My Software Invention Uses a New Recommendation Algorithm. Which Prior-Art References Should I Review Before Filing?
A worked example with a hypothetical recommender: two search tracks, a six-part feature split, CPC groups to start from, the US eligibility guidance to read alongside, and how to record the search. Sources are listed at the end.
The short answer: review US and foreign patents and published applications, papers and conference proceedings, preprints and forum posts, public code repositories, product documentation, and your own team’s earlier disclosures, and search each one feature by feature rather than by the name of your algorithm. As a general rule, 35 U.S.C. 102(a)(1) looks at what was already public before the effective filing date, whether patented, described in a printed publication, in public use, on sale or otherwise available to the public, subject to exceptions.1 That wording can reach online material such as papers, repositories or product documentation, because the MPEP treats an electronic publication, including an online database or Internet publication, as a printed publication if it was accessible to persons concerned with the art.2 Whether a particular document qualifies depends on its facts and dates.2 Ask counsel about any reference that matters.
One way to start is Patsnap Eureka Novelty Search. You describe the invention, and the Novelty Search Agent pulls out its technical features, runs multiple search strategies and ties each finding to source evidence, leaving the decision about what to file with you.11 Its results are input to your decision, not a finding on patentability or eligibility. Whatever literature the results include, run the non-patent track below yourself as well and add what you find to the same feature table.
The example is hypothetical and names no prior art. Eligibility under 35 U.S.C. 101 is a separate question, so read MPEP 2106 and the USPTO’s July 2024 eligibility update, which includes AI examples, alongside your results.8, 9
- Two tracks. The USPTO’s search steps end by broadening to foreign patents and to non-patent literature such as journals, websites and conference proceedings.4
- Split the invention first. Break it into input signals, representation, ranking or scoring method, training objective, feedback loop and serving constraints, then search each part.
- Run the search in Patsnap Eureka Novelty Search. Review or edit the extracted features, then confirm whether each is disclosed, partially disclosed or still novel.11
- Use CPC groups (the patent classification codes) with their warnings. The table below lists starting points such as G06Q 30/0631, G06F 16/9535, G06N 3/08, G06N 3/04 and G06N 20/00; the scheme says to consider the groups in each warning.5, 6, 7
- Date, record and re-run. Fix a cut-off date, record the databases, dates, keywords and documents, and re-run the search shortly before filing.4, 12 A search does not answer eligibility.8
Start from a description of your invention
Describe the invention in plain language, review the technical features the agent extracts, and read a feature-by-feature comparison against references linked to source.
- Extracted features to edit
- Source-linked findings
Why software needs two search tracks
What counts as prior art, in general
Section 102(a)(1) is the starting rule: a person is entitled to a patent unless, before the effective filing date, the claimed invention was patented, described in a printed publication, in public use, on sale, or otherwise available to the public.1 Section 102(a)(2) adds a patent issued under section 151, or an application published or deemed published under section 122(b), that names another inventor and was effectively filed before your effective filing date, subject to the exceptions in 102(b)(2).1 This guide describes the US rule only.
Documents that are not patents
The MPEP says a reference is considered publicly accessible if it was disseminated or otherwise made available to the extent that persons interested and ordinarily skilled in the subject matter, exercising reasonable diligence, can locate it, and notes that public accessibility has been called the touchstone for a printed publication.2 It adds that an electronic publication, including an online database or Internet publication such as a discussion group, forum, digital video or social media post, is a printed publication under 102(a)(1) if it was accessible to persons concerned with the art to which it relates.2 That is the general rule behind treating papers, conference proceedings, public code repositories and product documentation as places to look. The same section quotes the Federal Circuit calling the determination a case-by-case inquiry into the facts and circumstances of the disclosure, so a document’s format settles nothing by itself.2
Why patents alone may not be enough
The MPEP tells examiners that in very active, high-technology arts, patent documents may seriously lag behind invention and so be a reference source of limited value, and that examiners in those arts must take special care to include non-patent literature in their searches.3 If your field moves quickly, plan your search the same way. The USPTO’s multi-step search strategy ends with the same step: broaden the search with foreign patents, non-patent literature and/or a patent professional’s search, and search books, journals, websites, technical catalogs and conference proceedings in the applicable field.4
Split the hypothetical invention into features
Imagine a disclosure for a recommender in an online store. The items are hypothetical products. The recommender learns representations of users and items from interaction data, scores and ranks candidate products with a learned model, updates itself from the responses to its own recommendations, and returns results inside a response-time budget. This invention is made up for illustration. It describes no real patent, paper, product or company.
Search by feature, not by the word “recommendation”. The MPEP ties the analogous-art question to what the claimed subject matter essentially does, not to the name the applicant gives it, and says analogous arts must be searched regardless of where the claimed invention is classified.3 The same function can carry a different name in a paper, a repository and a patent, so give each feature its own vocabulary list.
| Feature | In the hypothetical | Vocabulary to try |
|---|---|---|
| Input signals | Clicks, views, purchases and dwell time, plus product attributes and context such as time of day or device | implicit feedback, interaction data, browsing history, session data, context features |
| Representation | User and item vectors learned from the interaction data | embedding, latent factor, latent vector, representation learning, collaborative filtering, matrix factorization |
| Ranking or scoring method | A model that scores candidate products for a user and orders them | candidate generation, ranking, re-ranking, scoring function, similarity measure, diversity |
| Training objective | A loss that weights some interactions more than others and handles products a user never saw | loss function, objective, weighting, negative sampling, regularization |
| Feedback loop | Logged responses to recommendations are used to update the model | online learning, incremental update, retraining, exploration, logged feedback |
| Serving constraints | Results returned within a response-time and memory budget | latency, caching, approximate nearest neighbor search, index, quantization, compression |
The vocabulary column is a starting list for your own brainstorming, not a list of the right terms for any real invention.
Collaborative filtering and matrix factorization are useful search words for the representation and scoring rows; they say nothing about what any reference discloses. The USPTO’s strategy begins by brainstorming alternative words for your concept and says to start with a broad keyword search and narrow it by adding features.4 Mark the one or two features you believe are new, and search each feature alone and in combination. Keep references that disclose only some features, and ask counsel whether they matter for obviousness under 35 U.S.C. 103.
Build the queries
In Patent Public Search’s Advanced Search, ADJn finds two terms within n terms of each other, in order and in the same sentence, and NEARn does the same in either order; WITHn and SAMEn widen the range to n sentences and n paragraphs.13 The USPTO’s efficient-searching guidance advises limited truncation with a small numeric modifier, such as comput$3 for compute and computing, over an open wildcard.14
Patsnap Eureka Novelty Search starts the same way. The agent picks out claim-like features, their dependencies and the inventive concepts, then maps them to patent claim dimensions for structured searching, and lets you review or edit the extracted features so the search matches your invention.11
CPC groups to search for the hypothetical recommender
The USPTO’s search strategy says to expand the keyword search with relevant CPC classifications, selecting the CPC system on the Classification Resources page and using a classification search in Patent Public Search’s Advanced Search with the .CPC. field code.4 The table lists groups to start from. Titles and warnings are as shown on the USPTO CPC scheme pages, scheme version 2026.08.5, 6, 7 Which groups fit your real invention is your judgment, and the “possible fit” column is a search-planning suggestion, not a statement of how any application would be classified.
| Group | Title on the USPTO scheme | Possible fit | Also consider, per the scheme |
|---|---|---|---|
| G06Q 30/0631 | Recommending goods or services (under G06Q 30/0601, Electronic shopping [e-shopping])5 | The recommendation use case and input signals, since the hypothetical items are products in an online store | Its warning names G06Q 30/06311 (based on purchase or browsing history), 30/06312 (based on similarity with other shoppers), 30/06313 (based on similarity of goods or services, e.g. substitute or alternate goods or services), 30/06314 (by determining complementary or compatible goods or services, e.g. accessory goods), 30/06315 (gift recommendations, e.g. for a specific gift recipient)5 |
| G06Q 30/0251 | Targeted advertisements, with subgroups G06Q 30/0255 (based on user history) and G06Q 30/0269 (based on user profile or attribute)5 | Worth a pass if the store also uses the recommender to choose which advertisements to show | No reclassification warning shown for these groups on the scheme page5 |
| G06F 16/9535 | Search customisation based on user profiles and personalisation (under G06F 16/953, Querying, e.g. by the use of web search engines, within G06F 16/95, Retrieval from the web)6 | Personalised ordering of results from a user profile | G06F 16/9536 (Search customisation based on social or collaborative filtering) and G06F 16/9538 (Presentation of query results; its own warning also names G06F 16/951, Indexing; Web crawling techniques)6 |
| G06F 16/335, 16/435, 16/635, 16/735 | Each titled Filtering based on additional data, e.g. user or group profiles, under Querying in G06F 16/30 (of unstructured textual data), 16/40 (of multimedia data), 16/60 (of audio data) and 16/70 (of video data) respectively6 | Profile-based filtering where the items are text, multimedia, audio or video; the scheme cross-references G06F 16/335 to G06F 16/9535 and 16/9536 for filtering in a web context | The parent Querying groups carry warnings: G06F 16/43 names G06F 16/432, 16/48, 16/483, 16/487 and 16/489; G06F 16/63 and 16/60 should be considered together; G06F 16/73 names G06F 16/7326 |
| G06F 16/24578 | Using ranking, under G06F 16/2457 (with adaptation to user needs), within G06F 16/24 (Querying) of structured data, e.g. relational data (G06F 16/20)6 | The ranking step, if the hypothetical ranks results of queries over structured data | No reclassification warning shown for this group on the scheme page6 |
| G06N 3/08 | Learning methods (under G06N 3/02, Neural networks)7 | Training objective and the way the model is trained or updated | G06N 3/0895 (Weakly supervised learning, e.g. semi-supervised or self-supervised learning), 3/09 (Supervised learning), 3/091 (Active learning), 3/092 (Reinforcement learning), 3/094 (Adversarial learning), 3/096 (Transfer learning), 3/098 (Distributed learning, e.g. federated learning), 3/0985 (Hyperparameter optimisation; Meta-learning; Learning-to-learn)7 |
| G06N 3/04 | Architecture, e.g. interconnection topology (under G06N 3/02, Neural networks)7 | The structure of a network that produces the representations, and possibly serving constraints | G06N 3/0464 (Convolutional networks [CNN, ConvNet]), 3/0475 (Generative networks; its own warning also names G06N 3/047, Probabilistic or stochastic networks), 3/0495 (Quantised networks; Sparse networks; Compressed networks), 3/0499 (Feedforward networks)7 |
| G06N 20/00 | Machine learning7 | Learning approaches that are not neural networks | No reclassification warning shown for this group on the scheme page. Its listed subgroups include G06N 20/10 (using kernel methods, e.g. support vector machines [SVM]) and G06N 20/20 (Ensemble learning)7 |
For each group with a warning, the scheme says all the groups listed in the warning should be considered in order to perform a complete search.5, 6, 7 Some of those groups carry warnings of their own, noted in the table. Check the live scheme when you search and note the scheme version in your record. If your recommender does not recommend goods or services, G06Q 30/0631 may not fit; look up the field you actually recommend in, and keep the groups for the learning method.
Include foreign patent documents
Run the patent track on foreign patent documents as well as US ones, and record them. The USPTO’s strategy says to broaden the search with foreign patents and names Espacenet, the European Patent Office’s worldwide patent publication database.4
In Patsnap Eureka Novelty Search, the third step expands scope with IPC and CPC codes alongside semantic, keyword, and family and citation strategies, and you can adjust terms or classifications with your domain knowledge.11 Bring the groups above to that checkpoint.
Why run a classification pass
The MPEP tells examiners that text search can work well when terminology is settled, yet a text search alone is rarely a thorough search of patent documents; some mix of text and other criteria, such as classification, would be a normal expectation in most technologies.3 It also says the field of search normally includes the classification locations where the claimed subject matter would be properly classified, and that the examiner should note every CPC group or subgroup that may hold pertinent material and extend the search to all probable areas.3
For software, consider a classification pass next to the keyword pass: a feature is named by whoever wrote the document, so a second route to the same documents that does not depend on wording is worth having. The USPTO’s steps pair the two, then review the references cited by the documents you found and the documents that cite them, which Patent Public Search can search in both directions.4 Use classes for the patent track. For the non-patent track, rely on the same feature vocabulary rather than CPC groups.
The non-patent track: what to review and what to keep
Where to look
- Papers and conference proceedings. The USPTO names journals and conference proceedings as places to search in the applicable field.4 Search each of the six features, then follow the references a good hit cites.
- Preprints, forums and posts. The MPEP lists discussion groups, forums, digital video and social media posts among Internet publications that can be printed publications if accessible to persons concerned with the art.2
- Open-source repositories. Read the README, documentation, release notes and history, not only the code. Whether a repository was publicly accessible on a given date is a legal conclusion that turns on the facts, so keep the evidence.2
- Product documentation and technical pages. The USPTO names websites and technical catalogs as sources.4 Capture the version you read, since pages change.
Capture the date and the copy
The MPEP says an electronic publication is treated as publicly available as of the date it was publicly posted, and that if a publication carries no publication date or retrieval date and there is no other evidence of the posting date, it cannot be relied on as prior art under 102(a)(1), though it may be relied on as evidence of the state of the art.2 It also says Office copies of electronic documents must be retained when the document may not be available later, especially for sources from the Internet and online databases.2 Do the same: save the page or file, the URL, the version, the date shown, how you found out when it went public, and the date you retrieved it.
On dating, the MPEP cites a Federal Circuit decision that relied on the Wayback Machine to establish a date of publication, and says publications obtained through it are prima facie deemed publicly accessible at the date and time in the time stamp.2 For social media and public forum posts, it says accessibility is decided case by case, considering where the information is posted, privacy restrictions on the posting, the length of time it was posted, and whether it is indexed for searching, and that timestamps on those sites may need careful scrutiny.2 Note those four points for any post you rely on.
Include your own team’s material
Under section 102(b)(1), a disclosure made within the year before the effective filing date does not count as prior art under 102(a)(1) if either of these holds:1
- the inventor, a joint inventor, or another person who obtained the subject matter directly or indirectly from them made the disclosure; or
- the subject matter had, before that disclosure, been publicly disclosed by the inventor, a joint inventor, or another who obtained it directly or indirectly from them.1
Both the one-year window and one of those conditions must be met. List your own papers, talks and repositories with their dates, and ask counsel whether the exception applies before relying on it. If you may file outside the US, ask counsel before relying on the 102(b)(1) exception, since other systems treat earlier disclosures differently.
Search without publishing your disclosure
The USPTO’s rule for its own examiners is that Internet searches relating to unpublished applications must be limited to the general state of the art and formulated in a way that protects the confidential proprietary intellectual property.3 Take the same care: use general technical terms in public search boxes, and keep the full disclosure out of them.
Turn your six-part feature split into a source-linked comparison
Edit the features the agent extracts, tune the terms and classifications, and mark each feature as disclosed, partially disclosed or still novel.
- Feature-by-feature table
- Edit terms and classes
- 14-day free trial
Running the search in Patsnap Eureka Novelty Search
The Novelty Search Agent starts from your description. It extracts the technical features, runs multiple search strategies and maps findings to source evidence, so that you can decide what to file.11
Five steps, with the hypothetical
- Submit an invention disclosure. Write the technical problem, the solution, its advantages and the implementation details in everyday language, or upload a PDF, image or Office document. The agent takes 200 to 500 words and extracts figures automatically.11 For the hypothetical, the six-part split from section 2 is a ready outline.
- Extract core technical features. The agent picks out claim-like features, their dependencies and the inventive concepts, then maps them to patent claim dimensions for structured searching. At the checkpoint you review or edit the features so the search matches your invention.11 Check that each of the six parts is present and that the feature you believe is new is marked.
- Build and run search strategies. Keywords, synonyms, semantic concepts, patent classifications and jurisdictions are expanded. Semantic search matches by meaning, four keyword searches cover synonyms, hypernyms, hyponyms and AI field search, classification and Boolean search expands scope via IPC and CPC codes, and family and citation tracing follows the citation chains of key prior art. At the checkpoint you adjust terms or classifications.11
- Compare against relevant references. The agent ranks the prior art and lines each reference up against your invention’s features in a comparison table, and every cell links back to the source patent text. At the checkpoint you confirm whether each feature is disclosed, partially disclosed or still novel.11
- Generate a novelty assessment report. The report covers the closest prior art, novelty risks, evidence gaps and patentability considerations, which your team can use when deciding whether to file, refine or stop.11
What stays with you
Patsnap Eureka Novelty Search is not a complete replacement for human IP professionals. The agent takes on the high-volume searching and analysis, and your IP experts audit what it returns and make the final strategic decision.11 Its results are input to your decision, not a finding on patentability or eligibility. Run the non-patent track from section 4 yourself as well, and add each paper, repository or document you find to your own record.
The free trial lasts 14 days and includes 10,000 trial credits.11
Read the US eligibility guidance alongside the search
MPEP 2106
MPEP 2106 states two criteria. The claimed invention must be to one of the four statutory categories, and it must qualify as patent-eligible subject matter, meaning the claim must not be directed to a judicial exception unless the claim as a whole includes additional limitations amounting to significantly more than the exception. The judicial exceptions are limited to abstract ideas, laws of nature and natural phenomena.8 The same section adds that involving a judicial exception does not by itself make an invention ineligible, and that a complete examination should be made under the other patentability requirements, including 102 and 103, whether or not a rejection under 101 is made.8
The July 2024 guidance update
The USPTO published “2024 Guidance Update on Patent Subject Matter Eligibility, Including on Artificial Intelligence” in the Federal Register on July 17, 2024 (89 Fed. Reg. 58128), effective that day.9 The notice says it is meant to help USPTO personnel and stakeholders evaluate the eligibility of claims involving AI inventions, and that it is to be used together with the MPEP.9 According to the notice, the existing eligibility guidance in MPEP 2103 to 2106.07(c) is used for all technologies, including AI inventions, which it describes as generally considered computer-implemented inventions.9
The notice describes Step 2A as a two-pronged inquiry: whether a claim recites an abstract idea or other judicial exception (Prong One), and, if so, whether the claim as a whole integrates it into a practical application (Prong Two).9 It identifies two areas of particular stakeholder concern for AI inventions, the evaluation of whether a claim recites an abstract idea (mathematical concepts, certain methods of organizing human activity and mental processes) and the evaluation of improvements at Prong Two, and it states that an improvement in the judicial exception itself is not an improvement in the technology.9 It also announces Examples 47 to 49: hypothetical claims on an artificial neural network used to identify or detect anomalies, on analyzing and separating speech signals, and on an AI model that assists in personalizing medical treatment.9 Read the examples on the USPTO site rather than relying on a summary.
The notice says it is not intended to announce any new USPTO practice or procedure, and that it does not constitute substantive rulemaking or have the force and effect of law. It also says that where earlier USPTO guidance, including any section of the current MPEP, is inconsistent with it, USPTO personnel are to follow this guidance.9 This guide does not predict how any claim would be treated.
Later USPTO materials
The USPTO’s eligibility page lists the 2024 update as effective July 17, 2024, and says current eligibility guidance is found in MPEP 2103 through 2106.07.10 It also lists later items, including:
- “Reminders on evaluating subject matter eligibility of claims under 35 U.S.C. 101” (August 4, 2025)10
- “Advance notice of change to the MPEP in light of Ex Parte Desjardins” (December 5, 2025)10
- a memorandum on “Best Practices for Submission of Rule 132 Subject Matter Eligibility Declarations (SMEDs)” (April 30, 2026)10
- a memorandum on “Subject Matter Eligibility Declarations (SMEDs) under 37 CFR 1.132” with a flowchart and best practices (September 29, 2026)10
Check that page for what applies when you file.
How this fits with the search
The search and the eligibility review start from the same place: what the disclosure actually describes. Writing the six-part feature split in concrete terms helps both conversations with counsel. Do not read a search result, with or without close references, as an answer on eligibility.
Record the search
The USPTO says it is important to record the details of your search process, such as the databases you used, the date and time of your search, the keywords you searched, and which patents and published patent applications were retrieved.4 For a software invention, extend that record.
- Cut-off date. Before searching, fix your cut-off date (your planned or claimed effective filing date) and note it in the record. Keep US patents and published applications whose effective filing date is earlier even if they were published after it, since section 102(a)(2) counts them.1
- Patent track. The CPC groups searched, including those named in each scheme warning, the scheme version, and the foreign patent documents covered.4, 5, 6, 7 Consider keeping the Patsnap Eureka Novelty Search report and its source-linked comparison table in this record.
- Non-patent track. For each paper, repository or product page: URL, version, the date shown and how its public date was established, the date you retrieved it, and a saved copy.
- Feature table. One row per reference, one column per feature of the six-part split, with the passage that discloses it, or a note that it is absent. Mark each feature as disclosed, partially disclosed or still novel.
- Decisions. The terms and groups you tried, what you excluded and why, and what you plan to do next: file, refine, stop, or commission a professional search.
- Re-run. Re-run the search shortly before filing, since applications filed earlier may have published in the meantime.12
The USPTO suggests you may want to hire a registered patent attorney or patent agent to review the search.4 A complete record lets that person start from your work.
Questions a search leaves open
- Documents you cannot see yet. Under 102(a)(2), a patent or an application published or deemed published under 122(b) that names another inventor and was effectively filed earlier counts as prior art, subject to the 102(b)(2) exceptions.1 Under 35 U.S.C. 122, the USPTO keeps applications in confidence and, subject to exceptions, publishes each one promptly after 18 months from the earliest filing date for which a benefit is sought, so a public search will not show an earlier-filed application that is still unpublished.12
- Public accessibility and dates. Whether and when a paper, repository or web page became publicly accessible is decided on its facts.2
- Eligibility. It is a separate requirement under 35 U.S.C. 101, analyzed under MPEP 2106 and the guidance described above.8, 9
- The legal conclusion. With Patsnap Eureka Novelty Search, auditing the results and making the final strategic decision stay with your IP experts.11
- Other jurisdictions. This guide covers the US only.
Frequently asked questions
My software invention uses a new recommendation algorithm. Which prior-art references should I review before filing?
Review US and foreign patents and published applications, and non-patent literature such as papers, conference proceedings, preprints, forum posts, public code and product documentation, plus your own team’s earlier disclosures, searching feature by feature: input signals, representation, ranking or scoring method, training objective, feedback loop and serving constraints. Under 102(a)(1), a person is entitled to a patent unless, before the effective filing date, the claimed invention was patented, described in a printed publication, in public use, on sale, or otherwise available to the public, subject to the exceptions in 102(b)(1), and the USPTO treats accessible electronic publications as printed publications.1, 2 Patsnap Eureka Novelty Search can start a search from a plain-language description, and you should run the non-patent track yourself as well; take close references to a patent professional.11, 4 Keep references that disclose only some features, and ask counsel whether they matter for obviousness under 35 U.S.C. 103.
Can open-source code or product documentation be prior art?
As a general rule, yes, if it was a printed publication or otherwise available to the public before the effective filing date, subject to the exceptions in 102(b)(1). The MPEP says an electronic publication is a printed publication if it was accessible to persons concerned with the art, and that this is a case-by-case inquiry into the facts, so dates and accessibility need evidence.1, 2 Ask counsel about any specific document.
Which CPC groups should I search for a recommendation algorithm?
Starting points include the groups in the table above, such as G06Q 30/0631 (Recommending goods or services), G06F 16/9535 (Search customisation based on user profiles and personalisation), G06N 3/08 (Learning methods), G06N 3/04 (Architecture, e.g. interconnection topology) and G06N 20/00 (Machine learning), together with the groups in each reclassification warning on the scheme.5, 6, 7 Confirm each group on the current USPTO scheme and note the version you used.
What does the USPTO’s July 2024 eligibility update cover?
It is a Federal Register notice, effective July 17, 2024, that addresses the eligibility of AI inventions, explains Step 2A, discusses abstract ideas and improvements, and announces Examples 47 to 49.9 It says it is not intended to announce any new USPTO practice or procedure.9 The USPTO’s eligibility page lists later materials, so check it, and ask counsel how the guidance applies to your claims.10
How does Patsnap Eureka Novelty Search fit into this search?
It starts from your description: it extracts technical features for you to review or edit, searches with semantic, keyword, classification and citation strategies, lines each reference up against your features in a comparison table whose cells link to the source patent text, and produces a novelty assessment report.11 You stay in control at three checkpoints.11 Run the non-patent track yourself as well.
Does a Novelty Search report show my invention is patentable or eligible?
How should I record a prior art search?
The USPTO advises recording the databases used, the date and time of the search, the keywords searched and the documents retrieved.4 Fix a cut-off date first, and for web material also keep the URL, version, date shown, retrieval date and a saved copy. Re-run the search shortly before filing, since applications filed earlier may have published in the meantime.12
Sources and verification
All pages accessed October 22, 2026.
- 35 U.S.C. 102, Conditions for patentability; novelty, subsections (a)(1), (a)(2), (b)(1) and (b)(2). law.cornell.edu/uscode/text/35/102
- USPTO, Manual of Patent Examining Procedure, section 2128, “Printed Publications” as Prior Art [R-01.2024], subsections I and II (electronic publications, date of availability, Wayback Machine, social media). uspto.gov/web/offices/pac/mpep/s2128.html
- USPTO, Manual of Patent Examining Procedure, section 904, How to Search [R-07.2022], including 904.01(c), 904.02, 904.02(a) and 904.02(c). uspto.gov/web/offices/pac/mpep/s904.html
- USPTO, Multi-Step U.S. Patent Search Strategy (page last updated May 7, 2024). uspto.gov/patents/search/patent-search-strategy
- USPTO, CPC Scheme, subclass G06Q, version 2026.08, groups G06Q 30/0631, G06Q 30/0251, G06Q 30/0255 and G06Q 30/0269 and the groups named in the G06Q 30/0631 warning. uspto.gov/web/patents/classification/cpc/html/cpc-G06Q.html
- USPTO, CPC Scheme, subclass G06F, version 2026.08, groups G06F 16/9535, G06F 16/335, G06F 16/435, G06F 16/635, G06F 16/735 and G06F 16/24578, their parent groups and the groups named in their warnings. uspto.gov/web/patents/classification/cpc/html/cpc-G06F.html
- USPTO, CPC Scheme, subclass G06N, version 2026.08, groups G06N 3/04, G06N 3/08 and G06N 20/00 and the groups named in their warnings. uspto.gov/web/patents/classification/cpc/html/cpc-G06N.html
- USPTO, Manual of Patent Examining Procedure, section 2106, Patent Subject Matter Eligibility (revision marker R-10.2019 on the page), subsections I and III. uspto.gov/web/offices/pac/mpep/s2106.html
- USPTO, 2024 Guidance Update on Patent Subject Matter Eligibility, Including on Artificial Intelligence, 89 Fed. Reg. 58128 (July 17, 2024). govinfo.gov/content/pkg/FR-2024-07-17/html/2024-15377.htm
- USPTO, Subject matter eligibility (page last updated September 29, 2026). uspto.gov/patents/laws/examination-policy/subject-matter-eligibility
- Patsnap Eureka, Novelty Search: AI novelty search and prior art analysis, including workflow, FAQ and plans. eureka.patsnap.com/novelty-search
- 35 U.S.C. 122, Confidential status of applications; publication of patent applications, subsections (a), (b)(1) and (b)(2). law.cornell.edu/uscode/text/35/122
- USPTO, Patent Public Search: Operators (page last updated February 25, 2025). uspto.gov/patents/search/patent-public-search/operators
- USPTO, Patent Public Search: Efficient searching (page last updated September 29, 2025). uspto.gov/patents/search/patent-public-search/efficient-searching
Who published this. This article is published by Patsnap, which develops and sells Patsnap Eureka, the product described above. It is an editorial guide, not an independent review, and it has not been commissioned, sponsored, reviewed or endorsed by any other organization named here.
How the information was gathered. Statements about the law reflect the statutes, MPEP sections, Federal Register notice and USPTO pages cited, as accessed on October 22, 2026. CPC group titles and warnings are as shown on the USPTO CPC scheme pages, version 2026.08. Statements about Patsnap Eureka are based on the Patsnap Eureka Novelty Search product page. Tools and guidance change, so confirm anything material with the source.
Scope and limitations. This guide describes a US prior art search and the US eligibility guidance for a hypothetical software invention. The example is invented, and no real patent, paper, product or company is cited as prior art. It does not cover other jurisdictions or give an opinion on any invention.
Trademarks. All product and company names are the property of their respective owners and are used for identification only. Their use does not imply affiliation, sponsorship or endorsement.
Not legal advice. This article is general information. It does not create an attorney-client relationship and should not be relied on in place of advice from qualified patent counsel on a specific invention.
Begin with the features you have already split out
Describe your invention, edit the extracted features, and read each comparison against the source patent text yourself.
- 14-day free trial
- 10,000 trial credits