Using ChatGPT to Audit Documentation for Style Compliance

In my previous article, Using ChatGPT to Audit Documentation for Missing Definitions, I acknowledged the magnitude of the challenge Documentation Teams face. Offloading audit activities to LLMs empowers these teams by returning time and simplifying the complexity of content creation and maintenance. For instance, consider using ChatGPT for style compliance audits as demonstrated below. Notice the explicit "Steps for Audit" detailing expectations for the model, this directly impacts the quality of the audit output. The first steps towards personalizing this approach for your need are replacing the Microsoft Manual of Style with references to your own modifying the "Steps for Audit" to meet your needs.

You are a technical writer 
You review documentation for compliance with Microsoft Manual of Style.
To begin, learn the Microsoft Manual of Style.
Then you will summarize the key points.
Once this is done, prompt the user for documentation to audit for style
 guide compliance.
Perform the audit following the steps below.

Steps for Audit:
1. Analyze each page in the documentation for style compliance.
2. For each page provide a summary that includes:
   - extract of compliance failure 
   - suggestion to be compliant
   - rewrite of failure per suggestion
   - example of where demonstrated well
3. Repeat until each page analyzed, prompt the user to continue.        

The Microsoft Manual of Style is summarized by ChatGPT for reference below. To further fine-tune this supply an annotated modified summary to the model to emphasize or deprioritize areas of focus.

For this example I audited the Wikipedia Page on Apache Iceberg to provide ample opportunity to find deviation from compliance. The "Steps to Audit" portion of prompt drives the focus and presentation of the output.

Introducing an LLM such as ChatGPT into Documentation Audits saves time and increases breadth of analysis. Consider scaling the review process with an LLM that serves as copy-editor for writing from external contributors. In an enterprise environment customized GPTs can learn from each other while maintaining a hyper-focus on a specific tasks. Good luck exploring my fellow pioneers, and follow for more educational content on creating educational content.

Thanks to my editor Gabriel Dash.