AI Newbie | 2026-07-14 | 9 min read
AI Writing Tells: Em Dashes Are Not the Real Problem
Everyone is joking about AI and em dashes, but the real problem is unedited AI output: repeated patterns, generic design, weak proof, and no final QA.
Direct answer: Em dashes can be a visible AI-writing tell, but they are not proof and they are not a ranking penalty. The real issue is publishing unedited AI output with generic structure, unsupported claims, weak examples, and no brand QA.
Written by: Esmail Hanif, AI Visibility Strategist & Founder, Martecks
Short answer
Em dashes are not the real problem. They are a social tell because many AI drafts overuse them, but one punctuation mark does not prove AI wrote a page and does not create a Google penalty by itself.
The real problem is unedited AI output: same rhythm, same transitions, unsupported claims, generic screenshots, blue-purple gradients, broken mobile layout, vague badges, and missing trust pages.
Why people notice em dashes now
The em dash became a meme because AI writing often uses it to stitch together polished clauses. Once people started seeing that pattern in emails, LinkedIn posts, websites, and blog intros, the punctuation became a shortcut for "this feels generated."
Research backs the broader pattern but not the overclaim. Papers on LLM prose have studied stylistic fingerprints from training and formatting habits, and a 2026 medRxiv analysis found a population-level rise in em dash frequency after ChatGPT became common. That makes the em dash a weak signal, not a detector.
Sources: arXiv: The Last Fingerprint, arXiv: Em-ergence of the em-dash
What Google actually cares about
Google’s public guidance is not "never use AI" and not "never use em dashes." Google says its systems aim to reward helpful, reliable, people-first content. It also says appropriate AI use is not inherently against Search guidelines; the problem is scaled, low-value, manipulative content.
So the SEO question is not whether a paragraph contains an em dash. The question is whether the page answers the query better than alternatives, shows experience or proof, cites what needs citation, and helps a real reader make a decision.
The real AI writing tells
Look for patterns across the page, not one punctuation mark.
| Tell | Why it feels AI generated | Better fix |
|---|---|---|
| Overused em dashes | Every sentence has the same polished pivot. | Keep them only where they improve rhythm. |
| Generic intro | The page starts with broad setup instead of the reader problem. | Open with the specific pain or decision. |
| Balanced but empty paragraphs | The draft sounds fair but says nothing concrete. | Add examples, edge cases, and tradeoffs. |
| Unsupported claims | Stats and trends appear without sources. | Cite primary sources or remove the claim. |
| Same CTA everywhere | The next step is disconnected from the topic. | Make the CTA match the reader intent. |
The real AI website tells
The same issue shows up visually. A site can feel vibe coded when it repeats default AI design patterns without judgment.
Common tells include blue-purple gradients, tiny hero badges above the headline, over-rounded cards, broken mobile spacing, fake SaaS copy, no privacy policy, no terms page, no real contact details, and no proof that the company exists.
The editing workflow
Use this before publishing AI-assisted writing or an AI-built website.
- Read the page out loud and remove repeated sentence rhythm.
- Search for overused punctuation, but edit for clarity instead of banning one mark.
- Replace broad claims with examples, screenshots, data, or source links.
- Add the real audience, use case, location, price context, or workflow detail.
- Run mobile QA before judging the design.
- Check trust basics: author, contact, privacy, terms, reviews, proof, and next step.
- Link to the related internal page that helps the reader continue.
Where a reusable style skill helps
The cleanest fix is not manually deleting every em dash. The cleanest fix is a reusable style skill that tells AI your audience, voice, banned patterns, source rules, examples, and final QA standard.
That turns "make it sound human" into a durable editorial system. It also makes your future drafts more consistent because the model has a house style instead of guessing from scratch.
Reference links
This topic came from TikTok source 88 and was validated against Google Search guidance and current research on LLM writing patterns.
Sources: TikTok source 88 idea trigger, Google: creating helpful, reliable, people-first content, Google: guidance about AI-generated content, arXiv: The Last Fingerprint, arXiv: Em-ergence of the em-dash
Final answer
Do not panic about em dashes. Fix the system that produced the default draft.
Use AI, but edit like a publisher: source the claims, add real examples, remove repeated patterns, test the page on mobile, and make the next step match the article.