Six Months of AI-Assisted Writing: Real Data and Hard Lessons

I run a technology newsletter with about 8,000 subscribers. For the first year, I wrote everything myself—research, drafting, editing, the whole thing. The content was good, but I was spending 15-20 hours per week on it, which started feeling unsustainable as my day job got busier.

Six months ago, I started experimenting with AI-assisted writing. The goal was simple: maintain quality while reducing time investment. Six months and roughly 150 articles later, I have real data to share and some lessons I wish someone had told me at the start.

## The Data First

Here’s what actually changed:

Metric	Before AI	After AI
Weekly output	2-3 articles	5-7 articles
Time per article	3-4 hours	1-1.5 hours
Reader engagement	Baseline	+12%
Comments/interaction	Baseline	-10%
Unsubscribe rate	Baseline	+4%
"Helpful" ratings	Baseline	-8%

The numbers tell a complicated story. I was faster and produced more. But engagement metrics went down slightly, and unsubscribes went up. Let me explain what I think is happening.

## Why Output Improved

**More content, more coverage**

Before AI, I was selective about what I wrote about because every article cost me significant time. I stuck to topics I was confident would perform well. With AI assistance, I could write about topics that seemed interesting but weren’t sure-fire hits.

Some of those “maybe” topics became surprise hits. More importantly, the wider coverage built a reputation for being a comprehensive resource. Readers started seeing me as more than just “that guy who writes about AI.”

**More time for refinement**

When writing took 3-4 hours, I was often rushing to meet my publishing schedule. With AI handling first drafts, I could spend the same total time but focus it on improvement—better structure, more thorough research, clearer writing.

The articles that got the most positive feedback were the ones where I used AI for drafting but spent significant time revising. The ones that got negative feedback were almost always the ones I rushed through with minimal editing.

## Why Engagement Dropped

**The writing felt flatter**

I knew this was a risk going in, but seeing it happen still surprised me.

My original writing had personality. I had strong opinions, made jokes that only I would make, referenced personal experiences that made readers feel like they knew me. AI-generated content, even when I edited it heavily, lost some of that character.

One longtime subscriber sent me a private message that hit hard: “Your recent articles feel like they were written by a professional content machine. The information is good, but I used to feel like I was getting your perspective, not just information about a topic.”

Ouch. But fair.

**The quality inconsistency problem**

Early on, I got overconfident. AI drafts looked so good that I started publishing with minimal review. That led to several articles with factual errors or awkward passages that slipped through.

Each error eroded trust a little. Readers who caught mistakes started questioning everything I wrote. This is hard to recover from.

**I became a quantity-over-quality person**

This one is on me, not the AI. Once I had the capacity to produce more, I started doing exactly that—even when quality should have come first. Some weeks I published every day, and the content suffered.

The best-performing articles from the past six months were the ones I spent the most time on, even with AI assistance. The worst were the ones I rushed through.

## The Mistakes I Made

**Trusting AI output without verification**

AI still hallucinates. It still makes up statistics, misattributes quotes, and confidently presents incorrect information. I’ve had to add a mandatory fact-check step for any specific claims or data points.

Now I never publish AI-generated content that contains statistics or claims without independently verifying them. This takes time, but it’s non-negotiable.

**Losing my writing voice**

I let AI homogenize my style. Each article started feeling like every other article, because AI was optimizing for generic “good writing” rather than my specific voice.

I’ve since created a detailed style guide for AI tools, emphasizing my preferences for sentence length, tone, the kinds of examples I use, and the humor I’m willing to include. The output is better now, but it took conscious effort to get there.

**Forgetting that AI should assist, not replace**

Early on, I fell into the trap of treating AI as a replacement for my thinking. I’d generate content and publish it without adding my own perspective. The result was competent but soulless.

Now my rule is: every article must contain something that only I could have written—personal experience, unique perspective, original analysis. AI handles the framework and supporting content. I provide the soul.

## My Current Workflow

After six months of experimentation, here’s how I use AI now:

**AI handles:**
– Initial research and source gathering
– First draft structure and framework
– Grammar, spelling, and readability improvements
– Generating variations and alternatives
– Filling in factual details after verification

**I handle:**
– Topic selection based on what I find genuinely interesting
– Core thesis and unique perspective
– Personal stories and experiences
– Final quality review and fact-checking
– Anything that requires current information or specialized knowledge

**The actual process:**
1. Identify a topic I find genuinely interesting (me)
2. Research the basics to form my own initial understanding (me)
3. Ask AI for a structured outline and supporting points (AI)
4. Write the core argument and personal perspective (me)
5. Use AI to draft supporting sections and fill gaps (AI)
6. Edit heavily, adding my voice throughout (me)
7. Fact-check all specific claims (me)
8. Final AI pass for readability and grammar (AI)

The key is that AI never produces final output. Everything it generates gets filtered through my judgment and revised to match my voice.

## What I’d Tell Someone Starting Out

If you’re considering AI-assisted writing, here are the things I wish I’d known:

**1. Start with quality, not quantity**

I went quantity-first and paid for it with engagement drops. You don’t have to make the same mistake. Start by using AI to improve quality, not just increase output.

**2. AI generates content; you generate value**

Readers subscribe because they find value in your unique perspective. AI can help you deliver that perspective more efficiently, but it can’t replace it. Every article should contain something only you could provide.

**3. Fact-checking is non-negotiable**

AI makes things up. Not always, but enough that you can’t skip verification. Build fact-checking into your process from the start.

**4. Your voice is your brand**

Consciously protect your writing style. Use detailed prompts, edit aggressively, and don’t let AI homogenize your content into generic “good writing.”

**5. Pay attention to feedback**

I tracked my metrics religiously and noticed engagement declining. If I hadn’t been watching, I might have gone months before realizing something was wrong.

—

## Is AI-Assisted Writing Worth It?

For me, yes. The time savings are real—I’m now spending about 8-10 hours per week on the newsletter instead of 15-20. That’s four to six hours I can spend elsewhere.

But I had to learn the hard way that efficiency gains come with quality risks. The readers who matter are the ones looking for genuine value, not just content volume.

If you’re thoughtful about how you use AI—starting with quality, protecting your voice, maintaining rigorous fact-checking—it can be a powerful tool. If you just want to publish more with less effort, you’ll probably end up with worse results than writing less with more care.

That’s my honest assessment after six months. Questions or experiences to share? I’d love to hear how others are navigating this.