# Values-explicit prompting closes the AI-human writing gap **Claim:** When you communicate *explicit organizational values* to an AI model (via detailed style guides that articulate culture, not just voice), AI output quality converges with human writing quality — measured by audience reception and engagement. **Scope & conditions:** - The organization must have articulable, specific values (not vague brand clichés like "friendly" or "approachable", which actually cause AI to output creepy or misaligned content) - The AI system must have sufficient context length to process a detailed style guide (not a 5-line prompt) - Evaluation must be on *audience reception* (growth metrics, engagement) not internal aesthetic judgment by the team **Evidence:** > [!quote] > "A key unlock for better writing has been communicating values to the model. Same as humans. Characteristics like 'friendly' often made the AI output 'creepy.' Getting specific here really helped." The WordPress Marketing Team invested in a 44,600-word style guide centered on explicit values. When automation was paused in late 2024 for a strategy review, subscriber growth dropped 38%; when automation resumed with the values-guide, growth recovered and outperformed manual management. This pause-and-resume creates a quasi-natural experiment validating the claim. **Why it matters:** Reframes AI prompt engineering away from "technical tuning" toward *organizational culture work*. Makes the problem tractable: if you can articulate your values as an organization, you can close the writing quality gap with AI. Also suggests a scalable template for open-source projects deciding whether to automate content production. ## Links - [[Perm · The middle is disappearing as AI commodifies previously-expert work]] — applies-in: values-rich prompting lets a producer compete after generic execution becomes commodity work - [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] — broader AI capability convergence - [[Perm · Automation removal reveals its own value via growth drop]] — evidence-for: the pause-and-resume result tests the growth contribution of the values-explicit automation rather than assuming it from output volume <!-- atomicity: 1 idea ✓ -->