The Canonical Product Overview and System Prompt: Powering Consistent AI Artifact Generation
You can’t generate good artifacts - content strategies, pitches, or business plans without consistent context and voice. AI chats erode both with every new session. The fix isn’t more copy-pasting. It’s two documents that work together: the Canonical Product Overview (CPO) and the AI System Prompt.
They define what your product is and how AI should reason and speak about it. Together, they turn AI into a reliable editor, not a forgetful intern.
The Problem They Solve
AI generation starts strong but drifts:
- Context loss: Facts get fuzzy across chats.
- Voice drift: Outputs revert to generic “AI workspace” hype.
- Semantic creep: Subtle positioning (e.g., “editor-first, context-first”) flattens.
Manual fixes? Endless pasting. Result: Slow, error-prone scaling.
What is a Canonical Product Overview?
A single source of truth for your product’s essence. Deliberately concise, stable, reusable.
Includes:
- Clear problem framing (e.g., tool fragmentation, context loss).
- Replacement narrative (what users switch from and to).
- Mental model of architecture.
- Outcome-first component descriptions.
- Canonical language (taglines, one-liners, promise).
- “What this is NOT” guardrails.
Avoids:
- Fluff, roadmaps, UI tactics, agent/model hype.
For Headgym, it’s the anchor: Editor + browser + notes + agents, centered on flow and memory, not “all-in-one AI tool.”
Inject it into any AI for factual grounding. Onboard teams. Align messaging.
The Power of the AI System Prompt
The CPO informs. The system prompt constrains.
It’s behavioral guardrails:
- Priorities: Emphasize outcomes first, replacement narrative always.
- Avoids: Jargon, overhyping agents/models, generic tropes.
- Tone: Calm, confident, product-led (no hype).
- Resolution: How to frame tradeoffs, ambiguities.
Short (10-15 lines). Injected at session start or within Pablo in a directory in the workspace named “_pablorules”. Acts as a “constitution” - prevents drift even with full CPO present.
How They Complement Each Other
CPO: The what (facts, framing).
System Prompt: The how (behavior, voice).
- Factual stability: CPO provides details; prompt ensures they’re used correctly (no dilution).
- Voice consistency: Prompt enforces style; CPO supplies the language snippets.
- Scalability: Stack them for any artifact—no repasting.
Example: Generating a Pitch
Prompt: "Using CPO and system prompt, write a pitch for Headgym to PMs."- CPO injects: Problem (context switching hell), replacement (unified flow).
- Prompt ensures: Editor-first tone, downplays agents, narrative arc.
Output: Sharp, on-brand, hallucination-free.
Real-World: Headgym Content Repurposing From 280 podcasts (marketing/PM focus):
- Pablo generates 300+ articles (“7 Insights from [Guest]”), LinkedIn posts, Twitter threads.
- CPO/system prompt: Consistent Headgym voice across blog/Medium/LinkedIn.
- Scale: Tag interviewees, hit 900-1000 networks. No drift.
Key Benefits
| Without CPO + Prompt | With CPO + Prompt |
|---|---|
| Frequent context loss per chat | Persistent truth across chats |
| Voice drift to hype | Stable, calm tone |
| Hallucinations | Guardrail-protected |
| Slow pasting | One-time injection |
| Inconsistent artifacts | Scalable, consistent output |
Best Practices
- Craft CPO: Human-readable, AI-injectable. Update rarely.
- Build Prompt: Behavioral rules only. Test on edge cases (e.g., “Describe agents”).
- Inject Selectively:
- ✅ Marketing, sales, docs, demos.
- ⚠️ Brainstorming.
- ❌ Pure tech debug.
- Evolve: Version as product changes. Use as “cheap insurance.”
- Pablo Workflow: Read workspace files (like this one), plan in messages, edit precisely.
Conclusion
In an era of AI wrappers, differentiation is subtle. CPO + system prompt preserve it.
They make AI an extension of your thinking: Consistent. Reliable. Scaled.
No more cut-paste drudgery. Just artifacts that sound like you.