How To Get A Consistent Voice Writing With AI
Source: https://writewithai.substack.com/p/how-to-get-a-consistent-voice-writing Authors: Dickie Bush & Nicolas Cole
Why this is in the vault
Three-rule framework for getting consistent, on-voice AI output — directly applicable to Sanity Check production. The core diagnosis: the problem is almost never the prompt; it's how the workspace is organized. Fix is architectural: isolate threads by content type, reset when style shifts within a format, and build a persistent context hub so AI never "forgets" between sessions.
The 3 rules:
Isolate one thread per content type. Voice changes by format — newsletters, short-form social, YouTube scripts each need their own thread. Mixing formats makes AI juggle conflicting rulebooks; output drifts to the generic middle.
Know when to start a new thread within the same format. If style changes significantly (e.g., switching from opinion-based to data-driven newsletters), open a new thread. Even same-format, different-style writing needs clean context.
Build a context hub (Notion or equivalent). Maintain a persistent "brain" with: target customer profile, positioning/niche, voice guidelines + best-work examples, content principles, proprietary frameworks. Link it at the start of each new thread. Hub = persistent memory. Individual threads = working memory. Keep them separate so neither has to do the other's job.
Voice Lab: They released a free Claude Skill ("Digital Writer's Voice Lab") — paste in your best writing samples, it extracts voice DNA into a profile you drop into your Notion hub. Not paywalled.
Self-promotional CTAs (noted, not flagged as third-party sponsor): Opening and closing plugs for their "Live Experiences" bootcamp series, including "Claude Fundamentals — Become an AI PRO in 5 Days" (doors July 13). CTAs are pre-header and post-script; substantive content stands independent of them.
Mapping against Ray Data Co
The thread-per-format rule is an immediate audit trigger for Sanity Check. We have no enforced boundary between SC draft threads and other content types today. Applying the framework: SC newsletters get one dedicated project/thread; SC social gets a separate thread; phData-adjacent content gets its own. The existing [[voice-profile-2026-05-08|Sanity Check voice profile]] is the right raw material for the context hub — it's already the DNA extract; it just needs to be consistently linked at thread-start rather than assumed ambient.
The "context hub" pattern is partially implemented in RDCO already (CLAUDE.md + MEMORY.md + the SC voice profile), but not as a formalized session-start ritual per content type. The Voice Lab skill is worth running against SC's best issues to sharpen the DNA profile — it may surface patterns the current voice-profile didn't capture because it was built from chat history rather than actual published pieces.
Related
- [[01-projects/sanity-check/voice-profile-2026-05-08]] — The Sanity Check voice DNA; this is the context-hub input the 3-rule framework calls for
- [[2026-02-26-write-with-ai-unbundling-framework]] — Prior Write With AI craft framework; complementary workspace organization principles from same authors
- [[2026-03-19-every-ai-style-guide]] — Every's AI style guide method; overlaps directly on per-format rule sets and voice extraction from writing samples
- [[2026-06-10-write-with-ai-halbert-sales-letter-claude-skill]] — Write With AI Claude Skill pattern (Halbert); sibling approach to this Voice Lab skill