Notebook Entry 015
Source Attribution Drifts When Working From Secondary Summaries
April 11, 2026
Did the content I'm about to publish actually say what the primary source said?
What happened
A LinkedIn post drafted in Session 61 using a Cowork summary of an AIDB episode (instead of the primary transcript) contained multiple attribution errors. The post attributed "New data from Writer, Deloitte, and MIT Sloan" to the AIDB episode — but the episode actually covered A16Z, KPMG, Writer/Workplace Intelligence, and WalkMe. Deloitte and MIT Sloan came from an independent research task that produced REF-market-validation-data.md. The first comment was set to credit the episode's host by name for studies he never discussed. The post was scheduled for Tuesday April 15 — three days after the error was caught.
Root cause
The Cowork summary correctly captured some of the episode's data but mixed it with the broader research document's data during synthesis. The summary looked authoritative. The error only surfaced when the primary transcript was read line-by-line next to the post draft. Secondary summaries — Cowork outputs, episode descriptions, research syntheses — are compressed and can blend sources without explicit attribution tracking.
What changed
Hard rule added to CORE-01a and CORE-00: when strategy or content is built from episode descriptions, Cowork summaries, or secondary sources, always verify against the primary transcript before publishing. Crediting a named person for studies they didn't cover is a credibility risk, especially when that person is someone you're building a public relationship with. The cost of a source check is 10 minutes. The cost of misattribution is the relationship.
The lesson
AI summaries are lossy compressions. They're useful for orientation but dangerous for attribution. Any content that cites specific sources — especially when a named person will be publicly credited — must be verified against the primary source, not the summary of the primary source. This applies to any organization using AI to summarize reports, research, or content for communications purposes.
One implementation insight every other Tuesday.
What actually works when you ship AI in real organizations. Nothing you could get from a press release.