How to Get AI to Not Sound Like a Robot

Step 1: Set up your project

Create a Claude Project for each client (or for your own brand if you're writing for yourself). This is where you'll store everything Claude needs to write consistently and well — think of it like onboarding a new team member.

Add all of this to the project knowledge before you start working:

→ Brand voice guide — tone, personality, words they use and avoid → Style guide — grammar, punctuation, formatting preferences → Examples of past work — blogs, emails, or copy that represents the brand → Brief templates — anything you use regularly for that client → Your working preferences and directives — more on this below → Your AI writing habits list — the list of things you don't want it to do (see Step 2)

Once everything is uploaded, start a new chat inside the project and send this message:

Hey Claude, I'm starting a new project and I'd like to make sure you have everything you need before we begin. I'm going to give you some context about this project — please just confirm you've received and understood each item as I give it to you. Don't generate anything yet. I'll let you know when it's time to start creating. Got it?

Then feed it anything that isn't already in the project knowledge — specific briefs, new examples, context about a particular piece you're working on.

Step 2: Give it your AI writing habits list

Even with a project set up, it helps to explicitly remind Claude of the writing habits you want it to avoid. I keep a saved list that I paste in at the start of any new chat or project. Here's mine — feel free to steal it:

✅ Always do: → Use en dashes with spaces – like this – instead of em dashes

🚩 Never do:

→ Overuse filler words — especially "that," "just," "simply," and "very"

→ Generic openers — "In today's digital age…" / "Whether you're a small business or a large enterprise…" / "Now more than ever…"

→ Symmetrical sentence structures — "Not only does this boost productivity, but it also saves valuable time." Avoid this pattern unless it genuinely fits.

→ Over-explaining common knowledge — don't define things readers already know

→ Tone shifts — don't go from conversational to corporate mid-paragraph

→ Keyword stuffing — don't repeat phrases unnaturally to hit a keyword

→ Wishy-washy point of view — "There are many pros and cons to each approach." Take a stance.

→ Repeating the same idea in different words — say it once, say it well, move on

→ Forced transitions — "On the other hand…" / "In conclusion, this demonstrates…" — let thoughts flow naturally

→ Emoji — unless specifically asked

Step 3: Make it sound like the client (or you)

Now give Claude everything it needs to emulate the right voice. This is where your project knowledge does most of the heavy lifting — but for each new piece, it helps to confirm Claude has what it needs.

You can give it:

→ Brand voice guide — if the client doesn't have one, ask Claude to create one by analyzing their website and social media presence → Style guide — if the client didn't provide one, ask Claude to build one from their existing materials → Examples — treat it like a new hire; give it examples of what good looks like, whether that's from the client, their competitors, or just pieces you admire → Directives — describe your role and its role clearly. Something like:

"You are my writing assistant. I am the lead on this project. Do not get ahead of me, do not take over, and do not generate anything I haven't asked for. We will work through this together, step by step. Your job is to support my process — not replace it."

Step 4: Ask it what it needs

At the start of every project setup — and honestly at the start of every new chat within that project — ask Claude what else it needs to do the job well:

"Before we get started, what other context would be helpful for you to write this in [brand]'s voice? What questions do you have?"

Keep asking this until Claude confirms it has everything it needs. It will tell you when it does. This one extra step produces dramatically better first drafts.

Example of how I use this:

When I'm working on a content series — say, a batch of LinkedIn posts and newsletters for a client — there are a lot of moving pieces. Instead of just dumping everything on Claude and hoping for the best, I ask it what it needs before each piece. It'll give me 5–10 questions. I answer the ones that feel most relevant (doesn't have to be in full sentences), and the draft it produces after that is so much better than anything it generates cold.

Step 5: Correct throughout the project

Don't wait until the end to fix things. Correct Claude as you go — it genuinely gets better as the conversation develops. You can:

→ Ask it to rephrase something → Point it back to the brand voice guide if it drifted → Ask it why it made a specific choice if something feels off → Ask it to think more deeply about the audience and what they actually need → Ask for sources if it stated something you're not sure about → Give it new resources mid-project — you can always paste in blog posts, style guides, or examples to help it recalibrate

Step 1: Create a custom GPT or start a new chat

For each client, create a dedicated Custom GPT under "Explore GPTs" → "Create." Load it with everything the client-specific context it needs to write well consistently.

If you're not using a Custom GPT, starting a fresh chat works too — just know you'll need to re-brief it every time. Either way, send this message before you do anything else:

Hey Chat, I'm starting a new project and I'd like to give you some resources to guide our work together. First, I'm going to give you some guidelines of what TO do and what to AVOID. Then, I'll give you more info about the voice I want you to use, the style I need you to write in, and more info about your role in this project. Please don't try to generate anything yet. Just let me know you've received and understood each item as I give it to you. I'll let you know when it's time to begin. Got it?

Step 2: Give it your AI writing habits list

Feed ChatGPT this list every time you start a new chat or custom GPT. It's the fastest way to cut out the worst AI writing habits before you even begin:

✅ Always do: → Use en dashes with spaces – like this – instead of em dashes

🚩 Never do:

→ Overuse filler words — especially "that," "just," "simply," and "very"

→ Generic openers — "In today's digital age…" / "Whether you're a small business or a large enterprise…" / "Now more than ever…"

→ Symmetrical sentence structures — "Not only does this boost productivity, but it also saves valuable time." Avoid unless it genuinely fits.

→ Over-explaining common knowledge — don't define things readers already know

→ Tone shifts — don't go from conversational to corporate mid-paragraph

→ Keyword stuffing — don't repeat phrases unnaturally

→ Wishy-washy point of view — "There are many pros and cons." Take a stance.

→ Repeating the same idea in different words — say it once and move on

→ Forced transitions — "On the other hand…" / "In conclusion, this demonstrates…"

→ Emoji — unless specifically asked

Step 3: Make it sound like the client (or you)

Now give ChatGPT what it needs to write in the right voice:

→ Brand voice guide — if the client doesn't have one, ask ChatGPT to build one by analyzing their website and social presence → Style guide — if the client didn't provide one, ask ChatGPT to create one from their existing materials → Examples — give it examples of great work from the client, their competitors, or pieces you like; treat it like a new hire who needs to see what good looks like → Directives — tell it clearly what its role is:

"You are my writing assistant. I am the lead on this project. Do not get ahead of me and do not generate anything I haven't asked for. We work through this together, step by step."

Step 4: Ask it what it needs

Before starting any piece, ask ChatGPT:

"What questions do you have? What context would be helpful to write this in [brand]'s voice?"

It'll usually give you 5–10 questions. Answer the ones that resonate — even partial answers help. The drafts you get after this Q&A are noticeably better than anything it produces without it.

Step 5: Correct throughout the project

Correct as you go. Don't save it all for the end. You can:

→ Ask it to rephrase something that doesn't feel right → Point it back to the brand voice guide if it drifted → Ask it why it made a specific choice → Ask it to think more deeply about the audience → Ask for sources on anything you're not sure about → Feed it new resources mid-project — blog posts, examples, style notes — to help it recalibrate