Chapter 7: Content Creation and Writing

17 min read

What It Is

Claude Code is not just a writing assistant — it is a writing system. The difference matters. A writing assistant helps you with one piece of text at a time. A writing system lets you analyze your own style, replicate it consistently, chain research into finished content, produce multiple format variations, and build repeatable workflows that turn raw material into polished output.

Because Claude Code has access to your file system, it can read your past work, learn your patterns, and apply them to new content. Because it can access the internet, it can pull in research and context. Because it can create files, it can save outputs in the exact format you need — RTF for pasting into LinkedIn, markdown for a blog, HTML for a web page.

For professionals who create content regularly — LinkedIn posts, newsletters, one-pagers, podcast descriptions, event summaries — this chapter covers the workflows that turn hours of content production into minutes.

Why It Matters for Non-Technical Professionals

Content creation is one of the most time-consuming activities for marketers, executives, researchers, and thought leaders. The bottleneck is rarely ideas — it is the production process. Turning a great insight into a polished, on-brand piece of content involves research, synthesis, writing, editing, formatting, and iteration. Each step takes time.

Claude Code compresses this pipeline. You can go from a raw transcript to a finished LinkedIn post in ten minutes. From a sixty-page research report to a styled one-pager in fifteen. From a dataset to a data-driven social media caption in five.

More importantly, Claude Code enables consistency. Once you teach it your style — your sentence structures, your formatting preferences, your content architecture — it can replicate that style across every piece of content it helps you create. This is something no chatbot interface can do as well, because Claude Code can actually read your past outputs from your file system.

What You Need

No special plugins required for basic content work. However, these enhance the experience:

document-skills plugin — For saving outputs as DOCX, PDF, or PPTX
frontend-design plugin — For creating infographics and visual content pages

Recommended permissions: - Bash(python3:*) — For creating visual content (infographics, charts)

The Workflows

Workflow 1: Analyzing and Replicating Your Writing Style

Before Claude can write like you, it needs to study how you write. This is the foundational workflow for all content creation.

As a Claude prompt:

I'm sharing two of my best-performing LinkedIn posts as images.
Analyze how I write these structured, insightful one-pagers:

/Users/krishna/Desktop/Content/viral-post-1.jpeg
/Users/krishna/Desktop/Content/viral-post-2.jpeg

Specifically analyze:
- Language patterns and sentence structure
- How I segment topics
- The depth and type of content
- How I use bold text and formatting
- The opening hook style
- How I structure the closing/CTA
- What makes these scannable on mobile

What Claude does: 1. Reads both images (Claude can analyze screenshots of posts) 2. Identifies structural patterns (header style, bullet formatting, paragraph length) 3. Analyzes language choices (active vs. passive voice, technical vs. accessible) 4. Documents the formatting framework (how bold is used, section breaks, whitespace) 5. Creates a “style guide” that it applies to future content

Why this matters: Once Claude has analyzed your style, you can reference this analysis in future prompts. “Write this in the style we analyzed earlier” becomes a shortcut to consistent, on-brand content.

Real example: I had Claude analyze two of my viral LinkedIn one-pagers. It identified patterns I was not even conscious of — that I always use exactly four to six topic segments, that my sentences stay under twelve words in bullet points, that I use bold for the first two words of each section as a scanning anchor. This analysis became the foundation for all subsequent content work.

This is my single most-used content workflow. It turns any long document — research reports, analyst papers, survey results, interview transcripts — into a structured, scannable one-pager in my personal style.

As a Claude prompt:

Read this research report:
/Users/krishna/Desktop/Reports/industry-outlook-2025.pdf

Create a one-pager summary following these rules:
- Use my established one-pager style (bold topic headers, concise
  bullets, data points prominent, scannable on mobile)
- Focus on non-obvious insights — skip anything that's common knowledge
- Include specific data points and statistics
- Aim for 6-8 topic segments
- Keep each bullet point under 15 words
- Do NOT mention any specific companies by name
- Save as RTF (not markdown) at /Users/krishna/Desktop/Content/
  industry-outlook-onepager.rtf

What Claude does: 1. Reads the entire PDF (even 60+ page reports) 2. Identifies the most interesting, non-obvious findings 3. Structures them into the specified one-pager format 4. Formats according to your style rules 5. Saves as RTF for easy copy-paste into LinkedIn or other platforms

Why RTF instead of markdown: Markdown formatting (asterisks for bold, hashes for headers) does not paste cleanly into LinkedIn, Google Docs, or email. RTF preserves formatting when you copy and paste.

Real example: I took a published investment research report — over sixty pages of analysis — and had Claude synthesize it into a structured one-pager following my exact formatting style. The result was a piece of content I could share on LinkedIn that would have taken me two to three hours to write manually. Claude produced a first draft in three minutes. After two rounds of iteration (adjusting which insights to emphasize, tweaking the framing of certain data points), it was ready to publish.

If you produce or appear on podcasts, this workflow turns episode transcripts into social media content.

As a Claude prompt:

Read this podcast transcript:
/Users/krishna/Desktop/Content/podcast-episode-12-transcript.txt

Create a LinkedIn caption for this episode. Follow these rules:
- Start with a compelling hook (2-3 options to choose from)
- Include 4-5 bullet points highlighting the key discussion topics
- Add a brief context line about the podcast and guest
- End with a subtle CTA (not "like and subscribe" — something organic)
- Tone: conversational but authoritative
- Length: 800-1200 characters (LinkedIn's sweet spot before "see more")
- Save as RTF

What Claude does: 1. Reads the full transcript 2. Identifies the most compelling discussion points 3. Crafts multiple hook options 4. Writes the caption in the specified format 5. Saves as RTF

Real example: For a podcast covering an investment thesis on a specific sector, Claude read the full transcript (a conversation between two team members), extracted the five most interesting discussion points, and produced a caption with three hook variations. I picked the best hook, adjusted one bullet point, and published. Total time: ten minutes including review.

Workflow 4: Data-Driven Post from Original Analysis

When you have done data analysis (Chapter 6), the natural next step is a social post about what you found.

As a Claude prompt:

I just completed an analysis of declining song duration across 33 years
of a music director's discography. The key findings are:

- Average song length declined from 5.8 min (1990s) to 4.2 min (2020s)
- The iconic songs of the 1990s-2000s averaged 6.47 minutes
- Songs over 6 minutes virtually disappeared after 2015
- The decline accelerated in the streaming era (post-2015)

The chart is saved at /Users/krishna/Desktop/Charts/song-duration-chart.png

Write a LinkedIn caption that:
- Opens with a provocative hook about attention spans
- Uses the data to tell a story (not just state facts)
- References the specific music director as an example of a broader trend
- Includes 3 alternative hooks so I can pick the best one
- Tone: thoughtful, slightly nostalgic, backed by data
- Save as RTF

What Claude does: 1. Takes your analysis findings as input 2. Crafts a narrative arc (not just a data dump) 3. Produces multiple hook options 4. Writes the caption with the data woven naturally into the story

Workflow 5: Multiple Hook Variations

The hook (opening line) of a LinkedIn post determines whether anyone reads the rest. I always ask Claude for multiple options.

As a Claude prompt:

Give me 5 different hook options for this LinkedIn post about
[topic]. Each hook should use a different technique:

1. A surprising statistic
2. A contrarian opinion
3. A personal observation
4. A question that provokes curiosity
5. A bold prediction

Keep each hook under 15 words.

This is a standalone prompt you can use for any piece of content. I often start the content process here — pick the best hook, then build the post around it.

Workflow 6: Batch Content from Large Datasets

As a Claude prompt:

I have a dataset of 1,000+ interviews about AI use cases in marketing.
The data is at /Users/krishna/Desktop/Data/ai-marketing-interviews.csv

The columns include: Interviewee Role, Company Size, Use Case Description,
Tools Used, Impact Reported.

Create a curated compilation of the 20 most interesting and surprising
use cases. For each one, write a 2-3 sentence description that
highlights what's novel. Group them by theme.

Save as an RTF document.

What Claude does: 1. Reads the entire dataset 2. Identifies the most distinctive and interesting use cases 3. Groups them thematically 4. Writes concise descriptions for each 5. Produces a formatted document

Workflow 7: Writing from Self-Analysis — Using Your Own AI Usage as Content

This is a workflow that most people will never think of, but once you see it, it is obvious: use Claude Code to analyze your own Claude Code conversation history, identify behavioral patterns, and write content grounded in real data about how you work.

As a Claude prompt (Step 1 — analyze your patterns):

Read my Claude Code conversation history and session logs.
Analyze how I actually use this tool:
- What categories of tasks do I perform?
- What prompting patterns do I use consistently?
- How do I iterate and give feedback?
- What tools and plugins do I use most?
- What's my workflow when starting a new project vs.
  continuing an existing one?

Produce a detailed analysis with specific examples from my
actual sessions.

As a Claude prompt (Step 2 — extract a style guide from existing writing):

I'm sharing two of my published essays:
/Users/krishna/Desktop/Content/essay-1.md
/Users/krishna/Desktop/Content/essay-2.md

Analyze my writing style in detail:
- Sentence structure and length patterns
- How I use analogies and examples
- Paragraph rhythm and section transitions
- Opening and closing techniques
- Where I use data vs. narrative vs. opinion
- My distinctive verbal fingerprints

Save this as a writing style guide I can reference for future
content.

As a Claude prompt (Step 3 — write the piece):

Using the behavioral analysis from Step 1 and the style guide
from Step 2, write a long-form essay about [topic].

Ground the essay in real behavioral patterns from my usage data —
not hypothetical claims. The essay should feel like it emerges
from lived experience, because it does.

Follow the style guide precisely. Save as RTF.

What Claude does: 1. Reads actual session logs and conversation history 2. Identifies real patterns (not speculative ones — patterns backed by evidence) 3. Analyzes existing writing to extract stylistic fingerprints 4. Writes a piece that marries the analytical findings with the author’s voice

Real example: I used this exact workflow to write a long-form essay about what it takes to be a future-ready employee in the AI era. Instead of theorizing, I had Claude analyze hundreds of my own Claude Code sessions to identify real behavioral patterns — how I prompt, how I iterate, how I chain tasks. It then analyzed two of my existing published essays to capture my writing style. The resulting essay was grounded in actual data about how a non-technical professional uses AI, not abstract claims. The patterns Claude surfaced — things like “contextual generosity” (front-loading context rather than prompting iteratively) and “infrastructure mindset” (building reusable systems rather than one-off prompts) — became the backbone of the piece.

Why this is powerful: Most AI-generated content about “how to use AI” is generic because it lacks real data. This workflow inverts that. Your own usage history becomes the primary source material, making the content authentic and specific in a way that no amount of prompting from scratch can achieve.

Workflow 8: Building a Personalized Writing Skill from Your Own Content

This is the workflow that takes everything in this chapter and turns it into a permanent, reusable system. Instead of re-analyzing your writing style every session, you build a Claude Code skill that encodes your patterns — so every future writing task automatically applies your voice, structure, and quality standards.

The process:

Step 1 — Gather your corpus. Collect your best-performing content. For LinkedIn, this means your posts with the highest engagement. For newsletters or blog posts, your most-read pieces. Aim for at least 30-50 pieces to give Claude a meaningful dataset. Put them in a single folder.

As a Claude prompt (Step 1 — gather your corpus):

Read all the posts in /Users/krishna/Desktop/Content/best-posts/

These are my highest-performing LinkedIn posts. Analyze the
entire corpus and identify:
- What separates high-performers from average ones
- Structural patterns (hook formulas, paragraph rhythm, closing)
- Language patterns (first person vs. second person, sentence
  length, vocabulary choices)
- Content patterns (what topics resonate, how data is used,
  what angles work)
- Formatting patterns (bold usage, line breaks, emoji usage
  or absence)

Create a comprehensive content playbook based on this analysis.
Save it as a markdown file.

Step 2 — Extract the framework. Claude will identify patterns you were not even conscious of. In my case, analyzing 76 LinkedIn posts revealed a clear 2x2 matrix: posts that were specific (named brands, real numbers) and personal (first-person experience) dramatically outperformed posts that were generic or distant. Claude also identified seven specific “underperformance traps” — patterns that consistently correlated with lower engagement — and six “viral triggers” that appeared in top-performing content.

Step 3 — Codify it as a skill. Take the analysis and turn it into a skill file — a set of instructions Claude follows whenever you invoke the skill.

As a Claude prompt (Step 3 — codify as a skill):

Based on the content playbook you just created, build a Claude
Code skill that:
1. Diagnoses any draft I give it (identifies which quadrant it
   falls in, which traps it hits, which triggers are available)
2. Reports the diagnosis in 3-5 lines (not a lecture)
3. Rewrites the draft applying all the patterns we identified
4. Presents two versions: an edited version (close to my
   original) and a fully rewritten version (optimized for
   maximum performance)

Save it as a skill file I can install.

Step 4 — Install and use. The skill becomes a permanent part of your Claude Code setup. Every time you write a LinkedIn post, blog entry, or any piece of content, you invoke the skill and it applies your entire analytical framework automatically.

What makes this transformative: You have turned tacit knowledge — the intuition you have built over years of writing — into explicit, repeatable instructions. The skill does not just “write like you.” It diagnoses drafts against a data-backed framework derived from your actual performance metrics. It catches the traps you fall into unconsciously. It surfaces the triggers you use instinctively in your best work but forget in your average work.

The meta-insight: This workflow is not limited to writing. The same pattern — gather your best work, analyze what makes it work, codify the patterns into a skill — applies to any repeatable creative task. Presentation design. Email outreach. Research briefs. Any domain where you have a corpus of past work and a measurable quality signal.

Workflow 9: Articles to LinkedIn Carousels — The Batch Visual Content Pipeline

This workflow turns long-form articles into branded LinkedIn carousel PDFs — the kind you would normally commission a designer for. No Figma. No Canva. Just HTML, CSS, and Cmd+P.

The setup: You need a folder containing the articles you want to convert (PDFs, markdown, or pasted text) and your brand guidelines (color hex codes, logo files, font preferences). Drop everything into a single project folder.

As a Claude prompt:

Read all the articles in /Users/krishna/Desktop/Carousel-Project/Articles/

These are blog posts from [brand/publication]. For each article,
create a 5-slide LinkedIn carousel:

Slide 1: Title slide with a hook headline that captures the
core thesis
Slides 2-4: Key arguments, frameworks, or data points — one
idea per slide, concise and visual
Slide 5: CTA slide driving readers to the full article

Brand guidelines:
- Colors: [dark background hex] for title/end slides,
  [light background hex] for content slides, [accent hex]
  for highlights
- Fonts: [heading font] for titles (uppercase, normal weight),
  [body font] for content
- Logo: /Users/krishna/Desktop/Carousel-Project/Assets/logo.png
  (use white version on dark slides)

Build each carousel as a standalone HTML file with a shared
styles.css. Slides should be 1080x1080px (square, LinkedIn
optimized). Include CSS print styles so each slide becomes its
own page when I use Cmd+P to save as PDF.

Also create a LinkedIn_Captions.md file with a ready-to-paste
caption for each carousel post.

What Claude does: 1. Reads each article and distills it into five slides — extracting the core thesis, key arguments, and a compelling hook 2. Creates a shared styles.css implementing your brand guidelines with CSS custom properties 3. Builds one HTML file per carousel, each linking to the shared stylesheet 4. Adds CSS print rules (@page { size: 1080px 1080px; margin: 0; }, page-break-after: always, -webkit-print-color-adjust: exact) that make PDF export a one-click operation 5. Writes LinkedIn captions for every carousel

The PDF export (the magic step): 1. Open any .html file in Chrome 2. Press Cmd+P 3. Set destination to “Save as PDF” 4. Check “Background Graphics” 5. Set margins to “None” 6. Save — each slide becomes its own PDF page

That is it. A multi-page branded carousel PDF, ready to upload to LinkedIn as a document post.

Why this workflow is remarkable:

First, it is batch production. The shared CSS means the same design system powers every carousel. Adding a new article is just creating another HTML file — Claude can produce five carousels in the time it takes to brief a designer on one.

Second, the content distillation is genuinely valuable. Claude reads a 3,000-word article and decides which three ideas deserve a slide, which data points to highlight, and how to frame the hook. This is editorial judgment, not just formatting.

Third, the three export paths give you flexibility. Cmd+P for instant PDFs. The Figma MCP plugin to push HTML slides into Figma as editable native layers (useful if a designer wants to refine them). Playwright for individual slide PNGs if you need them as separate images.

Real example: I used this workflow to convert five long-form blog posts from an investment firm’s AI thought leadership series into branded LinkedIn carousels. Each post was 2,000-4,000 words. Claude distilled each into five slides, created the design system from the brand’s color palette and fonts, generated all five HTML carousel files plus LinkedIn captions — all in a single session. The total output: five branded carousel PDFs, five LinkedIn captions, and a reusable template for future articles. What would have taken a designer a full day was done in minutes. And once the design system and shared CSS exist, each additional article becomes a near-instant conversion — Claude just generates a new HTML file using the established template.

Pro tip: Font strategy matters. If you plan to export carousels to Figma for further refinement, use Google Fonts (like Bebas Neue + Inter) instead of proprietary brand fonts. Google Fonts render identically in browsers and Figma, avoiding font substitution issues. You can always swap to the exact brand fonts in the final Figma file.

Workflow 10: The Publish Gate — De-AI Your Writing

I am aware of the irony here. A book written partly with AI is about to tell you to strip the AI out of your writing. It is still the best closing discipline in this chapter.

Wikipedia’s editors maintain a documented list of the signs of AI writing: the em-dash pileups, the rule-of-three sentences (“faster, smarter, better”), the “delve”-class vocabulary, symbolism inflation (“a testament to”), vague attributions (“experts say”). The humanizer skill is built directly on that list. It reads a draft, finds the tells, and rewrites them out.

Install it once:

In your terminal:

git clone https://github.com/blader/humanizer.git ~/.claude/skills/humanizer

Then, inside Claude Code, ask it to “humanize” the draft — or invoke the skill directly.

The placement matters more than the tool. Run it as the last step of every writing pipeline: draft with Claude, edit yourself, then let the humanizer strip the tells before anything ships. I run it as a standing final pass on everything I publish.

Think of it as the publish gate. Nothing goes out without clearing it — and your readers will never quite know why your writing does not read like everyone else’s.

Pro Tips

Tip 1: Always save writing outputs as RTF, not markdown. RTF preserves formatting (bold, italics, bullet points) when you paste into LinkedIn, Google Docs, Notion, or email. Markdown formatting characters will clutter your text if you paste into non-markdown environments.

Tip 2: Provide “anti-examples” along with examples. When teaching Claude your style, also show what you do NOT want. “Do not use exclamation points. Do not start with ‘In today’s fast-paced world.’ Do not use the word ‘leverage.’”

Tip 3: Keep a style reference file. After Claude analyzes your writing style, save the analysis as a file (e.g., my-writing-style.md) and reference it in future prompts: “Follow the style guidelines in /Users/krishna/Desktop/Content/my-writing-style.md.” This saves you from re-analyzing every time.

Tip 4: Use the AskUserQuestion workflow for long content. For long pieces (reports, multi-section documents), instruct Claude to present each section for your approval before moving on. This prevents Claude from writing an entire document that needs extensive revisions.

Tip 5: Draft in Claude Code, polish in your head. Claude produces excellent first drafts — typically 80-90% of the way to your final version. The last 10-20% is where your unique voice, timing, and judgment matter most. Think of Claude as getting you to the editing stage in minutes instead of hours.

Tip 6: /copy beats mouse selection. When you want Claude’s last response on your clipboard — a draft, a caption, a rewrite — type /copy. No dragging the cursor across terminal text, no missed lines. Draft, /copy, paste into LinkedIn. For anything with formatting worth keeping, the RTF route from Tip 1 is still better; /copy is for speed.

Tip 7: /export is your session archive. Type /export and Claude saves the whole conversation as a text file. I use it when a writing session produced thinking worth keeping — the discarded hooks, the reasoning behind an edit — not just the final draft. The transcript becomes raw material for future content (see Workflow 7).

What Others Are Doing with Content Creation

Several high-engagement content creation workflows have emerged among Claude Code users:

The “full creative team” approach. Claude Code as an entire creative stack: research, brief generation, asset creation, and publishing — all in one session. The workflow layers skills and hooks so that Claude moves from research to draft to final asset without manual handoffs. For marketers who currently coordinate between researchers, writers, and designers, this collapses the pipeline.

30 days of content in 2 hours. An comprehensive prompt system for batch content creation: generate topic ideas, write captions, design supporting visuals, and schedule posts — producing a month’s worth of social media content in a single extended session. The key insight is batching: instead of creating one post at a time, you front-load the creative decisions (topics, angles, formats) and then mass-produce the content.

Outbound campaign personalization. Claude Code for personalized outreach: research your ideal customer profile, identify targeting signals, and generate personalized emails and LinkedIn messages at scale. The workflow: provide Claude with your ICP definition and a list of prospects, and it researches each prospect, identifies relevant signals, and generates personalized messages.

GTM infrastructure as code. A newer pattern from the community uses CLAUDE.md to store your brand voice, ideal customer profile, and messaging framework. Every content task Claude performs automatically adheres to these parameters. Instead of re-specifying your brand voice in every prompt, it is embedded in the project memory — Claude produces on-brand content by default.