Claude Sonnet 5 used the most em dashes of the ten AI models we tested: 9.55 per 1,000 words, or about three in a 300-word draft. That is more than twice the rate of GPT-5.6 and Gemini, which used 3.3 to 4.4. Claude Haiku 4.5 came second, at 6.04, and GPT-6 Luna close behind at 5.39. Those two cannot be told apart.
Every model used them. Between 55% and 92% of each model's drafts had at least one. But the em dash is one AI writing tell out of 23 we count. Across all 23, the six models land close together. What differs is which tell each one leaves.
Claude uses the most em dashes
We asked each model for 35 short pieces: blog posts, emails, product copy, opinion pieces, personal essays, LinkedIn posts and how-to guides. Each ran three times, so each row below is 105 drafts of about 300 words.
Em dashes per 1,000 words, ten models, 1,049 drafts, September and October 2026
| Model | Dashes per 1,000 words | 95% range | Lowest and highest run | Drafts with any dash |
|---|---|---|---|---|
| Claude Sonnet 5 | 9.55 | 8.56–10.65 | 9.14–10.03 | 92.4% |
| Claude Haiku 4.5 | 6.04 | 5.23–6.97 | 5.49–7.01 | 83.8% |
| GPT-6 Luna (October) | 5.39 | 4.61–6.29 | 5.05–5.56 | 71.4% |
| GPT-5.6 Luna | 4.43 | 3.74–5.24 | 3.83–5.11 | 61.0% |
| GPT-5.6 Terra | 4.35 | 3.67–5.14 | 3.99–4.64 | 62.9% |
| Grok 4.7 (October) | 3.85 | 3.22–4.60 | 3.53–4.27 | 55.2% |
| Gemini 3.7 Flash | 3.84 | 3.20–4.61 | 2.83–4.67 | 60.0% |
| GPT-6.1 Sol (October) | 3.56 | 2.96–4.28 | 3.24–3.81 | 65.7% |
| Gemini 3.8 Flash | 3.56 | 2.94–4.30 | 2.99–4.20 | 63.5% |
| Gemini 4 Argon (October) | 3.27 | 2.71–3.96 | 2.88–3.91 | 59.0% |
Each row is one model, about 105 drafts in three runs. The 95% range shows where the true rate likely sits. Lowest and highest run gives the spread across those three runs.
The 95% range is where the true rate likely sits. We also ran each model three times, and the run column shows its lowest and highest run. We only call one model higher than another when their runs do not overlap. Sonnet 5 clears all nine others. Haiku 4.5 clears the September GPT and Gemini models but overlaps GPT-6 Luna. The rest overlap each other. Rows marked October ran on 15 October 2026. We ran GPT-5.6 Luna again that day as a check, and its rate moved from 4.43 to 4.33.
Sonnet 5 went past two em dashes in 59.0% of its drafts. More than two in one draft is where our AI slop detector flags them. The September GPT models did so in 16% to 19% of drafts, and the Gemini Flash models in 9% to 11%. Of the October models, GPT-6 Luna did so in 24.8% (95% interval 17.5% to 33.8%), Grok 4.7 in 14.3% (8.9% to 22.2%), GPT-6.1 Sol in 8.6% (4.6% to 15.5%) and Gemini 4 Argon in 4.8% (2.1% to 10.7%).
You may have read that Claude stopped using em dashes. A public test on GitHub found 0.05 per 1,000 words for Claude Opus 5.5. That is a different model. We had not tested Opus 5.5 when this table ran. On 20 November 2026 we did, and counted 0.12. The next section has those counts.
November: the newer Claude models dropped them
On 20 November 2026 we gave the same 35 requests, three times each, to 11 models, Claude Opus 5.5 among them, which we had left out for cost. GPT and Gemini ran through their own APIs. GPT-6 Luna and Gemini 4 Argon ran again as checks, and both matched their October counts.
Em dashes per 1,000 words, GPT and Gemini, 20 November 2026
| Model | Group | Dashes per 1,000 words | 95% range | Lowest and highest run |
|---|---|---|---|---|
| GPT-6.2 Astra | Frontier | 0.53 | 0.33–0.86 | 0.30–0.70 |
| Gemini 4 Carbon | Frontier | 4.32 | 3.66–5.10 | 3.80–4.76 |
| GPT-6.2 Sol | Middle | 1.35 | 1.00–1.82 | 1.34–1.35 |
| Gemini 4 Neon | Middle | 4.81 | 4.10–5.65 | 4.56–5.07 |
| GPT-6 Luna (check) | Fast | 5.53 | 4.75–6.45 | 4.71–6.02 |
| Gemini 4 Hydrogen | Fast | 5.07 | 4.33–5.94 | 4.69–5.73 |
| Gemini 4 Argon (check) | Check only | 3.44 | 2.86–4.14 | 3.00–3.69 |
Each row is one model, 105 drafts in three runs. The 95% range shows where the true rate likely sits. Compare rows only inside a group: Astra with Carbon, Sol with Neon, Luna with Hydrogen.
Em dashes per 1,000 words, Claude through Claude Code, 20 November 2026
| Model | Dashes per 1,000 words | 95% range | Lowest and highest run |
|---|---|---|---|
| Claude Sonnet 5 (check) | 11.46 | 10.34–12.69 | 11.36–11.60 |
| Claude Fable 5.2 | 0.30 | 0.16–0.56 | 0.09–0.46 |
| Claude Opus 5.5 | 0.12 | 0.05–0.32 | 0.00–0.37 |
| Claude Sonnet 5.5 | 0.00 | 0.00–0.12 | 0.00–0.00 |
Claude Code on a Claude subscription, not the API or the app. Compare these four rows only with each other. The 95% range shows where the true rate likely sits.
Run through Claude Code, Sonnet 5 wrote 11.46 em dashes per 1,000 words, against 9.55 in our September test, so do not set the Claude rows beside the tables above. Claude Fable 5.2, Claude Opus 5.5 and Claude Sonnet 5.5 each wrote fewer em dashes than Sonnet 5 in every run. Chance alone would give gaps that size about 1 time in 10,000 or less (p = 0.0001 each). Sonnet 5.5 wrote none in 32,831 words. None of them swapped in another dash: spaced hyphens stayed at zero. Our 0.12 for Opus 5.5 is close to the 0.05 the public test found.
In the frontier and middle groups, the GPT model wrote fewer em dashes than the Gemini one. Astra wrote 0.53 against Carbon's 4.32, and Sol 1.35 against Neon's 4.81, lower in every run. In the fast group, GPT-6 Luna and Gemini 4 Hydrogen were too close to call. Our best AI model for writing test compares all 23 tells for these models, group by group.
Across all 23 tells, the gaps are small
Em dashes are one of 23 patterns our detector counts. The others include stock words like "seamless", lists of three, "studies show" with no study named, and endings that start with "Ultimately,". Counted across all 23, the six September models scored between 4.2 and 5.2 tells per 1,000 words. The four October models scored 3.3 to 3.9. GPT-5.6 Luna, which we ran on both dates, also scored 0.36 lower in October, so part of that gap may be the date rather than the models.
All 23 tells per 1,000 words, same 1,049 drafts
| Model | Tells per 1,000 words | 95% range | Lowest and highest run |
|---|---|---|---|
| Claude Sonnet 5 | 5.18 | 4.46–6.01 | 4.57–5.69 |
| Gemini 3.7 Flash | 4.71 | 4.00–5.56 | 4.62–4.85 |
| Gemini 3.8 Flash | 4.57 | 3.86–5.40 | 3.99–5.29 |
| Claude Haiku 4.5 | 4.49 | 3.80–5.31 | 3.75–5.13 |
| GPT-5.6 Luna | 4.33 | 3.65–5.13 | 4.04–4.52 |
| GPT-5.6 Terra | 4.22 | 3.55–5.00 | 3.99–4.60 |
| GPT-6 Luna (October) | 3.85 | 3.21–4.63 | 3.44–4.21 |
| Grok 4.7 (October) | 3.53 | 2.93–4.25 | 3.15–3.89 |
| GPT-6.1 Sol (October) | 3.52 | 2.93–4.24 | 3.15–3.80 |
| Gemini 4 Argon (October) | 3.34 | 2.76–4.03 | 2.97–3.77 |
Each row is one model. Tells counts all 23 patterns our detector checks, em dashes among them. The 95% range shows where the true rate likely sits.
Every model's runs overlap at least one other model's, so none of them writes clearly more like AI than the rest. The four count rules score once per draft: too many em dashes, lists of three, "rather than" and "not Y" endings. Every other tell scores once per match.
So by count, no model writes most like AI. Which tell to look for depends on who wrote the draft.
Each vendor leaves a different tell
The totals tie because each vendor's models lean on a different habit. Claude leans on em dashes. GPT leans on lists of three. Gemini leans on stock AI words.
Share of drafts with each tell, by vendor, two models each
| Tell | Claude | GPT | Gemini |
|---|---|---|---|
| More than two em dashes | 42.4% | 17.6% | 10.0% |
| Too many lists of three | 56.7% | 86.2% | 70.3% |
| A stock AI word, like "seamless" | 17.1% | 5.7% | 23.9% |
| "Serves as" or "represents a" instead of "is" | 0.5% | 0.5% | 5.3% |
Each row is one tell. Each cell gives the share of drafts from that vendor's two models with the tell, about 210 drafts per cell.
Each cell pools about 210 drafts: two models, 35 prompts, three runs. A draft of 300 words is allowed one list of three before it counts. Claude's lead on em dashes and GPT's lead on lists of three are far beyond chance. Gemini's lead on stock words is clear against GPT, but not against Claude. In October, two of these habits held and one faded. Gemini 4 Argon put a stock AI word in 21.9% of drafts (15.1% to 30.7%), level with the Gemini Flash models. GPT-6 Luna and GPT-6.1 Sol wrote about 11 lists of three per 1,000 words, against 16.3 for GPT-5.6 Luna on the same day: about a third fewer, and clear of chance.
Here is one line from each vendor, from the first run:
A concert or major sporting event in a city can spike demand—and prices—for flights there.
Claude Sonnet 5, a blog post on airline prices
Stay cool, hydrated, and prepared with a water bottle that works as hard as you do.
GPT-5.6 Terra, a product description
The logistics and supply chain landscape is evolving faster than ever.
Gemini 3.8 Flash, a LinkedIn post about a new job
LinkedIn posts carry the most tells
The kind of writing mattered more than the model. LinkedIn posts carried 7.20 tells per 1,000 words. Personal essays carried 1.87, about a quarter as many. Opinion pieces and product copy sat near LinkedIn.
Tells per 1,000 words for each type of draft, all six models together
| Kind of writing | Tells per 1,000 words | 95% range |
|---|---|---|
| LinkedIn post | 7.20 | 6.22–8.34 |
| Opinion piece | 6.56 | 5.67–7.58 |
| Product or marketing copy | 6.33 | 5.42–7.38 |
| Blog post | 3.99 | 3.31–4.80 |
| How-to guide | 3.34 | 2.72–4.09 |
| 3.08 | 2.46–3.85 | |
| Personal essay | 1.87 | 1.42–2.46 |
Each row is one type of draft. The 95% range shows where the true rate likely sits.
If you draft LinkedIn posts or sales copy with AI, expect about twice the tells of an email from the same model. The model reaches for the same moves a copywriter would: three benefits in a row, a punchy aside, a word like "seamless".
More thinking did not remove them
Some models let you set how long they think before they write. We ran two of them at more than one level. GPT-5.6 Luna ran at all five of its levels, from minimal to xhigh. Gemini 3.8 Flash ran at low and at high.
Neither changed. Luna's thinking grew about tenfold, from 26 to 245 tokens (word pieces) per draft, and its tells stayed between 3.8 and 4.3 per 1,000 words. Gemini 3.8 Flash went from no thinking to about 1,000 tokens, and its tells went from 4.57 to 5.15. That gap is small enough to be chance.
So a higher effort setting will not get you fewer tells, and it costs more.
Why AI uses em dashes
Nobody has shown why for certain. One idea, from engineer Sean Goedecke in October 2025, is that newer models learn from scanned print books from around 1900, which use more em dashes than writing today.
Our data cannot test that idea. It does show the habit is not fixed across AI. Ten models got the same prompts and settings, and the heaviest user wrote 2.9 times as many as the lightest. That points to how each model was trained.
It also shifts when you ask. In November 2025, Tom's Guide reported that OpenAI's Sam Altman said ChatGPT now follows a custom instruction not to use them. We gave no such instruction, so these counts show each model's default.
An em dash does not prove AI wrote it
People used em dashes long before chatbots, and many still do. And the absence of one proves nothing either. About 38% of the GPT and Gemini drafts had no em dash at all, and they still came from AI.
A tell is a pattern. It proves nothing on its own. Our free AI slop detector runs the same 23 checks we used here and quotes each line it flags, so you can decide which ones to fix. The AI humanizer rewrites them for you.
A tell check cannot see whether a claim is true. A draft with no em dashes can still cite a study that does not exist. That is why TrueStandard checks the claims: paste your draft, and four models from different labs check each fact in about 60 seconds.
We wrote up each vendor's results in more depth:
- GPT-5.6 and GPT-6: fewer em dashes than Claude, and the most lists of three, though GPT-6 wrote a third fewer.
- Claude Sonnet 5 and Haiku 4.5: the most em dashes of any model we tested.
- Gemini 3.7 Flash, 3.8 Flash and 4 Argon: the fewest em dashes in September and October, and the most stock AI words.
How we ran it
We wrote 35 plain requests, five each for seven kinds of writing. One was "Write an email to my landlord asking them to fix a leaking kitchen faucet." Each asked for about 300 words and the text only. We set no tone, no style and no system prompt.
Each model got every request three times, with the same settings: low reasoning effort and room for 32,000 tokens, or word pieces, so no draft was cut short. We ran them on 26 September 2026, through each vendor's API, so no chat app added its own instructions. One Gemini 3.8 Flash draft ended in an error and was dropped, leaving 629. We ran GPT-6 Luna, GPT-6.1 Sol, Grok 4.7 and Gemini 4 Argon the same way on 15 October 2026, 420 more drafts with no errors.
A script then scored every draft against the 23 patterns in our AI slop detector. It is plain pattern matching. No AI model graded anything.
We tested short first drafts from a bare request. A style guide or a custom instruction would change the counts. A flag is a pattern match. It is not a verdict: a how-to guide that says "12 to 24 hours" trips the vague-range check even when that range is right. We did not measure human writing, and we left out four flagship models for cost: Claude Opus 5.5, Claude Fable 5.1, GPT-6 Astra and Gemini 3.1 Pro. On 20 November 2026 we added Claude Opus 5.5, Claude Fable 5.2 and GPT-6.2 Astra.
FAQ
Which AI uses the most em dashes?
Claude, in our test. Claude Sonnet 5 used 9.55 em dashes per 1,000 words and Claude Haiku 4.5 used 6.04. GPT-5.6 Terra and Luna used about 4.4, and Gemini 3.7 and 3.8 Flash used 3.6 to 3.8. That is 629 drafts across seven kinds of writing, run in September 2026. In October, GPT-6 Luna used 5.39, Grok 4.7 3.85, GPT-6.1 Sol 3.56 and Gemini 4 Argon 3.27. On 20 November, run through Claude Code, Claude Sonnet 5 used 11.46, and the newer Claude Fable 5.2, Opus 5.5 and Sonnet 5.5 used 0.30 or fewer.
Why does AI use em dashes?
No one has proven why. One theory holds that newer models learned from old print books, which use more em dashes. Our test shows the habit varies a lot by model: on the same prompts, Claude Sonnet 5 used 2.7 times as many as Gemini 3.8 Flash. So it comes from how each model was trained.
Are em dashes a sign of AI writing?
They are a weak sign on their own. People use them too, and about 38% of the GPT and Gemini drafts in our test had none at all. Look for several tells together, such as em dashes, lists of three and words like "seamless", instead of any single one.
Does ChatGPT still use em dashes?
Yes, but less than Claude. With no instructions, GPT-5.6 Terra and Luna used about 4.4 em dashes per 1,000 words, and about 62% of their drafts had at least one. GPT-6 Luna, tested in October, used 5.39. OpenAI says ChatGPT will follow a custom instruction not to use them.
What are the most common signs of AI writing?
In our 629 drafts, the most common was too many lists of three, in 71% of drafts. Next came too many em dashes, in 23%, and stock AI words like "seamless", "robust" and "landscape", in 16%. Which one you see most depends on the vendor.
Does higher reasoning effort make AI writing sound less like AI?
Not in our test. GPT-5.6 Luna at five effort levels and Gemini 3.8 Flash at low and high left the same number of tells, within the margin of chance, though thinking time grew about tenfold.
Keep reading
Multi-Agent vs Multi-Model AI in 2026
Multi-agent AI runs many roles of one model. Multi-model AI runs different models from different labs. Only multi-model can catch an error the first model is blind to.
Does Claude Use Em Dashes? Sonnet 5 and Haiku 4.5 Counted
We gave Claude Sonnet 5 and Haiku 4.5 the same 35 writing jobs as GPT and Gemini, three times each. Then we counted every em dash and 22 other AI writing tells.
Does Gemini Use Em Dashes? Fewer, but More AI Words
We gave Gemini 3.7 Flash and Gemini 3.8 Flash the same 35 writing jobs as Claude and GPT, three times each. Then we counted every em dash and 22 other AI writing tells. In October we added Gemini 4 Argon, and in November three more Gemini 4 models.
Best AI Model for Writing: 11 Models Tested
We counted 23 AI writing tells in 1,155 drafts. Two pairs split, and no model won overall.
Long Context vs RAG in 2026
Under about 100 pages, put your documents in a long context window. From 100 to 500 pages, Karpathy's wiki method works, and past 500, graph RAG wins. Better retrieval still does not make the answer true.
Fix the tells, then check the claims
You can strip every em dash and still publish a made-up statistic. Paste your draft into TrueStandard. Four models from different labs check each claim and flag what none of them can back up, in about 60 seconds.
TrueStandard is for writers who publish AI-assisted drafts. Paid plans check every claim and citation across four AI models before you hit publish.
Free: one check a day, no signup. Pro is $20 a month for 250 drafts.
Check Your Draft →