How to Draft a Software Patent: Explain the Technical Processing

Start by explaining how the computer solves a particular problem. Renaming a product feature list as a collection of modules does not explain the implementation. Data relationships, operating conditions and supporting evidence matter.

Prepare four kinds of information

Item	Useful detail
Input	Source, structure, acquisition conditions and preprocessing
Processing	Sequence, decisions and the role of the algorithm or model
Output	Data, control information or a device action
Effect	The effect connected to the stated mechanism and its validation

For an AI-related invention, separate training from use. Describe the model structure, data characteristics and operations on which the solution actually depends. Do not apply a universal parameter list to every invention.

A synthetic inspection example

Suppose a production-line camera supplies images. The system estimates lighting, adjusts the image, sends candidate regions to a detection model and produces defect locations for machine control. This is an explanatory example, not a searched or patentability-assessed invention.

Ask where the lighting estimate comes from, how adjustment changes the image, how candidate regions are selected and how the output becomes control information. Unknown accuracy or latency should remain an open question, not a fabricated measurement.

Choose claim categories from the implementation

Method, system, device and storage-medium claims can raise different drafting and examination issues. They are not a mandatory four-part package for every software invention. Identify the actor performing each step, involved components and supporting disclosure before deciding on the claim strategy.

For China, Patent Law Articles 2, 25 and 26 provide primary references concerning technical solutions, exclusions, disclosure and support. Adding the word “processor” does not by itself resolve every eligibility or disclosure issue. Other jurisdictions require their own analysis.

Keep the draft connected to engineering facts

For each operation, record input, output, trigger, exceptional conditions and source passage. Match the flowchart to these operations and use consistent data-object names. AI may help organize the record; it must not supply invented schemas, training results or performance figures.

Must a disclosure include source code?

Explain the technical solution. Whether code extracts, pseudocode or other material are needed depends on the invention and applicable requirements. Neither “we have code” nor “code is never needed” is an adequate universal rule.

Prepare the disclosure template, explore drawing choices and then use the appropriate drafting workflow.