessarily, though not expressly, present whenever the reference's express teaching is practiced). Under the post-AIA framework (patents with an effective filing date on or after March 16, 2013), § 102(a) defines prior art broadly to include anything patented, described in a printed publication, in public use, on sale, or otherwise available to the public before the effective filing date, anywhere in the world (removing the pre-AIA geographic restrictions on public-use and on-sale art), subject to the § 102(b) grace-period exceptions for the inventor's own disclosures within one year of filing.For pre-AIA patents (effective filing date before March 16, 2013), the older first-to-invent framework applies, which introduces additional complexity: the critical date can be the invention date rather than the filing date if the inventor can establish an earlier invention date through corroborated evidence (lab notebooks, witness testimony, reduction to practice records), meaning a reference dated after the patent's filing date but before a claimed earlier invention date is not necessarily disqualified as prior art without further diligence into the actual invention timeline, a nuance modern searchers sometimes overlook when reflexively applying AIA rules to an older patent family. In either framework, the § 102 analysis should conclude with an explicit statement, per claim, of which category of prior art the reference qualifies as (patent, printed publication, public use, on sale, or otherwise available to the public) since this affects both the strength of the anticipation argument and which forum (district court versus IPR, which is limited to patents and printed publications) can consider it.

Question 12

How do you analyze a patent claim under 35 USC §103?

Accepted Answer

A § 103 (obviousness) analysis follows the Graham v. John Deere framework: (1) determine the scope and content of the prior art, (2) ascertain the differences between the prior art and the claims at issue, (3) resolve the level of ordinary skill in the pertinent art, and (4) evaluate objective indicia of non-obviousness (secondary considerations) where present. In practice this means identifying a base reference that discloses most of the claim, identifying one or more secondary references that supply the missing limitation(s), and the step most frequently done poorly articulating why a person of ordinary skill in the art, at the time of the invention, would have had reason to combine the base and secondary references to arrive at the claimed invention (the 'motivation to combine,' informed post-KSR International Co. v. Teleflex Inc. by common sense, market demand, design need, and predictable variation, not only by explicit statements in the references themselves).After the prima facie obviousness case is constructed, it must be tested against objective indicia the patent owner is likely to raise defensively: commercial success of a product embodying the claim (with a required nexus between the success and the specific claimed features, not just general product success), long-felt but unsolved need that the invention addressed, failure of others who tried and failed to solve the same problem, copying by competitors, unexpected or superior results compared to what the combination would have predicted, and industry praise or licensing activity recognizing the invention's significance. A rigorous § 103 analysis does not treat the prima facie combination as the end of the inquiry; it proactively researches whether strong secondary-considerations evidence exists for the patent in question (commercial success data, licensing history, industry commentary) and weighs it honestly, because a technically sound combination can still fail to invalidate a claim if the patentee has strong, well-documented secondary considerations evidence.

Question 13

Can multiple prior-art references be combined to invalidate a patent?

Accepted Answer

Yes, this is the essence of an obviousness (§ 103) invalidity theory, as distinct from the single-reference anticipation (§ 102) theory. There is no fixed cap on the number of references that can be combined, but as a practical matter, combinations of more than two or three references are disfavored and harder to sustain, both because each additional reference requires its own independent motivation-to-combine showing and because a fact-finder (judge, jury, or PTAB panel) becomes increasingly skeptical that a person of ordinary skill would have assembled an increasingly elaborate, multi-source combination absent the benefit of hindsight from already knowing the claimed invention. The strongest obviousness cases typically use exactly two references: a primary reference supplying the bulk of the claimed structure or method, and a secondary reference supplying one or two specific missing limitations, combined via a clearly articulable, reference-grounded reason.The motivation to combine can come from several recognized sources post-KSR: an explicit teaching, suggestion, or motivation within one of the references themselves (the older, narrower TSM test, still valid as one basis but no longer the exclusive test); the references being drawn from the same field of endeavor addressing the same problem, such that combining known elements according to known methods to yield predictable results is itself sufficient motivation; market demand or design incentive that would have prompted a skilled artisan to look for exactly the missing feature; and simple substitution of one known element for another known element performing the same function, where the result is predictable. A chart supporting a combination argument should identify, explicitly and separately from the limitation mapping itself, which of these motivation categories applies and cite the specific textual or contextual basis for it. An obviousness chart that maps limitations correctly but never explains why the combination would have been made is analytically incomplete and vulnerable to a hindsight-bias rebuttal.

Question 14

How do you build an invalidity claim chart?

Accepted Answer

An invalidity chart is a two-column, row-per-limitation document with the claim language on one side and, on the other, evidence drawn from prior art rather than from an accused product. Claim construction must happen before any reference is selected: the claim is parsed into its limitations and ambiguous terms are resolved against intrinsic evidence (the claim language, specification, and prosecution history) first, because mapping against an unconstrued claim risks either falsely rejecting a good reference reading a term more narrowly than it will actually be construed or falsely accepting a weak one by reading it more broadly than it should be. For anticipation charts, every row must be filled from the single candidate reference alone; for obviousness charts, each row should specify which reference (primary or secondary) supplies that limitation, and the chart should include a distinct section, not folded silently into the limitation rows, articulating why a person of ordinary skill in the art would have had reason to combine the references. A well-built invalidity chart also documents what it does not find: limitations the reference set only partially supports, alternative claim constructions that would change the outcome for a given row, and importantly, an assessment of whether the reference was already considered during prosecution (if the reference was cited and distinguished by the examiner, an invalidity argument based on it faces a materially higher burden, since it must show the examiner erred, whereas art the examiner never saw carries no such presumption of correctness to overcome). Best practice numbers rows to match the claim's own limitation numbering (or assigns consistent [1.1], [1.2] labels if the claim has none), cites pinpoint page/column/line or paragraph references for every quoted or paraphrased disclosure, and where the analysis will support a PTAB petition explicitly separates § 102 grounds from §103 grounds even when relying on overlapping reference sets, since these are distinct statutory grounds requiring separate treatment in a petition.

Question 15

What makes prior art strong enough for patent invalidity?

Accepted Answer

The strongest prior art satisfies several criteria simultaneously: it discloses every claim limitation (ideally as a single anticipatory reference rather than requiring a combination), its date is unambiguous and well before the critical date with clear documentary support (a printed publication date, a certified mailing or filing date, an archived and dated web capture), it was not considered by the examiner during prosecution (so there is no presumption of examiner correctness to overcome), and its disclosure is explicit rather than requiring an aggressive inherency or claim-construction argument to reach the missing limitation. A reference meeting all five criteria, often called a 'knockout reference' in practice, can resolve an entire invalidity case on its own and is the primary target of any competent invalidity search before the analyst settles for a multi-reference obviousness theory.Beyond these baseline criteria, strength is also a function of the reference's resistance to the patentee's likely counter-arguments: a reference from a highly credible, hard-to-dispute source (a peer-reviewed publication, an official standards document, the defendant's own admitted prior product) is harder to attack on authentication or interpretation grounds than an obscure or ambiguous source; a reference whose relevant disclosure is stated in unambiguous, technical (not merely suggestive) language reduces the risk that the patentee's expert can offer a plausible narrower reading; and, for obviousness references specifically, a reference explicitly identifying the problem the claimed invention solves (even if it doesn't itself solve it the same way) strengthens the motivation-to-combine argument considerably. Weak art, by contrast, requires the analyst to stack multiple assumptions: an aggressive claim construction, an inherency argument, and a speculative motivation to combine, and each additional assumption independently reduces the odds a fact-finder accepts the overall invalidity theory.

Question 16

How do you identify the strongest prior-art references?

Accepted Answer

Identifying the strongest references among a candidate pool is a triage process applied after an initial broad search has generated candidates, not a search strategy in itself. The first triage pass eliminates references that fail on date alone (post-dating the critical date once the priority chain is correctly established) or that clearly fail to disclose a dispositive limitation on close reading, even if they looked promising from an abstract or summary. The second pass ranks survivors by how many limitations each covers on its own; a reference covering all limitations moves to anticipation analysis immediately; references covering most but not all limitations become candidates for the base reference in an obviousness combination, ranked by how few and how easily justified the remaining secondary references would need to be to close the gap.A third, frequently decisive ranking factor is whether the reference was already before the examiner during prosecution: references not of record carry no presumption to overcome and are strongly preferred over references the examiner already considered and implicitly or explicitly distinguished, all else being equal. A fourth factor specific to combination searches is source diversity: a combination drawing its references from clearly independent sources (different inventors, different assignees, different countries, different publication venues) is generally a more persuasive obviousness case than a combination drawing from closely related sources (e.g., two patents from the same inventor or assignee, which can invite an argument that the combination was only apparent to that specific inventor, not to a generic person of ordinary skill). The output of this ranking process is typically a short list of two to five top candidates carried forward into full claim charts, rather than exhaustively charting the entire candidate pool.

Question 17

How do you rank prior-art references?

Accepted Answer

Ranking formalizes the reference-strength criteria coverage, disclosure quality, date certainty, and prosecution history status into a comparative framework so multiple candidates can be weighed against each other rather than assessed only in isolation. A practical ranking rubric scores each candidate reference across several axes: coverage (how many, and which, claim limitations it discloses with limitations weighted by how dispositive or difficult-to-satisfy they are, since covering nine easy limitations and missing one hard one ranks lower than covering all ten), disclosure quality (express versus inherent versus arguable), date certainty (a reference with an unambiguous, well-documented date outranks one requiring corroborating evidence to fix its timing), procedural posture (whether it's usable in the intended forum IPR requires patents/printed publications only, while district court litigation can use public-use and on-sale art), and prosecution history status (not-of-record outranks previously-considered-and-distinguished).For obviousness candidates specifically, an additional axis scores the strength of the available motivation to combine: a combination supported by an explicit cross-reference in one document to the other, or by both documents addressing the identical stated problem, ranks well above a combination whose only connection is that both happen to be in the same general technology area. The highest-ranked references (typically the top two to five) are carried into full invalidity claim charts; mid-ranked references are retained in a supporting list (useful as secondary references for obviousness combinations built around a top-ranked base reference, or as fallback options if a top candidate is later disqualified on some technical ground like a defective date); low-ranked references are documented as 'considered and rejected' in the search methodology memo rather than silently dropped, since a defensible search record should show the analyst's full universe of considered art, not just the winners.

Question 18

How do you search patents by technical concept rather than keywords?

Accepted Answer

Concept-based patent search relies primarily on classification systems: the Cooperative Patent Classification (CPC), jointly maintained by the USPTO and EPO, and its predecessor/parallel International Patent Classification (IPC), which assigns every patent to one or more hierarchical categories based on the technical subject matter of the invention, independent of the vocabulary used to describe it. The workflow is to first identify the CPC subclass(es) covering the claimed concept (starting from a known relevant patent, either the target patent itself, a close competitor patent, or an initial keyword hit and reading off its assigned CPC codes, then browsing the CPC classification tree itself, which includes definitional notes describing exactly what each subclass covers), then retrieving the full population of patents and applications within that subclass, which becomes a curated candidate pool largely immune to the terminology problem that undermines pure keyword search, since two patents describing the same underlying feature in entirely different vocabulary will still generally share a classification code.Beyond formal classification, concept-level search increasingly uses semantic and embedding-based retrieval, in which claim language or a technical concept description is converted into a vector representation and compared against a similarly embedded corpus of patents and publications, retrieving documents by conceptual proximity rather than lexical overlap. This is the feature-level retrieval approach the field is actively moving toward, and it is particularly effective because patent classification, while powerful, is coarse relative to individual claim limitations: two patents in the same CPC subclass can still differ enormously in which specific sub-features they disclose, and a semantic search operating at the level of individual claim limitations (rather than whole-document classification) can identify the exact passage within a large, only-generally-relevant document that discloses a specific narrow claim feature, which neither pure keyword search nor pure classification search reliably surfaces on its own.

Question 19

How do you find earlier versions of a patented technology?

Accepted Answer

Earlier versions of a patented technology are typically found by tracing three chains simultaneously: the patent family and priority chain of the target patent itself (checking whether it claims priority to an earlier provisional, parent, or foreign application whose own disclosure or filing date might differ from the issued patent's and checking whether continuation or divisional applications reveal how the claims narrowed or broadened over prosecution, which often points directly at what the examiner considered closest prior art); the inventor's and assignee's own historical portfolio (earlier patents, abandoned applications, and publications by the same individuals or company frequently contain earlier, less-refined versions of the same underlying idea, sometimes filed years before the eventually-issued claim and sometimes never patented at all, appearing only in an academic paper or an earlier abandoned filing); and the product history of the technology itself, where predecessor products, earlier standard revisions, or earlier industry implementations often embody an earlier, unpatented (or differently-patented) version of the same concept.For fast-moving technical fields, especially software and AI, a particularly productive source is open-source and academic preprint history: many claimed 'inventions' in these fields are refinements of published research or open-source implementations that predate the patent filing by months or years, and platforms like arXiv, GitHub commit histories, and conference proceedings archives can pin down precise disclosure dates for early versions of an algorithm or architecture. It is also worth deliberately searching for the technology's evolution narrative in industry press and technical retrospectives (product-history articles, 'how X was built' engineering blog posts, oral-history interviews with the original developers), which, while not always independently citable as formal prior art frequently name the specific earlier products, papers, or internal projects that a more targeted search can then run down and formally document.

Question 20

How do you search for prior art using patent citations?

Accepted Answer

Citation-based search exploits the fact that patents form a directed graph: each patent cites earlier references it (or the examiner) considered relevant, and is in turn cited by later patents that considered it relevant. Backward citation search pulling the references cited by the target patent and by its close family members surfaces the art already identified as relevant by the applicant and examiner during prosecution, which is a fast way to map the immediate prior-art landscape the patent itself acknowledges, though art already of record meaning the examiner already considered and, implicitly or explicitly, distinguished it carries less invalidity weight than art the examiner never saw, since there is no presumption of examiner error to overcome with a fresh reference. Forward citation search finding later patents that cite the target patent is useful less for finding prior art against the target (later patents can't be prior art to an earlier one) and more for competitive intelligence and for identifying other patents in the same family of technology that might share close prior art, since patents that cite the same target often cite overlapping prior art themselves.The more powerful technique is citation-chaining one or two hops beyond the immediate reference set: examining what the target patent's own cited references themselves cite (their backward citations), and what cites those same cited references in turn (a shared-citation or co-citation analysis). Two patents that are both cited heavily by the same set of later, unrelated patents are very likely addressing closely related technical problems even if they never cite each other directly and use entirely different vocabulary this indirect signal is often how a searcher discovers a highly relevant reference that no direct keyword or even classification search would have surfaced, because it sits one inferential step removed from the obvious candidate pool. Examiner citations specifically (as opposed to applicant-submitted citations via an IDS) are worth flagging separately in a search record, since they reflect independent examiner judgment about relevance rather than the applicant's own disclosure obligations, and a pattern of examiner citations clustering around a particular earlier inventor or company is often a useful lead for further targeted searching of that source's broader portfolio.

Prior Art & Invalidity Search