How to Stop ChatGPT/Claude From Overusing Em-Dashes and Robotic Transition Buzzwords

If you have used ChatGPT, Claude, or Gemini to draft articles, technical docs, or emails, you have likely run into the dead giveaways of AI-generated text:

• The Overused Em-Dash (—): The AI drops an em-dash into almost every paragraph to append parenthetical context mid-thought.

• Robotic Transition Words: Every transition relies on words like “Furthermore,” “Moreover,” “In addition,” “Delve,” or “It’s worth noting.”

• The “Not X, but Y” Pattern: Sentences structured around contrasting points (e.g., “It’s not just a tool, but a complete transformation”).

• Syrup-Sweet Closings: Ending every informative response with “Let me know if you have any questions!” or cheerleading phrases.

When these patterns repeat, your writing instantly loses its human voice and signals to readers (and search engines) that it was spun out by a Large Language Model.

Here is why AI models fall into these habits—and the exact custom prompts and system settings you can use to strip them out for good.

Why LLMs Love Em-Dashes and Transition Words

Large Language Models generate text sequentially, predicting one token (word or sub-word) at a time. When an AI starts writing a sentence and realizes mid-stream that it needs to append extra context or qualify a statement, it cannot pause to rewrite the beginning of the sentence.

Instead, it picks the easiest bridge available:

1. The Em-Dash: Allows the model to insert an extra thought without breaking grammatical structure.

2. Formal Transitions: Words like “Furthermore” or “Moreover” act as high-probability connectors between independent paragraphs when the AI needs to maintain logical continuity.

What works fine as an occasional stylistic choice becomes an annoying habit when repeated across thousands of words.

The Solution: A Copy-Paste System Prompt

To stop this behavior, telling the AI “don’t write like an AI” or “humanize this” isn’t enough. LLMs respond much better to explicit negative rules combined with structural constraints.

Copy and paste this instruction block into your prompt, system prompt, or Custom Instructions setup:

[STYLE & PUNCTUATION RULES]
1. PUNCTUATION: Do not use em-dashes (—) or en-dashes (-). If you need to connect ideas, use standard sentence structures, coordinating conjunctions (and, but, so), or split thoughts into two separate sentences.
2. BANNED TRANSITIONS: Never use robotic transition words or AI filler, including: "Furthermore", "Moreover", "In addition", "Delve", "Testament", "Beacon", "Tapestry", "It's worth noting", "In conclusion", or "To sum up".
3. PATTERNS TO AVOID: Do not use the "Not X, but Y" sentence structure. Avoid excessive rhetorical questions.
4. SENTENCE STRUCTURE: Write direct, clean sentences. Limit independent clause stacking. Vary sentence length naturally.
5. TONE: Objective, human, and direct. Skip cheerful sign-offs (e.g., "Let me know if you have any questions").

How to Set This Up Permanently (No Re-typing)

Instead of pasting this block every time you start a new conversation, configure it directly in your platform settings.

1. In ChatGPT

1. Click your profile image in the bottom-left corner and select Personalization (or Custom Instructions).

2. Under “How would you like ChatGPT to respond?”, paste the style rule block above.

3. Click Save.

2. In Claude (Anthropic)

• Projects Workspace: If you use Claude Pro or Team, create a Project and paste the rules directly into the Project Instructions panel.

• Claude Code / Developer CLI: Add the style rules into your workspace CLAUDE.md file.

How to Fix Existing Drafts (Find and Replace)

If you already have a 2,000-word draft filled with em-dashes and AI buzzwords, you don’t need to rewrite it manually line by line.

Strategy A: The Second-Pass Prompt

Paste your existing draft into a new chat session with this command:

“Clean the following text without changing its core meaning. Remove all em-dashes (—) by replacing them with commas, periods, or conjunctions. Replace robotic transitions (like ‘Furthermore’ or ‘In addition’) with natural phrasing. Output clean ASCII text only.”

Strategy B: Quick Manual Find & Replace

If you prefer manual control inside Google Docs, Microsoft Word, or VS Code:

1. Press Ctrl + F (or Cmd + F) and search for —.

2. Apply one of three instant fixes:

   • The Period Fix: Replace the dash with a period and capitalize the next word.

   • The Conjunction Fix: Replace the dash with a comma followed by “and”, “but”, or “so”.

   • The Deletion Fix: Delete the parenthetical phrase between the dashes entirely—it’s usually non-essential filler.

By constraining sentence structure and explicitly banning robotic crutches, you can force AI models to generate text that reads cleanly, naturally, and authentically on the first try.

For a visual breakdown of how to update your AI preferences and enforce clean punctuation, check out this guide on stopping ChatGPT from using em dashes. This tutorial provides a quick walkthrough on adjusting settings so your drafts stay clean automatically.