How to Use AI for Advanced Patent Search
Advanced patent search uses structured fields, filters, and counting rules to answer a defined research question. AI can help interpret the query, organize returned records, explain technical differences, and produce a source-linked report—provided the search logic and evidence boundary remain visible. This guide shows how the Run Advanced Patent Query Skill applies that approach to a focused battery-cooling example.
The report excerpt below comes from a real Skill run on a narrow CATL liquid-cooling plate query. It shows the complete result returned by that query and the verified relationship between the two publications.
From a research question to a family-level insight
Focus the search on the technology in the title and the applicant identity.
TTL:("liquid cooling plate") AND AN:(CATL OR "CONTEMPORARY AMPEREX")The Skill preserves the exact syntax so the search can be reproduced.
Two publications were returned by the focused query.
Both records describe the same multi-channel water-cooling plate approach. The useful result is a cross-jurisdiction family path—not two independent inventions.
Use this result to
Review the shared mechanism, family members, and controlling claims.
Do not use it to claim
Complete portfolio coverage, FTO, validity, or technical superiority.
Run the Skill on the applicant, technology, and scope you need to review.
What advanced patent search means
A general keyword search asks whether a few words appear in a document. An advanced patent search defines where those words must appear, which applicants or dates matter, how records are counted, and what evidence must be returned. A professional query may combine title, abstract, claims, applicant, publication date, jurisdiction, classification, or patent-number fields.
The result is not automatically an answer. It is a controlled patent set that can be reviewed, reconciled across families, summarized by applicant or technology, and traced back to individual publications.
Where AI helps—and where it does not
AI is useful for translating a research question into searchable concepts, explaining fielded syntax, normalizing record fields, summarizing technical approaches, and turning a result set into a readable report. It can also flag missing dates, inconsistent applicant names, or the risk of treating publications as unique inventions.
AI should not silently rewrite an approved query or infer that every hit is relevant. Search scope, exclusions, family treatment, current status, and claim interpretation still require explicit rules and, where the decision is legal or high-stakes, qualified review.
How the Skill turns a query into a report
The Skill keeps the user-supplied query unchanged unless a correction is proposed and approved. It then turns retrieval into a traceable reporting sequence:
Validate the query, report title, dates, jurisdictions, cap, and counting unit.
Record the total, returned count, sort order, cap, and identifiers needed for verification.
Separate publication, application, priority, applicant, assignee, status, and technical fields.
Create the overview, company summaries, grouped patent details, Markdown, and accessible HTML.
This is more than asking a chatbot to list patents. The exact search logic, returned evidence, interpretation, and gaps remain separate and inspectable.
Prepare, install, and run
Start with a valid Patsnap professional query. Also define the report title, maximum record count, date interpretation, jurisdictions, counting unit, report language, and whether literature context or AI synthesis is required.
Run this Patsnap query: TTL:("liquid cooling plate") AND AN:(CATL OR "CONTEMPORARY AMPEREX"). Create an English patent report titled “CATL liquid-cooling plate patent report.” Use a maximum of 50 records, preserve the exact query, record the counting rule and evidence cutoff, verify family relationships for the returned publications, and separate retrieved facts from AI interpretation.How to read and verify the result
Read the overview first: confirm that the exact query, date field, cap, sort order, and counting unit match the question you intended to ask. Then check whether the company summary is supported by enough records and whether each technical statement links to a retrieved publication.
In this Sample, two publications do not mean two inventions. Family verification shows that they represent the same underlying filing family. The useful conclusion is a cross-jurisdiction publication path and a shared cooling-plate mechanism—not a two-patent portfolio trend.
Review next: inspect the controlling claims, current legal status, relevant family members, and any important records excluded by the query before using the report for a decision.
When you need a full patent landscape instead
Use this Skill when you already have a fielded query and need a bounded report from its returned records. A full patent landscape is a broader project: it normally requires iterative search expansion, documented exclusions, comprehensive family review, taxonomy development, tagging, trend analysis, and quality checkpoints across a larger corpus.
The separate landscape workflow can combine Search, Analyze, Tag, and Report stages. Those stage Skills are not mandatory follow-on steps for every advanced query.
Refine or extend your patent search
Connect your agent to structured patent-search tools when the next question requires a broader query, current records, or deeper family and claim review.
Frequently asked questions
Do I need to know Patsnap query syntax before using the Skill?
You need a meaningful fielded query or a clearly defined query to run. AI can explain or propose corrections, but the Skill should show any change instead of silently rewriting the search.
Can AI decide whether every returned patent is relevant?
AI can support screening and summarize evidence, but relevance depends on the research question and documented inclusion rules. Important records and exclusions should remain reviewable.
Does the report automatically collapse patent families?
The report records the selected counting unit and can use family information when available. It must not treat a publication count as a family count without verification.
Is this the same as a patent landscape?
No. This Skill executes a defined advanced query and reports its returned records. A landscape usually needs iterative retrieval, broader quality controls, taxonomy work, and corpus-level analysis.
