Direct Answer
What is a 23-column prompts knowledge base?
A knowledge base built by running 23 sequential prompts — each producing one named column in a structured row — that together represent a complete company identity. The 23-column structure is the load-bearing minimum required for downstream libraries (Content, Outreach, Sales) to read without ambiguity. Every downstream consumer reads the same 23 columns, so nothing in the platform can disagree with itself.

A brand without a synchronized knowledge base is twenty disconnected documents pretending to agree with each other. The 23-column prompt chain is the architectural answer — one row, twenty-three fields, a single source of truth that every downstream library reads. The chain is not how content is written. It is how content stops contradicting itself.

Why 23 Column Prompts and Not One

The intuitive shortcut is to write one mega-prompt that asks an LLM to generate the entire brand identity in one go. The mega-prompt produces a sleek-looking document and a giant blob of text that nothing downstream can use. You cannot query it. You cannot version individual fields. You cannot tell the Content library to read only the positioning section without parsing prose.

The 23-column chain solves all of that by being a database row instead of a document. Each column is one named field — Mission, Positioning, ICP Profile, Brand Voice, Tagline, Bold Claim. Each field is produced by a dedicated prompt. Each field is independently queryable, diffable, version-controllable, and consumable by exactly the downstream library that needs it.

The 23-column chain is not how content is written. It is how content stops contradicting itself.

Tommy Saunders · Founder, Windfield Real Estate

The 23-Column Prompt Chain

The chain is shown below as 23 tiles. Each tile is one prompt. The order is load-bearing — early columns capture intake, middle columns synthesize identity, late columns produce the public-facing artifacts. Every column reads every column to its left as structured context.

Figure 01 · 23-Column Prompt ChainSequential · Hybrid Tooling
01Notion
Business Type
Industry, model, stage.
02Notion
Stage
Pre-seed → Growth.
03Notion
Founder Story
Origin and why.
04Notion
Origin Insight
Wedge insight that started it.
05External
Mission
One sentence, present tense.
06External
Vision
10-year world if we win.
07External
Core Values
3–5, with anti-patterns.
08Notion
Customer Type
Title, company size, segment.
09Notion
Pain Points
Top 5 named pains.
10External
ICP Profile
Where they hang out, buy triggers.
11External
Value Prop 01
Headline benefit.
12External
Value Prop 02
Supporting benefit.
13External
Value Prop 03
Proof-point benefit.
14Notion
Competitors
Named, with one-line note.
15External
Differentiators
Why we win vs. each competitor.
16External
Positioning
For X who Y, we are Z.
17External
Brand Voice
Adjectives + rationale.
18External
Tone Spectrum
Formal ↔ Casual band.
19Notion
Vocabulary
Say / don't-say lists.
20External
Tagline
≤6 words. Defendable.
21External
Elevator Pitch
30 seconds. Conversational.
22External
Manifesto
Long-form internal narrative.
23External
Bold Claims
3–5 defensible public claims.
Eight Notion AI columns. Fifteen External AI columns. The tool indicator on each tile tells you which engine produces that field. The order is fixed because each column reads all the columns to its left as structured input.

Notion AI vs External AI

Eight columns are Notion AI prompts. Fifteen are external AI prompts (Claude or GPT). The decision rule is simple: if a column needs to read other Notion pages or databases as context — competitor pages, customer records, vocabulary indexes — it runs on Notion AI. If a column needs long-form generation under tight structural control — mission, positioning, manifesto — it runs on external AI.

This is the part most prompt-library projects get wrong. They commit to one engine and pay for it in either speed or output quality. The chain is hybrid because the work itself is heterogeneous; some fields are lookups, some fields are generation. Different jobs deserve different tools.

The tool indicator on every tile

Each column is tagged with its tool — Notion or External — so the operator running the chain knows what is happening when. The tag is also the routing instruction for the orchestrator: Notion-tagged columns route to the Notion AI API; External-tagged columns route to the hosted model. The same diagram is the documentation and the execution plan.

Before / After — Manual vs the Chain

Before the library, defining brand identity was a multi-week, multi-stakeholder process that produced a deck nobody read after the first three weeks. After the library, it is a 30-minute questionnaire and a four-minute chain run.

  1. Manual brand definition (before). Founder writes a draft. Marketing manager redrafts. Three weeks of revisions. A 40-page deck. Six months later, no two team members can quote it consistently. The deck is forgotten; the brand drifts.
  2. Library run (after). Founder completes a 30-minute intake questionnaire. The 23-column chain runs in 200–230 seconds. The output is a single row of structured data — 23 named fields, all internally consistent, all queryable by downstream libraries.
  3. Downstream regeneration. The Content, Outreach, Sales Enablement, and Visual Asset libraries read the row and regenerate their own artifacts. The website’s About page updates. The sales script primer updates. The LinkedIn description updates. No manual sync, no copy-paste, no drift.
  4. Maintenance. When the brand evolves — a new positioning, a new value prop — the founder updates the relevant column and the whole downstream system regenerates. Brand evolution becomes a database update, not a six-month rebrand.

Brand Consistency Across Libraries

The reason this matters is the same reason every other library in the platform exists: consistency across surfaces is an architecture decision, not a discipline initiative. The 23-column row is the contract. Every library that consumes the row commits to reading it the same way every time. If the Content library reads positioning as “X for Y”, the Outreach library cannot reframe it as “Y for X”. The row is the source of truth and the source of truth wins.

That is what makes the platform feel like one studio. The articles, the LinkedIn posts, the cold emails, the deck, the website, the press release — they all sound like the same company because they were all generated from the same row. The salesperson, the marketer, and the founder cannot disagree with each other because none of them are writing the brand voice; the brand voice is a database field.

23
Columns
230s
Chain Runtime
1
Source Row

One row, twenty-three columns, every downstream library reading the same data. That is the whole story. The chain is not the impressive thing; the row is. The chain is just how you get the row produced in four minutes instead of four weeks.

Frequently Asked Questions

5 Questions
What is a 23-column prompts knowledge base?
A knowledge base built by running 23 sequential prompts — each producing one named column in a structured row — that together represent a complete company identity. Downstream libraries (content, outreach, sales) read the same 23 columns without ambiguity.
Why a column chain and not a single mega-prompt?
Because a mega-prompt produces one blob of text you cannot query, version, or route. A column chain produces 23 named fields, each auditable, each diffable, each individually consumable by downstream agents. The chain shape is what makes the knowledge base useful, not just generated.
How are Notion AI and external AI mixed in the chain?
Eight columns run on Notion AI because they need to read other Notion fields as context — competitors, customer pain points, vocabulary. Fifteen columns run on external AI (Claude or GPT) because they need long-form generation with controlled structure. The chain is hybrid by design.
What is the runtime?
Under four minutes total for the 23-column run. The intake questionnaire is the slow part (about 30 minutes of founder time). The chain itself executes in 200–230 seconds. After the chain completes, downstream artifacts regenerate automatically.
What downstream systems read the knowledge base?
Every other library in the platform. Content reads positioning and voice. Outreach reads value props and pain points. Sales Enablement reads competitors and differentiators. Visual Assets reads tone and mood. All of them read from the same 23-column row, so nothing can disagree with itself.
About the author
T
Tommy Saunders
Founder, Windfield Real Estate
Building the AI-native content operations system for business operators who need predictable output, not AI experiments.