The Em Dash Isn't the AI Tell You Think It Is — 2026 Guide

If you've been deleting every em dash from your writing out of fear someone will think it's AI-generated, you can stop. New research says you've been watching the wrong punctuation mark.

For the last two years, the em dash has been treated as the ultimate AI writing tell. Teachers flag it, editors side-eye it, and writers have started avoiding a perfectly good piece of punctuation because they're scared of being mistaken for a chatbot. But a widely discussed 2026 analysis from The Economist — covered by Fast Company and Dataconomy — found something surprising: of the major AI models tested, only Claude actually used em dashes more often than human writers do. ChatGPT, which used to lean on the em dash constantly, now uses it less than the average human writer.

So if the em dash isn't the giveaway anymore, what is? This guide breaks down does an em dash mean AI writing, what actually triggers AI detectors in 2026, and how to clean up your text either way — including a free tool that does it in one click.

Where the "em dash AI detector" myth came from

The em dash didn't become suspicious by accident. AI language models are trained on huge volumes of published writing — books, journalism, essays — where the em dash shows up constantly. Everyday writing (texts, emails, quick notes) barely uses it at all, which made the mark stand out the moment AI tools started generating that everyday-style writing at scale. MDPI's writing blog traces this exact pattern: the em dash is common in polished, literary writing and rare in casual writing, so its sudden appearance in casual contexts felt like a red flag.

From there, the idea spread fast. Screenshots, Reddit threads, and even a viral moment where social media accused a major brand's marketing post of being AI-written — just because of a dash — turned "em dash AI myth" into common knowledge almost overnight, according to FIU News's interview with AI researcher Mustafa Ocal.

The problem: it was never a reliable signal, and it's an even weaker one now.

Why does ChatGPT use em dashes? (And why it's changing)

Language models don't think in punctuation rules — they predict the next most likely piece of text based on patterns in their training data. Early GPT models leaned hard on the em dash because it showed up often in the high-quality writing samples used to train them.

That's changing model by model:

ChatGPT (GPT-5.1 and later): Now suppresses em dashes in most default outputs. According to the Economist analysis, it uses them less than human writers do.
Claude: Still the one model that reliably out-dashes humans, per the same report.
Gemini and others: Trending downward, following the same pattern as ChatGPT.

That alone should tell you why building a detection habit around one punctuation mark is a losing strategy — the models keep moving the target.

AI detector false positives in 2026: what's actually happening

Here's the part that matters more than any punctuation mark: real people are getting flagged for writing that sounds "too AI," and it's causing genuine problems — students called into academic-integrity meetings, freelancers losing clients over a false accusation. Plagiarism Today's breakdown of em dash usage across models found wildly inconsistent dash usage between tools — some AI models barely use the em dash at all — which means anyone using "lots of dashes" as their detection method is going to be wrong constantly, in both directions.

Signs of AI-generated text that actually hold up

If punctuation isn't the answer, what is? Based on the Economist research and existing detection literature, a few patterns are far more consistent than any single character:

Light punctuation overall. AI-generated text tends to use fewer commas, semicolons, and parentheses than human writing — the opposite of the em dash myth.
Long, uniform sentences. Instead of varying sentence length the way people naturally do, AI text often runs long sentences back to back — a pattern researchers call "burstiness" when it's missing.
Overreliance on "and." The Economist report specifically flagged AI's tendency to string ideas together with "and" instead of varying sentence structure.
Verbosity and rare words. AI-written text tends to run longer than necessary and reach for fancier vocabulary — polysyllabic words, scientific-sounding phrasing, and nominalizations (turning verbs into noun phrases) — more often than people do.
Predictable rhetorical patterns. Phrases like "it's not X, it's Y" and rule-of-three lists show up far more often in AI writing than in natural human writing.

None of these are perfect on their own either — but together, they're a far more honest picture of how to spot AI writing in 2026 than counting dashes ever was.

How to remove em dashes from ChatGPT text (if you still want to)

Even though the em dash isn't the smoking gun it used to be, plenty of writers still want a cleaner, more natural-sounding draft before publishing — especially if a teacher, editor, or client still associates dashes with AI. A few ways to do it:

Manually swap them out. Replace with a comma, a period and new sentence, or parentheses, depending on what the sentence needs.
Ask the AI tool itself. Most chat tools will follow a style instruction like "avoid em dashes" — though it doesn't always stick for long.
Use a text cleaner. This is the fastest option if you're editing a full draft rather than one sentence at a time.

That's exactly what our free AI Text Cleaner is built for — it strips em dashes, markdown symbols, smart quotes, and other AI-formatting leftovers from ChatGPT or Claude output in one click, right in your browser. Nothing is uploaded or stored.

How to make ChatGPT text sound human (beyond punctuation)

Since punctuation isn't the real fix, focus on the patterns that actually matter:

Vary your sentence length on purpose. Mix short, punchy sentences with longer ones instead of letting everything run at a similar length.
Read it out loud. Uniform, monotone rhythm is easier to hear than to see on the page.
Cut unnecessary words. AI drafts tend to run long — run your draft through a Word Counter and look for sentences that repeat an idea already covered.
Check your keyword and phrase repetition. Overused connector words and repeated phrasing are a giveaway — our Keyword Density Checker will show you exactly which words and phrases you're leaning on too hard.
Rewrite the predictable patterns. If you spot an "it's not X, it's Y" line or a rule-of-three list you didn't consciously choose, that's a sign the draft needs a human pass.

FAQ

Is the em dash still an AI tell in 2026?

Not a reliable one. Of the major models tested in the 2026 Economist analysis, only Claude used em dashes more than human writers — ChatGPT now uses them less often than people do.

What actually triggers AI detectors?

Detection tools weigh dozens of signals, but the strongest ones identified so far are light punctuation overall, long and uniform sentence structure, overuse of "and," excess verbosity, and predictable rhetorical patterns like rule-of-three lists.

How do I remove em dashes from text online for free?

Paste your text into our AI Text Cleaner — it removes em dashes, smart quotes, and markdown formatting instantly, with nothing uploaded or saved.

Do AI detectors give false positives?

Yes, regularly — especially when a detector (or a person) relies on a single signal like punctuation. That's part of why relying on the em dash alone as evidence of AI writing has caused real problems for human writers.

What should I use instead of an em dash?

Depends on the sentence: a comma or semicolon for a related thought, a period and new sentence for a cleaner break, or parentheses if the aside is truly secondary.

Tools used in this guide: AI Text Cleaner · Word Counter · Keyword Density Checker · AI Token Counter