Primary-Source Content Distillation
Turn expert conversations into structured outlines without outsourcing the voice
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 6
- Confidence
- 98%
Feed AI a high-quality primary source such as an expert interview, podcast transcript, recorded call, or collection of relevant videos. Ask it to identify important themes, unexpected observations, disagreements, and exact supporting passages. A human editor then chooses the most promising angle and has the model organize the selected evidence into a detailed outline suited to the destination channel. The writer adds independent research, interpretation, and a genuine voice rather than publishing the generated text directly. This workflow uses AI for compression and structure while preserving the human functions that make content distinctive: taste, point of view, synthesis, and accountability. It also creates leverage from valuable conversations that would otherwise remain trapped in audio or video formats.
Origin
Gail Axelrod at Jellyfish uses recorded conversations with subject-matter experts to generate highlights and detailed outlines that she then develops into finished content.
Core principles
- 01Begin with primary-source insight
- 02Use AI to extract and organize, not impersonate the author
- 03Preserve traceability to the source material
- 04Adapt insight to the destination channel
- 05Keep human judgment responsible for the final point of view
How to run it
- 1
Choose primary sources
Select interviews, calls, videos, or podcasts containing firsthand expertise and useful tension. Prefer material unavailable in generic search summaries.
Pro tip Combine multiple sources when you want patterns and disagreements rather than a single viewpoint.
Watch out Low-information transcripts only produce better-organized low-information content.
- 2
Extract candidate insights
Ask the model for key ideas, surprising claims, repeated themes, disagreements, and exact supporting passages. Require clear links back to source locations.
Pro tip Request unexpected or counterintuitive observations separately from general themes.
Watch out Verify excerpts against the transcript before publication.
- 3
Select an editorial angle
Use human judgment to choose the audience, promise, and distinctive point of view. Discard material that is interesting but irrelevant to that angle.
Pro tip Look for a tension that the source resolves rather than a broad topic summary.
Watch out Letting the model choose the prevailing view can erase the author's distinctive perspective.
- 4
Build the outline
Organize the chosen evidence into a detailed beginning, development, and conclusion for the target channel. Attach supporting source material to each section.
Pro tip Ask what additional evidence would strengthen or challenge each section.
Watch out Do not treat an outline as a publishable draft.
- 5
Add research and interpretation
Supplement the primary source with checked data, examples, and counterarguments. Explain why the extracted observations matter to the intended reader.
Pro tip Use cited research to test rather than merely decorate the source's claims.
Watch out AI-generated citations and statistics require direct verification.
- 6
Write in a human voice
Draft and edit the piece using the author's real vocabulary, judgment, and experience. Remove formulaic language and unsupported transitions.
Pro tip Use AI for targeted revision questions rather than copying an entire generated article.
Watch out Publishing an unedited AI draft can damage audience trust.
In the wild
A content director records a conversation with a subject-matter expert, uploads the transcript, and asks ChatGPT to extract highlights and create a detailed outline. She then brings in outside data and writes the finished piece in the publication's voice.
→ The team reaches a strong first structure in substantially less time without surrendering editorial control.
A creator collects transcripts from twenty YouTube videos about one topic. The model identifies common themes, overlooked issues, and the spiciest disagreements, which the creator uses as raw material for an original analysis.
→ A large source set becomes a navigable map of possible angles and evidence.
Common mistakes
Publishing the generated draft
The workflow's value comes from extracting and organizing insight, not replacing the writer's judgment and voice.
Losing source traceability
Extracted claims without locations or quotations become difficult to verify and easy to misrepresent.
Choosing only consensus themes
Summarizing what every source says can produce accurate but undifferentiated content.
Is it for you?
Best for
It is best for content teams repurposing interviews, podcasts, webinars, or expert calls into written pieces.
Not ideal for
It is not ideal for generating personal thought leadership from generic source material with no distinctive insight.
From the transcript
“she uses it to produce detailed outlines from primary source material.”
“She pulls that into chatbt and has it pull out key highlights from the conversation and then essentially produces a detailed outline of what the…”
“it's always about pairing the human expertise with you know the chatbt or the AI expertise.”
From the episode
9 ChatGPT Use Cases from the Top 1% of Marketers