Content workflows: voice-consistent writing and repurposing

Most AI-assisted content tools have the same problem: they know you told them your voice, but they don't actually know your voice. You paste in a description — "direct, conversational, no jargon" — and get output that sounds like a description of your writing, not your writing. The reason is that descriptions of voice are not the same as examples of voice.

Cowork solves this differently. Instead of describing your voice, you point at it. "Read my last five articles in context/" isn't a style instruction — it's a reading assignment. Cowork extracts patterns from actual examples: sentence length distribution, how you open paragraphs, the ratio of declarative to interrogative sentences, specific phrasing patterns that recur. The output reflects what you actually write, not what you think you write.

Voice-consistent article writing

The basic pattern:

"Read my last five articles in context/, then write a new article on [TOPIC] in the same voice. Flag any phrases that sound like AI-tells and offer alternatives."

The flag-and-offer step is not optional if you care about quality. AI-generated content has recognisable patterns — em-dash overuse, "it's not just X, it's Y" constructions, sentences that start with "Ultimately" or "At the end of the day." When you ask Cowork to flag these and suggest alternatives, you get a draft where the tells are visible and correctable, not buried in prose you have to read twice to spot.

The better your context/ folder, the better the output. Three articles produce a sketch. Ten produce a reliable voice model. Pair this with a voice file (a dedicated markdown document listing your patterns, preferences, and examples of good and bad output) and you stop repeating the same voice instruction in every prompt — Cowork reads the file and applies it automatically.

Blog-to-LinkedIn repurposing

Template-based repurposing tools turn a blog post into a LinkedIn post shaped like a LinkedIn post. That's not the same as your LinkedIn posts.

Cowork's approach: before generating the repurposed content, it browses your actual LinkedIn profile to read your recent posts, identifying your current format and style — not a generic "LinkedIn style." The result matches what you're already posting, including structural habits like whether you use line breaks between every sentence, how you end posts, whether you use bullet points or paragraphs.

The prompt pattern:

"Read the article in context/article.md. Browse my LinkedIn profile at [URL] and read my last ten posts. Repurpose the article into a LinkedIn post matching my current style and format."

Slide deck generation from briefing docs

Point at the briefing doc, specify the audience, and describe the desired output:

"Read context/product-brief.md. Produce out/launch-deck.pptx for a sales audience. Each slide should have a headline, three supporting points, and speaker notes. Use a structure of: problem, solution, proof, ask."

Cowork loads the pptx skill and produces an actual PowerPoint file with layouts, speaker notes, and a structure that follows what you specified. This is a first draft — not a finished deck — but it's a first draft that takes fifteen minutes instead of three hours.

Case study production from raw client interviews

Drop raw interview transcripts into raw/ — messy, unedited, timestamped. Ask for a case study in out/:

"Read all interview transcripts in raw/. Produce out/case-study.docx in the format: challenge, approach, outcome, client quote. Extract the three strongest quotes into out/quotes.md."

The document reflects what the client actually said, not a cleaned-up paraphrase of what you remember them saying. Review it, adjust the framing, polish the language — but start from something real.

The anti-slop discipline

Every content workflow should start by pointing at two files: your voice file and your anti-slop file. The anti-slop file is a list of explicit bans — phrases, constructions, and patterns you never want in your content. Em dashes used for emphasis. "It's not just X, it's Y" sentence structure. Hedging phrases like "it's worth noting that" or "in many ways." Emoji.

These files replace the instruction you'd otherwise have to type in every prompt. They live in your ClaudeCowork/ folder and persist across all content tasks. As you find new tells to ban, update the file. Cowork reads it before generating anything, and the quality floor rises without you having to think about it.