06-reference

writewithai commodity vs personality content

2026-09-07·reference·source: Write With AI·by Nicolas Cole (guest; cross-posted from "The Art & Business of Writing")
writing-craftai-writingvoicecontent-ownershipghostwritingsanity-check-voice-study

"The Right Way To Write With AI" — Nicolas Cole (cross-post via Write With AI)

Why this is in the vault

Cole's cleanest articulation yet of a distinction Ray Data Co already leans on implicitly — that "AI-written" and "generic" are not the same failure, and the fix isn't avoiding AI, it's feeding it a real data set.

The core argument

Cole splits all writing into two buckets. Commodity Content is anything derived from the shared, open-source substrate of language — conventional wisdom, how-to advice, technical explainers — that no individual owns and that a competent LLM can now produce as well or better than a human, because it requires no unique information, only fluency. Personality Content is the opposite: life facts, opinions, insights, stories, and lessons that only exist because one specific person lived them — a data set LLMs structurally cannot have unless the person first articulates it themselves. His claim: you cannot out-write, out-speed, or out-cheapen AI on Commodity Content, so the only defensible position is building a "Digital Brain" — his term for a written library of Personality Content — and then letting AI handle repetition and packaging of ideas you've already claimed as yours. He extends this to "voice," arguing most people conflate "I typed it" with "it sounds like me," when voice is actually decomposable into vocabulary, rhythm, and purpose (his "EDAN" framework — Explanation/Description/Action/Narration), and AI can approximate any of those once it has examples to draw from. Ghostwriting, in his framing, isn't hallucinating someone's identity — it's extracting Personality Content via interview and organizing it, which is functionally identical to what an LLM does with a well-fed prompt. He closes by distinguishing "Original Content" (net-new thinking) as the only category untouched by either bucket, and the only one worth the majority of a writer's remaining time as Commodity Content gets fully commoditized.

Mapping against Ray Data Co

Direct, load-bearing overlap with the Sanity Check voice-study effort and the harness-engineering thesis: Cole's "Personal Language Model" is the same object RDCO's WRITING-rdco.md standing contract and the per-brand DESIGN-.md files are trying to encode — a durable, personal data set that keeps AI output from reading as generic even as the founder scales output volume. It also reframes a recurring internal question (is a given draft "too AI-sounding") away from "did AI write the sentence" toward "does the sentence carry any Personality Content" — a sharper, checkable test than vibes-based AI-slop detection. Worth stealing directly: EDAN as a lightweight voice-QA lens for /draft-review, and "you can't automate what you can't articulate" as a one-line justification for why the vault (the founder's own life facts, opinions, stories, lessons) is the actual moat, not the writing tools layered on top of it.

⚠️ Sponsorship

sponsored: true / sponsor_entity: self. No third-party paid placement. Two self-promotional elements: (1) a mid-essay CTA to join Cole's "AI Writing Skool" community; (2) a footer bio block listing all of Cole's ventures he co-founded or founded — Different Publishing, AI Writing Skool, Ship 30 For 30, Premium Ghostwriting Academy, Typeshare, Ghostbase, and Write With AI itself (the sending publication). This is house cross-promo across Cole's own venture portfolio, not a disclosed third-party sponsor — treat the argument as credible but note the author has a direct commercial incentive to position "build a Personal Language Model" as valuable, since several of his own products (AI Writing Skool, Ghostbase) sell exactly that service.

Related