Skip to main content

Understanding AI Features in Delve

K
Written by Kate G

Overview

Delve’s AI features are designed to assist researchers in analyzing qualitative data faster and more efficiently. AI in Delve acts like a research assistant, helping you brainstorm ideas, generate codes and subcodes, summarize transcripts, and uncover patterns within your data. However, it is important to remember that AI tools do not replace your expertise - they assist, but you still need to critically evaluate results and understand your research context.

Note: All AI features require joining Delve’s AI beta program. You must accept the Beta Terms to access AI Chat, Transcript AI, and Apply Codes Using AI.

Delve provides three AI tools to assist with qualitative analysis: Transcript AI, AI Chat, and Apply Codes Using AI. Each tool has unique capabilities, inputs, and recommended uses. Understanding how they work and their limitations will help you make the most of Delve’s AI features.


Overview of Delve AI Features

  • Transcript AI: Analyzes a single transcript using your existing codebook to suggest high-level codes, themes, and relevant quotes.

  • AI Chat: Allows deep interaction with coded snippets across your project, helping you brainstorm subcodes, refine codes, summarize snippets, and clarify insights.

  • Apply Codes Using AI: Automatically applies your existing codes to transcripts based on code names and descriptions (deductive coding).

These tools can be used together in sequence for efficient qualitative analysis:

  1. Start with Transcript AI for a high-level overview of one transcript.

  2. Use AI Chat to refine codes, create subcodes, summarize snippets, and explore patterns across transcripts.

  3. Apply Codes Using AI to systematically code transcripts using your refined codebook.


How Each Tool Works

Transcript AI

Purpose: Generate high-level codes, summarize transcripts, and suggest quotes.

Inputs:

  • Individual transcripts (you can only use the AI tool to analyze one transcript at a time)

  • Your existing codebook

Outputs:

  • Suggested codes/themes

  • Summarize transcript

  • Suggest quotes from transcript to support themes or codes

  • Citations linking to your transcript to verify quotations and sources

Limitations:

  • Only sees the selected transcript and your codebook.

  • Cannot see previously coded snippets.

Recommended Use:

  • Start your analysis with Transcript AI for a single transcript.

  • Use its output to guide deeper coding in AI Chat.


AI Chat

Purpose: Assist in deeper analysis and refinement of coded snippets.

Inputs:

  • Selected coded snippets or transcripts (you can filter by code, transcript, or both).

Outputs:

  • Ideas for subcodes, new codes, summaries, or code descriptions

  • Identifying potential snippet outliers

  • Suggestions you can evaluated for accuracy and relevance

  • Citations linking to your snippets to verify quotations and sources

Recommended Use:

  • The AI works with a subset of your data at a time, so filtering by code, transcript, or both will give you the most targeted insights. (It will let you know if it's only analyzing a portion of your data.)

  • Refine codes and subcodes after a first round of broad coding.

  • Summarize code collections and create code descriptions.

  • Explore nuanced patterns across multiple transcripts.

  • Works best iteratively with specific prompts

  • Use AI generated Code Descriptions to improve your Apply codes using AI.


Apply Codes Using AI

Purpose: Deductive application of your refined codebook to transcripts.

Inputs:

  • Individual transcript

  • Existing codebook (code names and descriptions)

Outputs:

  • Applies codebook to selected transcript (AI-applied snippets)

  • Memos explaining why each code was applied to the snippet

When to use:

  • After your codebook is complete or refined

  • To speed up coding across transcripts

Recommended Use:

  • Works best with a well-defined, specific codebook - clear, distinct code descriptions lead to the most accurate results

  • Performs optimally when codes have minimal overlap - refining definitions to reduce ambiguity helps the AI apply codes more precisely

  • Designed for a human-in-the-loop workflow - iterating with human review ensures the highest quality and reliability over time


Suggested Workflow

  1. Transcript AI – Generate high-level codes and summaries for a transcript.

  2. AI Chat – Refine codes, create subcodes, and summarize snippets.

  3. Apply Codes Using AI – Systematically apply refined codes to transcripts.

  4. Iterate – Use AI Chat and manual review to adjust codes as needed.


AI Feature Comparison Table

Feature

Inputs

Outputs

Ideal Use Case

Key Limits

Transcript AI

Single transcript + codebook

Transcript summary, suggested initial codes, highlighted quotes

Early-stage coding, high-level overview

Limited context, does not see already coded snippets

AI Chat

Filtered coded snippets

Subcodes, code summaries, new code ideas, customizable answer formats

Refining codes, summarizing snippet collections, brainstorming

The AI can only analyze a subset of snippets at a time, so filtering by code, transcript, or both is recommended for larger projects

Apply Codes Using AI

Transcript + codebook

Automatically coded transcript, AI memos explaining decisions

Applying finalized codebook efficiently

Accuracy depends on codebook quality, human review suggested

Did this answer your question?