Prompt Loops (used within Flow AI)

Prompt Loops apply a user instruction to each section of a document, and the results are combined to provide a comprehensive answer.

You can split the document into paragraphs, sections, or chapters. What each of these mean may be different per document, because documents can be formatted in many ways. When we do a Deep Dive, we parse out the structure of a document and infer what the chapters and sections are. This isn't perfect, but many documents with clearly demarcated sections that fit within this framework.

After you choose the split type, we iterate through each paragraph, section, or chapter (whatever is selected), and then apply the user instruction to the snippet of text. We then combine the result into one file and display that to you for review. The smaller the split type, the longer the final result will be.

Optionally, you can choose Focus Areas, where the prompt is only applied to the selected Focus Areas. This can help save Prompt Loop segments.

Here are some example Prompt Loops for inspiration:

• Provide all examples of "Topic X" in this transcript
• I'm writing a blog post on "Topic X". Help me extract key insights that I can add to the post.
• Extract all mentions of "Topic X", and tell me how it fits within "Topic Y"
• Generate questions to test understanding
• Summarize using these rules: "Rule X..."
• List all final invoice amounts