How to Stop Claude Writing Like an AI

You can tell when something’s been written by AI. You might not be able to pinpoint exactly why, but you feel it. The text is technically fine. Grammatically correct. Well-structured. And completely lifeless.

I use Claude every day to help produce content, from social media posts to newsletters to training materials. And without the right guardrails, it writes like every other AI: a flood of “pivotal moments,” “broader landscapes,” and sentences that sound profound but say absolutely nothing.

The good news is you can fix this in about two minutes.

Why AI writing sounds like AI writing

Wikipedia’s editors got so fed up with AI-generated text being pasted into articles that they built a detailed guide cataloguing the patterns. It’s called “Signs of AI Writing,” and it’s one of the most useful things I’ve read on this topic, because it’s based on thousands of real examples rather than speculation.

The patterns they identified fall into a few categories.

The vocabulary. AI models lean on a small rotation of words that sound impressive but add nothing. “Delve,” “underscore,” “pivotal,” “robust,” “seamless.” Research from the Max Planck Institute found that words like these have spiked in usage by over 50% in published writing since ChatGPT launched, and not just in AI-generated text. Humans have started copying the AI’s vocabulary because it feels safe and professional. That’s how pervasive this has become.

The sentence structures. AI loves a particular kind of false contrast: “It’s not just X, it’s Y.” Or: “This isn’t about X. It’s about Y.” These constructions mimic the shape of insight without actually containing any. Wikipedia editors flag them constantly.

The throat-clearing. Every other response starts with “In today’s rapidly evolving digital landscape…” or “It’s worth noting that…” These are filler phrases. They exist to pad out the text and make it sound authoritative, but they actually make it sound less human, because real people don’t talk like that.

The uniform structure. AI produces paragraphs of nearly identical length, sentences of nearly identical rhythm, and lists where every item follows a “Bold term: explanation” format. The result is text that feels like it was extruded from a machine, because it was.

The em dashes. This one surprised me. AI massively overuses em dashes, often multiple times per paragraph, in places where a comma or parenthesis would be more natural. It’s become so notorious that some AI companies have started suppressing them in newer models.

How to fix it in Claude

Claude has a feature called Custom Instructions (you’ll find it in Settings > Profile). Whatever you put there gets applied to every conversation. It’s the single most effective way to change how Claude writes, because you set it once and it shapes everything from that point on.

I’ve put together a set of instructions based on Wikipedia’s research, supplemented by analysis from GPTZero, Grammarly, and my own experience of producing hundreds of pieces of content with Claude over the past year.

Here’s what to do:

1. Open Claude (claude.ai or the app)
2. Go to Settings
3. See the Profile section
4. Paste in the text below
5. Hit ‘Save changes’

That’s it. Every conversation from this point will follow these rules.

The custom instructions (copy and paste this)

## Voice

Be direct. Have opinions. Use specific examples and names, not vague claims. State your point first, then support it. Trust the reader to recognise what matters without labelling it as "significant" or "important."

## Banned words

Never use these — they are the most flagged AI-writing markers:

delve, dive into, navigate (figurative), underscore, bolster, foster, harness, leverage, unpack, shed light on, pave the way, pivotal, groundbreaking, cutting-edge, transformative, game-changing, innovative, robust, comprehensive, seamless, intricate, nuanced (as empty praise), vibrant, multifaceted, holistic, testament, landscape (figurative), realm

Never use these phrases:

- "In today's [fast-paced/rapidly evolving/digital] world..."
- "It's important/worth noting that..."
- "One of the most [important/significant/crucial]..."
- "When it comes to..." / "At its core..." / "At the end of the day..."
- "This is where X comes in" / "Let's break it down"
- "Plays a crucial role in..." / "It cannot be overstated..."
- "...underscoring the importance of..." / "...highlighting the need for..."
- "...reflecting a broader trend toward..." / "...marking a significant shift in..."

Never use these structures:

- "It's not just X — it's Y"
- "Not only X, but Y"
- "This isn't about X. It's about Y."
- "No X. No Y. Just Z."

These mimic insight without providing any.

## Structure

- Vary paragraph and sentence length. Don't write uniform blocks.
- Never use the "Bold term: explanation sentence" list format. It's the single most recognisable AI pattern.
- Don't signpost ("Let's explore," "Now let's turn to"). Just make your point.
- Don't open with a sweeping contextual statement. Don't close with a summary or inspirational wrap-up. Start and end on substance.
- Don't restate the question back before answering it.

## Style

- Use contractions. "It's," "don't," "won't."
- Maximum one em dash per response. Use commas or parentheses instead.
- Don't over-format. Plain prose is often clearer than headers and bullet points.
- Drop preamble ("Great question!"), performative enthusiasm ("exciting," "incredible," "powerful"), and unsolicited caveats.
- Match tone to context. Casual question, casual answer.

## Before finishing, check:

1. Read it out loud. Does any sentence sound like a press release? Rewrite it.
2. Are you repeating the same point in different words? Say it once.
3. Does your opening sentence set the scene with a grand statement about the state of the world? Delete it, start with the second sentence.

A few things worth adding on top

The instructions above are a general-purpose fix. They’ll make Claude sound dramatically more human regardless of what you’re writing. But if you want to go further, add a short paragraph above them describing your own voice. Something like:

“I write in a warm but direct tone. I’m knowledgeable without being showy. I use British English. I prefer short sentences and plain language over jargon.”

Even two or three sentences of personal voice description will make a noticeable difference. Claude is good at adapting to a described style, but it needs something to adapt to. Without it, you get the default, and the default is that polished, lifeless, AI-sounding prose that everyone’s learned to recognise.

The other thing I’d recommend: if you use Claude for a specific type of content regularly (social media posts, emails, blog articles), build a Custom Style for that format. Custom Instructions set the baseline. Styles let you dial it in further for specific contexts.

Does it actually work?

Yes. Noticeably. The banned words list alone makes a significant difference, because so much of what makes AI text feel “off” comes down to vocabulary choices. When you remove the option to reach for “pivotal” or “foster” or “it’s worth noting,” Claude is forced to find more natural, specific ways to express the same ideas. And it does.

The structural rules matter too. Once you stop Claude from producing those uniform “Bold term: explanation” lists and “In today’s fast-paced world” openings, the output starts to read like something a person actually wrote.

It’s not perfect. You’ll still want to edit. But you’ll be editing for substance and accuracy rather than spending your time stripping out AI clichés, which is a much better use of your time.