Most brand documents drift the day after they ship. The founder approves a positioning statement. The marketing manager writes it down. Six months later the deck says one thing, the website says another, and the sales team is using a third version they cobbled together because the first two did not survive contact with a customer. The Company Identity Prompt Library is what we built so that does not happen anymore.
Why Brand Documents Drift
A brand document drifts because it has no enforcement layer. It sits in a Google Doc. It gets copy-pasted into a deck. It gets reinterpreted by every new hire who reads it. Each interpretation is a tiny act of drift, and tiny acts of drift compound. The fix is not a better brand document. It is a brand document that downstream content cannot bypass.
That is what the library produces. The 23 column prompts generate a knowledge base, not a manifesto. Every downstream library — Content, Outreach, Sales Enablement, Visual Assets — reads from the same 23 columns. There is no interpretation layer because there is no document; there is structured data. A salesperson cannot reinterpret a value proposition that they read from a database row.
A brand document drifts because it has no enforcement layer. Fix the layer, not the document.
What Column Prompts Are
A column prompt is one prompt that produces one field in a structured knowledge base. The library is organized like a spreadsheet — 23 columns, each filled by one prompt. The prompts run in sequence, and each one reads the columns to its left as context. The output is a database row, not a free-form essay. That single architectural decision is what lets the system stay coherent.
Column prompts are different from a single mega-prompt that asks an LLM to produce a brand document end-to-end. A mega-prompt produces one giant blob of text that you cannot query, cannot version-control field-by-field, and cannot route to specific downstream consumers. Column prompts produce a row with 23 named fields, each one auditable, each one diffable, each one reusable.
The 23-Column Chain
The chain runs left to right. Column 01 is the seed — business type and stage. Column 23 is the bold market claim — the most ambitious public statement the brand is willing to defend. Every column in between depends on the columns to its left. The chain is not parallel; the chain is a relay.
Notion AI + External — A Hybrid Chain
Eight of the 23 columns are Notion AI prompts. Fifteen are external AI prompts (Claude or GPT). The split is not arbitrary. Notion AI is fast, reads other Notion fields natively, and is the right tool for fields that need cross-page context — pulling from existing customer data, prior brand documents, related Notion databases. External AI is slower but handles long-form generation better and gives the orchestrator finer control over output structure.
Columns 01–04 are intake; Notion AI handles them. Columns 05–07 are foundational identity (mission, vision, values); external AI handles them because the output needs structured prose. Columns 08–09 are customer signals; Notion AI. Columns 10–13 are ICP and value propositions; external AI. Columns 14, 19 are inventory tasks (competitors, vocabulary); Notion AI. The rest — positioning, voice, tagline, pitch, manifesto, claims — are external AI because the output is the brand's literal voice and external models do that work better today.
A pure external-AI chain would be slower and would lose Notion's native ability to read the rest of the workspace. A pure Notion-AI chain would lose the long-form quality of the external models. Hybrid is not a fallback. It is the right architecture — different tools for different jobs, orchestrated by one chain.
What the Library Produces
The end state is a 23-field row in the brand knowledge base. Every downstream library reads from those 23 fields. The library produces, in addition to the row itself, nine derived artifacts that pull from the row and shape it for specific consumers — the deck-ready one-pager, the website's About page, the LinkedIn company description, the sales script primer, the brand voice guide, the visual brief template, the press-release boilerplate, the investor narrative, and the manifesto for internal use.
Nine artifacts from one row. Every artifact is regenerated when the underlying row changes — no manual sync, no copy-paste. When the founder updates the positioning column, the about page regenerates. When the value props change, the sales script primer updates. The system stays in sync because the source row is the only thing anyone has to touch.
The founder interview is the slow part — usually 30 minutes. The chain itself runs in 200–230 seconds. The output is a complete, internally consistent brand knowledge base, ready for downstream libraries to read. There is no revision loop because there is nothing to revise. There is one row, the row is the truth, and the truth propagates.