Stop using AI to write your emails UNLESS

Your readers know.

They can tell when you fed last week's news into ChatGPT and hit send.

The phrasing is off.

The structure is predictable.

The voice sounds like every other "AI-assisted" newsletter clogging their inbox.

You're saving time, but you're training your list to ignore you.

The issue isn't that you're using AI.

The issue is you're using it wrong.

Most people think AI newsletters mean one thing: paste a prompt into ChatGPT, copy the output, and schedule it.

That's not a system.

Here's the contrarian take: AI shouldn’t replace your thinking process - it should replace the boring manual work.

It can do that, but only if you use multiple models in sequence, each doing what it does best, with human gates at the right points.

I run this exact system for clients.

A B2B expert comes in with 50+ blog posts, 200 tweets, 10 LinkedIn articles.

We need a weekly newsletter that sounds unmistakably like them, not like a language model.

Step 1: I drop their name, URL, content samples (pasted text), and industry into a Google Sheet.

Step 2: ChatGPT analyzes everything and outputs a 3,000–5,000 word voice document. Voice profile. Vocabulary lists. Sentence structure patterns. Example snippets. Contraindications (words they never use). This is the foundation.

Step 3: That document goes to Claude, which extracts the intel and creates an ultra-specific brief: outline, topic ideas, audience pain points, structural preferences.

Step 4: Claude writes the full newsletter draft using that brief. It exports to a Google Doc.

Step 5: Human intervention - this is the step where I polish, edit the formatting, make voice tweaks, and tighten up the structure. The draft is 80% there. I'm finishing the last 20% that only a human should touch.

The whole system runs in Make.

Setup takes time.

Production per issue takes minutes.

Remember - one model can't do this.

Custom GPTs can't do this.

You need task separation, provider selection, and layered context.

Step 1: Map your content sources

Pull 10–20 representative samples (emails, posts, articles).

Paste them into a structured sheet with metadata: client name, URL, industry, content type.

This is your raw material.

Step 2: Use ChatGPT to build the voice profile

Feed it everything.

Prompt it to extract voice patterns, vocabulary, sentence rhythm, contraindications, and tone markers.

Export a 3,000+ word document.

This becomes your voice bible.

Step 3: Use Claude to create the execution brief

Pass the voice document to Claude.

Have it synthesize into a tight brief: outline format, topic suggestions, audience context, and structural rules.

This is your per-issue instruction set.

Step 4: Use Claude to draft, then polish manually

Claude writes the newsletter using the brief.

You review for formatting, voice fidelity, tone, and structure.

You're not writing from scratch.

You're editing for the 20% that matters.

Step 5: Orchestrate with Make (or equivalent)

Automate handoffs between models.

Sheet → ChatGPT → Claude → Google Doc.

Human review happens at the end, not scattered throughout.

Google Sheets — Input layer for content samples, metadata, and client context.

ChatGPT (via API or Make) — Summarization and voice profiling. Cheaper and faster for pattern recognition.

Claude (via API or Make) — Drafting and long-context synthesis. Better instruction-following and tone matching.

Make — Orchestration layer. Connects models, manages handoffs, and routes outputs to Google Docs.

The stack is simple.

The sequencing is what matters.

Mistake 1: Using Custom GPTs for everything

Custom GPTs can't handle task separation.

You're asking one model to profile voice, generate topics, write drafts, and self-edit.

It fails at all four because it's optimized for none of them.

Mistake 2: Skipping the voice profiling step

If you go straight to "write a newsletter about X," the output is generic.

Voice profiling front-loads the context so every draft starts 80% of the way to your actual voice.

Mistake 3: No human review gate

AI drafts need human polish on formatting, voice fidelity, and structural flow.

If you skip this step, your readers will know.

Every time.

Try it and let me know how it goes.

Keep dominating,

Tanyo