AI patent search reads an invention disclosure, turns it into concept and classification queries, and searches patent databases, journals, standards and the company's own lab notes and earlier publications by meaning rather than by keywords alone. It ranks the closest references and charts each claim element against a cited passage. A patent searcher checks the hits, and the attorney decides on novelty, claim scope and whether to file.
Trigger: An invention disclosure reaches the IP committee, or counsel needs prior art before drafting, filing or challenging a patent. Owner: Head of Intellectual Property. Primary output: Patent search report with strategy, query log and ranked references. Consequential actions require approval.
Assess your workflowKeyword strings miss references written in other vocabulary or languages, decisive disclosures hide in journals, standards and product manuals, and nobody checks the lab notes for what the company already made public.
VDF AI agents turn the disclosure into concept queries, search licensed databases and internal records together, rank references by how many claim elements they cover and draft a chart with a cited passage for each match.
Breaks the invention disclosure into features, alternatives and the problem solved, and drafts candidate claim elements for the searcher to edit.
Builds concept, synonym and classification-code queries and logs every string, database and run date.
Searches licensed patent databases, journals and standards alongside internal reports, notebooks and earlier filings.
Charts each claim element against the leading references with the paragraph, figure or claim that discloses it.
Assembles the strategy, ranked references, uncovered elements and open questions into a search report for counsel.
Each patent prior art search source has a defined purpose, freshness expectation, quality gate, and sensitivity boundary.
Purpose: Define the features and alternatives the search has to cover.
Freshness: Loaded on submission and updated after each inventor interview.
Quality: Each disclosure versioned with its submission date.
Sensitivity: Unfiled inventions; restricted to the IP team and named inventors.
Purpose: Find what the company itself has already disclosed and establish internal dates.
Freshness: Indexed nightly from the notebook and document systems.
Quality: Entry dates and authors kept exactly as the source system records them.
Sensitivity: Trade secrets; read-only access scoped by project.
Purpose: Supply patent and non-patent literature for comparison.
Freshness: Queried live at search time, with the database and date noted.
Quality: Publication and priority dates read from the record, never inferred by the model.
Sensitivity: Query strings reveal inventive concepts, so they are logged and reviewed.
Twice a year, rerun the search on recently granted or refused cases and check whether the examiner's references appear in the agent's ranked list.
It pays off for companies that file regularly in fields where vocabulary shifts quickly, such as materials, medical devices, electronics and software. Under 35 U.S.C. 102, in the first-inventor-to-file form in force since March 2013, prior art covers anything patented, described in a printed publication, in public use, on sale or otherwise available to the public before the effective filing date. Earlier-filed US patents and published applications naming other inventors count too, and the one-year grace period shields only disclosures that came from the inventors themselves.
Europe casts the net just as wide. The EPO’s 2026 Guidelines for Examination define the state of the art as everything made available to the public before the filing date, in writing, orally, by use or any other way, with no restriction on where, in what language or how long ago (verified October 2026). That is why non-patent literature matters: a conference poster or an old product manual can defeat novelty as surely as a granted patent, and a short keyword list rarely reaches either.
It is a weaker fit for an occasional search in a field nobody on the team knows, where an experienced outside searcher still earns the fee.
The USPTO’s free Patent Public Search gives public access to US patents and published applications, and the office says that from 7 November 2026 it will require users to sign in with a USPTO account (verified October 2026). Teams that lean on it should set up accounts now and check each database’s terms before an agent queries it automatically.
The search report is the evidence that matters later. A log of every query string, database and run date shows a future searcher, an examiner response or a validity challenge exactly what was covered and what was not.
VDF AI Agents run on servers you control, air-gapped if needed, so unfiled disclosures and notebook entries never reach a public model service. Hybrid search blends exact terms with semantic similarity, chunk citation pins each chart entry to its page and paragraph, and the OpenAPI invoke tool calls licensed patent databases through their documented APIs. The VDF AI Router can keep invention material on local models only.
Once counsel decides to file, the patent drafting agent picks up the disclosure and the search report. For case law and statutes instead of patents, see AI-assisted case law research and the legal research agent; engineering teams indexing their own test reports and designs can start from R&D knowledge retrieval. More workflows for counsel sit under the legal and compliance use cases.
Control: Outbound queries go only to approved databases, and the searcher approves any string describing the core concept.
Accountable owner: Patent searcher
Control: Every chart entry cites a paragraph or figure, and the searcher reads the full text of each top-ranked reference.
Accountable owner: Patent attorney
Control: Dates come from database records and are compared with the effective filing date before the report is issued.
Accountable owner: IP paralegal
New disclosures from one technology area for a quarter, searched by the agent and by the usual searcher in parallel, with both reports going to the attorney.
Assign these prebuilt tools to the bounded agents in Patent Prior Art Search, or browse all VDF AI tools.
These sources inform the governance and evaluation approach for Patent Prior Art Search. They do not certify a specific deployment.
Written by VDF AI Editorial Team. Last reviewed 29 October 2026.
Answers for In-house IP counsel, patent searchers and R&D leads evaluating this workflow's data, controls, measures, and operating boundaries.
Talk to an expertIt can run the first pass: generating queries, reading thousands of abstracts and ranking the closest references. A trained searcher still reads the full text of the top hits, decides where to dig further and signs the report, and the attorney decides what the results mean for patentability.
The exposure sits in the queries as much as in the disclosure. Keep the disclosure and notebooks on your own servers, send queries only to databases whose terms you have approved, and have the searcher sign off any string that spells out the core inventive idea before it leaves the network.
In the US, 35 U.S.C. 102 covers patents, printed publications, public use, sales and anything else available to the public before the effective filing date, plus earlier-filed US patent filings by other inventors. EPO examiners treat anything made public before the filing date as state of the art, whatever its place, language or age.
The same retrieval and charting serve an invalidity search against an asserted patent. Freedom to operate asks a different question, whether a product falls within live claims in a given country, and that conclusion belongs to counsel, so the agent only gathers and charts the candidate patents.
Start building it free in the cloud, or describe your Patent Prior Art Search workflow and we will help map the appropriate governed agent network for your environment.