Using Claude on my Blog

I've been writing on this blog for a while now, and like a lot of developers, I have a backlog of half-finished drafts and post ideas scrawled in random Markdown files. Last month I decided to stop pretending I'd magically find the time to polish them all by hand, and I started using Claude as part of my writing workflow. Here's how that's gone.

Why Claude, specifically?

I've tried a few different LLMs for writing assistance, and Claude consistently feels the most "writerly" to me. It tends to:

• Keep a consistent voice across long outputs.
• Push back when I ask for something vague.
• Avoid the over-bulleted, over-bolded "AI listicle" style unless I explicitly ask for it.

For a personal blog where tone matters, that's a big deal. I don't want every post to read like a LinkedIn announcement.

The workflow I landed on

After some experimenting, I settled on a pretty simple loop. It's not magic, and it's intentionally not "press a button, get a post."

1. I write the seed myself

Every post still starts with me. Usually it's a title, a few bullet points, and a paragraph of raw thoughts — the kind of thing I'd jot down before I forgot the idea. Claude is good, but it can't generate opinions I haven't had yet.

2. Claude does the structural pass

I hand the seed to Claude and ask it to suggest an outline: sections, subheadings, and what each part should cover. This is where it earns its keep. I'm a developer, not an editor, and I tend to either ramble or be too terse. Claude is decent at finding the middle.

3. I draft, Claude critiques

Here's the part that surprised me: I get more value from Claude as an editor than as a writer. I'll write a rough draft, paste it in, and ask things like:

• "Where does this drag?"
• "Which sentences sound generic?"
• "What would a skeptical reader push back on?"

The answers are often uncomfortably accurate.

4. Final polish stays human

I don't let Claude write the final version. Once the structure and arguments are solid, I rewrite the prose myself. This is the step that keeps the blog feeling like mine and not a generic content farm. It also keeps me from accidentally publishing one of those tell-tale phrases ("In today's fast-paced world…") that scream LLM.

What I've learned

A few things I didn't expect going in:

Prompts are throwaway, principles are not. I stopped trying to craft the perfect mega-prompt. Instead, I keep a short note of how I want my posts to read — voice, structure, things to avoid — and paste it in as context when I need it.

It's better at "less" than "more." Claude is great at trimming, sharpening, and reorganizing. It's noticeably worse at inventing fresh insight. So I lean on it for compression, not creation.

Disclosure feels right. I'm not going to slap an "AI-assisted" badge on every post, but I also won't pretend I wrote 100% of every word from scratch. If a post is mostly Claude's prose, I'll say so. If it's mine with editing help, I treat that the same way I'd treat a friend reading a draft — no disclosure needed.

Is it worth it?

For me, yes. I'm publishing more often, my drafts are tighter, and I'm spending less time staring at a blinking cursor. The trade-off is that I have to be more disciplined about keeping my own voice in the loop, because it's genuinely easy to let the model take over.

If you're sitting on a pile of unfinished drafts like I was, give it a shot. Just don't let it write the parts that make the blog yours.