After 100 AI-Written Articles, Here's the Workflow That Actually Works

Night N of editing. Finally stopped fighting with AI-generated prose.

Let me start with something embarrassing.

When I first started using AI for writing last year, I was naive beyond belief. I’d type “write a Medium article about remote work,” hit enter, and 30 seconds later a full draft appeared. My first thought: this industry is doomed, and here I am still typing with my own hands?

Then I actually read the thing. “In today’s fast-paced digital landscape…” Three paragraphs of perfectly correct nothing, and an ending that insisted on elevating everything into “the future of remote work.” I tried to fix it — and realized editing took longer than writing the article myself. Every sentence required me to judge whether it was filler, and whether anything real was hiding under the filler. (Usually nothing was.)

Over the past year-plus I’ve written more than 100 pieces with AI in every format you can think of: Medium articles, newsletters, blog posts, marketing copy, business emails, social media threads. I’ve eaten every kind of loss. Fluff. Copy-pasted claims that readers called out because the data was fabricated. Enough scars to know what actually works.

Here’s the workflow I landed on. I use it for everything now, and it’s made me two to three times faster than writing from scratch — with quality that doesn’t collapse. The real problem was never the AI. It was my method.

The Right Way to Think About AI Writing in 2026#

One sentence: don’t let AI write for you. Let it speed you up.

What AI is genuinely great at: taking things already in your head and expanding, restructuring, and fleshing them out. What it’s terrible at: living your life, making your judgment calls, holding actual opinions. Direction and quality have to stay in your hands. Hand over the steering wheel and it will drive straight into the ditch of “key insights and takeaways.”

Once that clicked, I stopped treating AI as an author. It’s a very fast typist who knows nothing about me. So you have to break the work into pieces and feed it properly.

Seven steps. Skip one and you’ll pay for it in rework.

My 7-Step Workflow#

Step 1: Nail the topic and angle (before you touch AI)#

Before writing anything, answer three questions: Who is this for? What problem does it solve for them? What makes my angle different from the other 500 articles on this? AI can’t do this step for you — only you know your readers. Once you’ve got an angle, pressure-test it:

I want to write about [topic] for [audience].
My core angle: [your take].
Stress-test it for me: what are the 3 biggest
weaknesses of this angle? What would a skeptical
reader push back on?

Step 2: Generate outlines — three of them#

Ask for three structurally different outlines in one go, pick the one you like, and cut it up by hand. This is measurably faster than generating one outline, hating it, and regenerating:

Based on this angle, give me 3 outlines with
different structures:
Version A: problem → analysis → solution
Version B: the 3 mistakes I made, one section each
Version C: contrarian hook, conclusion first
Titles and bullet points only. Max 6 sections each.

Step 3: Inject your own experience (the most important step)#

This is the soul of the whole workflow. AI has none of your real experiences. If you don’t feed them in, it will invent them. My move: once the outline is locked, I manually drop material under every section — the mistake I actually made, the actual numbers from that month, the actual email a reader sent me, the actual screenshot. Even one sentence helps: “The first time I built an automation with n8n, I burned an entire evening on one webhook.” Give AI that, and suddenly the writing has a pulse.

Step 4: Expand section by section (never the whole draft at once)#

One-shot the entire article and you get a beige, committee-approved essay. Go section by section with specific requirements each time:

Write ONLY section 2 now. ~200 words.
Material (must use): [paste your real story]
Requirements: first person, open with the specific
moment it went wrong for me. No summary sentences.
No "In today's..." filler. Conversational tone.

Step 5: Fact-check and verify every claim (non-negotiable)#

I learned this the expensive way. An article about newsletter performance quoted “average open rates of 45%” — plausible-sounding, completely wrong. Industry average is somewhere in the 20s. Now I have a hard rule: every number, quote, and case study that comes from the AI gets verified before publishing. This prompt makes the audit easy:

List every fact, number, and reference used in
this article. For each one, label it: from your
training data, or provided by me? Put all the
training-data ones in a separate list so I can
verify them.

Step 6: Human editing and polish (make it sound like you)#

At this point the piece is readable, but it still reeks of AI. This step is mine, and I focus on three things: voice (turning formal prose into how I’d actually say it), transitions (AI sections butt up against each other like train cars), and filler removal. Every “delve,” “landscape,” and “unlock” gets deleted on sight.

Step 7: Optimize per platform#

Same content, different skin for every channel. Medium wants depth. A newsletter should read like a letter to a friend. LinkedIn wants professional-but-personal. Twitter/X wants it short and punchy. This one actually IS AI’s strength:

Turn this article into a LinkedIn post:
keep the core argument and the real story,
under 250 words, hook in the first 2 lines,
end with an open question.

A Real Example, Start to Finish#

Take the Medium article on remote work productivity. Step 1: my angle wasn’t “remote work tips” (yawn) — it was “After 5 years working remotely, only 3 counterintuitive things actually moved my productivity.” Step 2: three outlines from AI, I picked the “3 mistakes” structure. Step 3: I stuffed each section with real material — the strange emptiness of my first two weeks after cutting meetings in half, the daily automation I built with Make, the time a Bangkok coffee shop’s Wi-Fi taught me to work offline.

Step 4: section by section, ~250 words each. Step 5: AI cited a Stanford remote-work study — it’s real, but one of the numbers was off, and I corrected it. Step 6: polish, cut two paragraphs of filler. Step 7: after publishing, I pulled the opening and turned it into a Twitter thread. Total: about 3 hours. Writing it from scratch would’ve cost me a full day.

3 Fatal Mistakes (All Mine)#

Mistake one: generating the full draft and shipping it. Every minute you “save,” you pay back double in editing. AI’s one-shot generation is an exercise in averaging — it serves you the most common version of the article on the internet.

Mistake two: skipping fact-checking. Readers don’t care that you used AI. But publish one wrong number and trust evaporates. Readers online are especially ruthless — someone will show up in your comments with a source.

Mistake three: zero personal opinions. Without your judgment, your article is a search engine summary. Even a minority take, even a slightly spicy one, beats “on one hand… on the other hand…” by a mile.

5 Editing Tricks That Make AI Writing Sound Human#

• Read it out loud. Anywhere you stumble, rewrite it into what your mouth would actually say. Dumbest trick, best results.
• Break the sentence rhythm. AI loves uniform long sentences. I insert a short one. Five words. The rhythm instantly feels human.
• Cut the first 20%. AI warms up too slowly. Whole opening paragraphs can go — the piece usually gets stronger at paragraph two.
• Add one detail only you could write. A specific place, a specific dollar amount, a specific disaster. Right is good. Specific is better.
• Run it through Hemingway Editor. It’s free. Split every highlighted monster sentence and watch readability jump.

What Changes by Content Type#

Medium articles: spend most of your time on steps 3 and 5. Depth and authenticity are the whole game — 1,500 honest words beat 3,000 padded ones. Newsletters: double down on step 6; it has to feel like a letter to a friend (mine holds a 40%+ open rate largely on voice alone). Marketing copy: in step 4, every section brief must name the target action, or AI defaults to writing a product manual. Business emails: you can skip most of the workflow — give full context, generate once, then cut it in half yourself. Emails are always shorter than you think. Social posts: step 7 is the main event; one long article becomes four or five posts, and the hook is life or death.

Final Thoughts#

AI is a tool, not a body double. After a year and a hundred-plus articles, I’m more convinced than ever: method beats model. Give the same AI to two people and one mass-produces sludge while the other amplifies their craft. The difference is whether they broke the process into the right steps — and whether they put something of themselves into it.

Don’t expect to win by lying flat. Keep your hands on the wheel, and AI becomes the assistant it was always supposed to be.

Every step above was paid for in tuition. If you’re writing with AI too, tell me in the comments — what’s the worst mistake it’s made on your watch?