My AI-Powered Writing Workflow: Step-by-Step Guide

People ask how I use AI in my writing work. Not theoretically—practically. What tools, in what order, with what prompts? Here's my actual workflow, refined over 18 months of daily use.

Overview: The Five Phases

My writing process has five distinct phases, and I use different AI tools (or no AI) for each:

1. Research: Understanding the topic deeply
2. Planning: Structuring the content
3. Drafting: Getting words on the page
4. Editing: Refining and polishing
5. Optimization: SEO and formatting

Let me walk through each phase with specific examples.

Phase 1: Research

Primary tool: Perplexity AI + Claude

Before I write anything, I need to understand the topic thoroughly. My research stack:

Perplexity for Quick Understanding

I start with Perplexity because it gives me sourced answers. "What are the current trends in email marketing automation?" returns a summary with links I can verify.

Key advantage: citations. Unlike ChatGPT, I can check where information came from. Essential for factual accuracy.

Claude for Deep Analysis

Once I have the basics, I use Claude to think through the topic more deeply. Claude handles nuance and complex reasoning better than alternatives in my experience.

This conversation often reveals angles I hadn't considered. Claude's tendency to think deeply before responding surfaces insights that quick searches miss.

Phase 2: Planning

Primary tool: Claude or ChatGPT

With research done, I structure the piece. Here's my actual outline prompt:

Notice how specific this is: word count, audience psychology, structural requirements, and what I want for each section. Vague prompts get vague outlines.

I rarely use the AI outline directly. It's a starting point. I rearrange, add sections, remove others. But having that structure accelerates my thinking.

Phase 3: Drafting

Primary tool: ChatGPT with my voice trained

Here's where it gets interesting. I don't ask AI to write the whole article. Instead, I write section by section, using AI to overcome blank-page paralysis.

The Section-by-Section Method

For each section in my outline, I either:

• Write it myself if I know exactly what to say
• Use AI to generate a draft if I'm stuck
• Use AI to expand bullet points into paragraphs

When I use AI for drafting, my prompt includes heavy context:

Custom Instructions Matter

My ChatGPT Custom Instructions include my voice preferences: conversational but not sloppy, specific examples over abstractions, skeptical of marketing truisms, preference for short paragraphs.

With these set, drafts already sound closer to me. Less editing required later.

Phase 4: Editing

Primary tool: My brain + Hemingway Editor + Grammarly

This is where AI takes a back seat. Editing is where my voice and judgment matter most.

First Pass: De-AI-ifying

Any AI-generated sections need human fingerprints. I read through specifically looking for:

• "Leverage" and "utilize" (replace with simple words)
• Lists of three adjectives (AI loves these)
• Overly balanced "on one hand... on the other hand" structures
• Generic transitions ("Furthermore," "Moreover")
• Anything that sounds too safe or hedge-y

I rewrite these in my own words. Sometimes entire paragraphs. The goal: someone who knows my writing shouldn't be able to tell which parts started as AI.

Second Pass: Clarity and Flow

Hemingway Editor catches long sentences and passive voice. Grammarly catches errors and awkward phrasing. Neither is AI-powered in the generative sense—they're just helpful tools.

AI for Specific Editing Tasks

I do use AI for specific editing tasks:

• "Suggest 5 stronger alternatives for this weak headline: [headline]"
• "This paragraph is too long. How might I break it up while maintaining flow?"
• "What's missing from this conclusion? What might readers still wonder?"

Phase 5: Optimization

Primary tool: ChatGPT for meta content

Once the article is polished, AI helps with optimization:

• Meta descriptions: "Write 3 meta description options (max 155 characters) for this article. Hook readers while including the keyword 'email automation'."
• Social snippets: "Write 3 Twitter-length summaries that would make someone want to click."
• Internal linking suggestions: "Given this article about email automation, what related topics should I link to?"

These are tasks where AI is genuinely faster and often better than me. I'm not emotionally attached to meta descriptions—let AI handle them.

Total Time Investment

For a 2000-word article:

• Research: 20-30 minutes
• Planning: 10-15 minutes
• Drafting: 45-60 minutes
• Editing: 30-45 minutes
• Optimization: 10-15 minutes

Total: About 2-2.5 hours

Before AI tools, the same article took 4-5 hours. The time savings are real, but more importantly: the quality is often higher because I spend less energy on grunt work and more on thinking.

What I Don't Use AI For

Equally important—what I never delegate to AI:

• Personal anecdotes (those must be real)
• Specific opinions and hot takes (those must be mine)
• Final judgment on whether something is good enough
• Understanding what the client/audience actually needs

AI is the engine. I'm still the driver.