Why Three-Pass AI Summarization Beats Single-Shot
Our 3-pass pipeline with GPT produces dramatically better meeting summaries than a single prompt. We explain the architecture and share real comparisons.
When we first built Huddix's summarization system, we started with a simple approach: transcribe the meeting, feed the transcript to an LLM with a prompt like "summarize this meeting," and return the result. It worked — but the summaries were often generic, missed key points, and lacked the nuance that makes meeting summaries truly useful.
After extensive experimentation, we developed a three-pass pipeline that produces dramatically better results. Here's why.
The Problem with Single-Pass Summarization
- Lack of focus — A single prompt must balance multiple objectives simultaneously, often doing none perfectly.
- Inconsistent structure — Without explicit guidance, LLMs produce summaries with varying structures that are hard to compare across meetings.
- Missing nuances — Important details get lost in the generalization process.
- Factual errors — LLMs can hallucinate details when summarizing long transcripts in a single pass.
Our Three-Pass Pipeline
Pass 1: Structured Extraction
In the first pass, we use GPT to extract structured data from the transcript: speakers identified, topics discussed, questions raised, and any explicit commitments made. We use a JSON schema to ensure consistent structure and minimize hallucination. This pass also performs heuristic sentiment analysis.
Pass 2: Thematic Analysis
The second pass analyzes the extracted data to identify themes and patterns. Which topics received the most discussion? Where did people disagree? What decisions were made? This pass also enriches the analysis by pulling in cross-meeting memories.
Pass 3: Summary Generation
The final pass synthesizes the structured extraction and thematic analysis into a coherent, well-structured summary with explicit sections: Executive Summary, Key Decisions, Action Items, Discussion Topics, Unresolved Questions, and Past Context.
Real Comparisons
We tested both approaches on 100 meetings and had human evaluators rate the quality. Results:
- Completeness: 4.2/5 (3-pass) vs 2.8/5 (single-pass)
- Accuracy: 4.5/5 vs 3.9/5
- Action item extraction: 92% vs 67%
- Useful for later reference: 4.4/5 vs 2.9/5