Marvelous examination of how to spot 2026-era AI writing, via the Economist - AI likes long sentences with less punctuation; “and” is its most overused word - relatedly, lists of three things, which drives up use of "and" as well - polysyllabic adjectives: “significant”, “increasingly” - scientific jargon ("rate-limiting," “parameter”) - nominalizations (making nouns from verbs: eg, “expansion” from “expand”) - ofc, everyone's favorite: "it's not X, it's Y"

Aug 3, 2026 · 1:32 PM UTC

119
441
3,554
616,033
link: economist.com/culture/2026/0…

How to spot AI writing

Large language models like long words and em-dashes—or do they?

economist.com
5
12
155
45,085
Nice job, Pangram :)
9
92
26,309
Sort replies: Relevant Recent Liked
Replying to @DKThomp
'table stakes' is one I look for.
4
50
14,376
genuinely
1
1
45
12,682
Replying to @DKThomp
Of course, the tricolon (list of three) has been regarded as an important rhetorical device since antiquity.
5
4
91
21,388
Replying to @DKThomp @Jesse_Livermore
And em dashes - which makes me sad, given how much I love them.
8
2
58
19,133
Replying to @DKThomp
>use of three > its not X, it's Y so something I've thought that was unspoken here, is that AI is clearly trained on management consulting writing?
1
16
776
Replying to @DKThomp
So you’re just basically saying that what you think is AI writing is writing by intelligent people.
13
788
Replying to @DKThomp
Anyone else like "goddamnit run-ons and lists of three this is just how I sound!"?
1
8
1,295
Replying to @DKThomp
Claude has one of the worst writing styles known to man.
5
818
Replying to @DKThomp
1
752
Replying to @DKThomp @DynamicWebPaige
Final boss:
1
3
603
Replying to @DKThomp
The best AI-assisted writing still benefits from human editing to add rhythm, personality, and variety.
1
358
Replying to @DKThomp
Seems like @pangram should then be less forceful in its conclusions. I edit a daily newsletter on @Substack. Five years running. 120+k subs. 184 countries. All hand-crafted. Like many, I'm still getting "100% AI-written" results.
1
301
Replying to @DKThomp
"and" perhaps was fed the King James Bible
111
Replying to @DKThomp
Sharp observation, spotting AI patterns is becoming a new literacy...
53
Replying to @DKThomp
Shitty for those of us who can't write short sentences.
310
Replying to @DKThomp
Well, shit. On the other hand, nobody is gonna think 'It's time to flay the ocelot' came from the fucking AI.
1,001
Replying to @DKThomp
The strongest signal isn't a single phrase or pattern. It's whether the writing feels genuinely purposeful rather than statistically fluent.
856
Replying to @DKThomp
Have some fun watching this ultimate face off. Who’s your favourite? Choose one before watching the outputs.
Gerard Sans | Axiom 🇬🇧
@gerardsans
Aug 3
Replying to @gerardsans
Claude 🇺🇸 vs Qwen 🇨🇳 Prompt: Open ended text /🧵
913
Replying to @DKThomp
“That is load-bearing.” “The distinction matters.”
126
6,080
Replying to @DKThomp
A big one is unnecessary lead-ins. Here’s the thing: AI uses them all the time. Here’s why it matters. No one speaks like this.
1
27
932
Replying to @DKThomp
This is also, unfortunately, mostly how I've written since my undergrad days more than 20 years ago.
1
22
1,784
Replying to @DKThomp
Claudes have a different definition of a "long" sentence:
1
21
5,498
Replying to @DKThomp
Outside of X all the people i talk to, openly admint using AI to write stuff, and they dont care that much if its AI ( also a lot of them are not reading anymore, they just ask AI to summarize) my point being... should we really be waisting time caring about AI writing patterns?
5
2
11
4,287
Replying to @DKThomp
I fear that people will continue to change their writing style to avoid being labeled as AI.
15
1,934
Replying to @DKThomp
This is ridiculous. All of these things were taught in my ELA and writing classes in school. We're just going to stigmatized good writing conventions as AI.
1
11
660
Replying to @DKThomp
This is the analytic explanation. The continental explanation is "I know it when I see it"
1
10
1,254
Replying to @DKThomp
Another big one is negatives for things that aren’t logical introduced in the first place. “The Incas made better pyramids. They didn’t cheat. They didn’t use a shirt cut.”
1
8
934
Replying to @DKThomp
controversial take: lots of human writing suffers from many of these tells + the more AI content you read (with/without realizing it) the more you might write like AI as a human, so your 100% human writing might trigger Pangram-esque classifiers and this will be very unfair
7
398
Replying to @DKThomp
Essentially you're saying AI writes mostly better and more complex while average human is less complex and more simplistic..
6
6
3,501
Replying to @DKThomp
cover letters have every one of these now. the newer problem runs the other way: people who just write like that naturally get flagged as bots. we're speedrunning a world where clean grammar reads as suspicious
6
935
Replying to @DKThomp
I mean, now compare the AI with a legal brief, published academic article, or college textbook.
1
6
938
Replying to @DKThomp
Quietly or Silently and their cousin “the question no one’s asking”
1
4
592
Replying to @DKThomp
Style fingerprints don't stay useful forever. They disappear as soon as people start optimizing around them.
4
286
Replying to @DKThomp
Important: it looks like The Economist is comparing these other sources to LLMs *when the LLMs are asked to write news articles*. Some of these findings probably change when the LLMs are asked to produce different outputs—especially social media posts, for sentence length.
4
1,133
Load more