Something Happened to Claude’s Writing!

Has Claude's writing been sucking for you lately?

Because it has for me.

And for a while, I genuinely thought we were doing something wrong.

We have writing rules.

We have context.

We’ve spent a lot of time defining how we want things to sound.

Yet lately I’ll ask Claude to write something, read the output, and think...

What happened?

The writing feels more generic.

More obviously AI.

Sometimes it ignores rules we've explicitly given it.

And recently it drafted an email for me that was just...

bad. 😂

Not “I would change a few sentences” bad.

I don't want to use this bad.

Which was particularly frustrating because writing was one of the reasons I started using Claude in the first place.

When Claude first came onto my radar, I was tired of the writing I was getting elsewhere.

Claude felt different.

The writing sounded better.

More natural.

More human.

And it became one of the reasons I kept coming back.

So when the quality started changing, my assumption was:

Something must be wrong with our system.

Maybe our instructions weren't clear enough.

Maybe our context wasn't right.

Maybe we needed to change the prompts.

You know how this goes.

When AI gives you bad output, the first thing we're taught to do is blame the prompt.

So that's where I looked.

Until I heard something that changed how I thought about the problem.

I was listening to an interview with one of Claude's engineers.

He was talking about the newer models and explained something I found fascinating.

According to him, they've been working on getting Claude to communicate effectively not only with humans, but also with other agents.

And apparently, striking that balance isn't easy.

The way a model needs to communicate with another AI agent isn't necessarily the way you or I want it communicating with us.

Suddenly, something clicked for me.

Because that's almost exactly how some of the writing had started to feel.

Like AI talking to AI.

Technically fine.

Structured.

Competent.

But not something I necessarily want to send to another human.

And hearing that gave me a little relief.

Maybe we hadn't suddenly forgotten how to work with Claude.

Maybe the tool itself had changed.

That led me to a bigger realization.

I think a lot of us are still treating AI like we're choosing a favorite sports team.

You're a Claude person.

Or a ChatGPT person.

Or a Gemini person.

And once you've built your workflows around one of them, switching feels like you're starting over.

But I'm increasingly convinced that's the wrong way to think about this.

The best AI for one job doesn't have to be the best AI for every job.

I still think Claude is fantastic for certain things.

Coding?

Reasoning?

Thinking through difficult problems?

I'm still using it.

But writing?

Right now, I'm experimenting.

And I'm looking beyond the big frontier models too.

One of the models I'm testing right now is DeepSeek 4.1 Flash.

And for some of our writing tasks, particularly blog posts, I've been really impressed with what we're getting.

In my own testing, it has repeatedly produced writing I prefer over what I've recently been getting from Claude.

And it's cheaper.

That's interesting to me.

Because it means the question isn't:

“What's the best AI?”

That's becoming almost meaningless.

The better question is:

“What's the best AI for THIS job?”

That's a very different way to build with AI.

And I think that's the lesson here.

Don't become loyal to the tool. Become loyal to the outcome.

If Claude is still producing fantastic writing for you?

Great.

Keep using it.

If it isn't?

You don't necessarily need another 37-line prompt trying to force it back into being something it currently isn't.

Test something else.

Compare the output.

Let the work decide.

That's what I'm doing.

Because AI is changing too quickly for me to assume the tool that was best six months ago is automatically the tool I should still be using today.

But now I'm really curious about your experience. 🤔

If you use Claude for writing, have you noticed the writing getting worse lately too?

For me, it was emails where I really started thinking:

Something is different here.

What about you?

Hit reply and tell me in one sentence.

Even something as simple as “Yep, I've noticed it too” or “Nope, Claude still works great for me” is perfect.

I genuinely want to know if you're seeing the same thing.

Want the full story? 👇

I recorded a video explaining what I noticed, what I heard that changed my understanding of the problem, and why I'm now testing other models for different writing jobs.

For the physicians reading this 👨‍⚕️

Our AiM (AI & Marketing) Rounds community has been growing, and the conversations are getting better because more doctors are sharing what they’re actually trying, questioning, and figuring out in real time.

We’re talking about practical AI, marketing, what’s working, what isn’t, and how all of this actually fits into a medical practice.

No pressure to be an expert. No need to have the perfect answer.

Just doctors learning from doctors.

If you want in, reply “JOIN AIM” and I’ll send you the WhatsApp link.

A few things worth checking out 👀

Just sharing a few things I've been exploring lately:

• 🤝 Let’s connect: I share more of my AI experiments, failures, and what we're learning on LinkedIn.

• 📘 Book I keep thinking about: The Delegation Trap — exploring what happens when leaders stop being the bottleneck.

• 🤖 What we’re building with AI: A look at what we’re working on with AI

• 📖 Currently reading: The book on my nightstand right now

• 🛠 Tool I’ve been using: GoHighLevel (GHL) — keeps the important stuff in one place.

Was this email helpful? 💬

I'd love your quick feedback.

❤️ Yep — I’ve noticed this too.
💛 Interesting — now I want to test other models.
💜 Claude is still working well for me.

Just hit reply with ❤️,💛, or 💜.

I read every response, and it helps me figure out what experiments and lessons are worth sharing next.

— Atiba