Yes, but less than Claude or GPT. Gemini 3.8 Flash used 3.56 em dashes per 1,000 words and Gemini 3.7 Flash used 3.84, the lowest of the six models we tested. That is about one in a 300-word draft. Claude Sonnet 5 used 9.55.
Gemini's tell is the stock AI word. About one Gemini draft in four had one, such as "seamless", "landscape" or "streamlined". GPT drafts had one about one time in 18.
Gemini uses the fewest em dashes we measured
We asked each model for 35 short pieces in seven kinds of writing, from emails to LinkedIn posts. Each ran three times, so each row is about 105 drafts of about 300 words. We called the models directly through Google's API. The Gemini app was not involved.
Em dashes per 1,000 words, Gemini against Claude and GPT, September 2026
| Model | Em dashes per 1,000 words | 95% range | Drafts with one or more | Drafts with more than two |
|---|---|---|---|---|
| Gemini 3.8 Flash | 3.56 | 2.94 to 4.30 | 63.5% | 8.7% |
| Gemini 3.7 Flash | 3.84 | 3.20 to 4.61 | 60.0% | 11.4% |
| GPT-5.6 Terra | 4.35 | 3.67 to 5.14 | 62.9% | 16.2% |
| Claude Sonnet 5 | 9.55 | 8.56 to 10.65 | 92.4% | 59.0% |
The 95% range is where the true rate likely sits. Gemini's ranges overlap GPT's, so the two are about even. Both sit well below Claude. More than two in one draft is where our AI slop detector flags them. No Gemini draft at low effort had more than four.
Gemini also rarely swapped in a spaced en dash ( – ) instead, under 0.4 per 1,000 words for either model. So its low count is real. It did not just swap one dash for another.
Its tell is the stock AI word
Words like "seamless", "robust" and "landscape" are the tell most readers already know. Gemini used them most. 25 of 105 Gemini 3.7 Flash drafts had one, and 25 of 104 Gemini 3.8 Flash drafts.
Drafts with at least one stock AI word
| Model | Drafts | 95% range |
|---|---|---|
| Gemini 3.8 Flash | 24.0% | 16.8 to 33.1% |
| Gemini 3.7 Flash | 23.8% | 16.7 to 32.8% |
| Claude Haiku 4.5 | 20.0% | 13.5 to 28.6% |
| Claude Sonnet 5 | 14.3% | 8.9 to 22.2% |
| GPT-5.6 Terra | 5.7% | 2.6 to 11.9% |
| GPT-5.6 Luna | 5.7% | 2.6 to 11.9% |
Gemini's lead over GPT is far beyond chance: 50 of 209 drafts against 12 of 210. Its lead over Claude, 50 of 209 against 36 of 210, could still be chance.
Each model had its favourites. Gemini 3.8 Flash used "landscape" in 7 drafts and "seamlessly" in 4. Gemini 3.7 Flash used "seamless" in 7 and "streamlined" and "robust" in 4 each. Both were asked for a LinkedIn post about a new job in logistics:
The logistics and supply chain landscape is evolving faster than ever.
Gemini 3.8 Flash, first run
We have big goals ahead, and I could not be more energized to lead such a talented operations team as we redefine what seamless logistics looks like.
Gemini 3.7 Flash, first run
It writes "serves as" where a person writes "is"
Gemini had one more habit the other vendors did not share. It wrote "serves as", "acts as" or "represents a" in 11 of 209 drafts. Claude and GPT did so in 1 of 210 each. The gap is too large to be chance.
For students who cannot afford private tutoring, AI serves as an equalizer, leveling an academic playing field that has long favored the wealthy.
Gemini 3.8 Flash, an opinion piece on AI and homework, first run
One email request drew the same phrase again and again. One email asked Gemini to tell a client that a website would cost more. In four of its six drafts, Gemini said the online store "represents a" major, substantial or significant "expansion" of the scope. That held across both models and all three runs.
More thinking did not change it
At low effort, Gemini 3.8 Flash wrote with no thinking at all. We also ran it at high, where it thought for about 1,000 tokens, or word pieces, before each draft and cost three and a half times as much.
Its tells went from 4.57 to 5.15 per 1,000 words, and its em dashes from 3.56 to 4.33. Both gaps are small enough to be chance. Stock AI words showed up in 27.6% of drafts at high, against 24.0% at low.
How to clean up a Gemini draft
Search for the words first. "Seamless", "landscape", "streamlined", "robust" and "empower" came up most often. Then look for "serves as" and "represents a", and try "is" instead.
Our free AI slop detector does both checks, plus 20 more, and quotes each line it flags. The AI humanizer rewrites them for you.
Swapping words leaves the facts alone. A Gemini draft without "seamless" can still quote a figure nobody measured. That is the part TrueStandard checks: paste your draft, and four models from different labs test each claim in about 60 seconds.
How we ran it
We wrote 35 plain requests, five each for seven kinds of writing. One was "Write an email to a client explaining that their website redesign will cost $4,000 more than quoted." 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, on 26 September 2026, through Google's API with low reasoning effort. Gemini 3.8 Flash also ran at high. We gave every draft room for 32,000 tokens. One Gemini 3.8 Flash draft ended in an error mid-sentence and was dropped. A script scored the other 314 Gemini drafts against the 23 patterns in our AI slop detector. No AI model graded anything.
This covers Gemini 3.7 Flash and 3.8 Flash only. We left out Gemini 3.1 Pro for cost. These are short first drafts from a bare request, and the Gemini app adds its own instructions. A flag is a pattern match. It is not a verdict.
We ran the same test on Claude and GPT. The six-model comparison has all of them side by side.
The other two vendors:
- Claude Sonnet 5 and Haiku 4.5 used the most em dashes of any model we tested.
- GPT-5.6 Terra and Luna used the most lists of three and the fewest stock AI words.
FAQ
Does Gemini use em dashes?
Yes, but the fewest of the six models we tested in September 2026. Gemini 3.8 Flash used 3.56 per 1,000 words and Gemini 3.7 Flash 3.84, about one in a 300-word draft. Claude Sonnet 5 used 9.55.
What words does Gemini overuse?
In our test, "landscape", "seamless", "seamlessly", "streamlined", "robust" and "meticulous" came up most. About one Gemini draft in four had at least one stock AI word, against about one in 18 for GPT-5.6.
How can you tell if Gemini wrote something?
Look for stock words like "seamless" and "landscape", and for "serves as" or "represents a" where a person would write "is". No single tell proves it. Gemini used fewer em dashes than Claude or GPT, so a draft without dashes can still be from Gemini.
Does Gemini use fewer em dashes than ChatGPT?
A little, but not by a margin we could confirm. Gemini 3.7 and 3.8 Flash used 3.6 to 3.8 per 1,000 words and GPT-5.6 Terra and Luna about 4.4. Their ranges overlap. Both used far fewer than Claude.
Keep reading
Multi-Agent vs Multi-Model AI in 2026
AI builders use both terms as if they meant the same thing. They are different architectures with different strengths. The difference matters most for the one job neither term sells: catching AI errors before you publish.
AI Agent Workflow Patterns: When Each One Works (and When It Fails)
Six patterns cover almost every agent you'll build. Five are routine. The sixth, verification, breaks when you wire it with a single model, and most teams wire it that way.
Every Type of AI, Explained
From large language models to coding agents: what each type of AI does, which tools lead each group, and how to pick the right one for your work.
What Karpathy's AI Methods Don't Fix
In six weeks, Andrej Karpathy and AI builders shipped three viral reliability methods. Each is real and useful. None of them solves the checking problem for writers.
3 AI Stress Tests from Q2 2026
In April 2026, top AI builders ran real experiments instead of demos. The results were more interesting than the demos. Here is what each test reveals, and why none of them fully answers the question writers care about.
Check the claims in your Gemini draft
A Gemini draft can lose every stock word and still cite a number nobody published. Paste yours into TrueStandard. Four models from different labs check each claim and flag what none of them can back up, in about 60 seconds.
Check Your Draft →