AI and LinkedIn ·
Why a ChatGPT post gets spotted (and how not to fall into it)
Posts made entirely with generalist AI (ChatGPT, Claude, Gemini with no further tuning) are becoming instantly recognisable on LinkedIn. There are concrete linguistic marks that give them away, and a trained audience is starting to discard them before reading the second paragraph.
This guide collects those marks so you can identify them, whether to spot them in others or to avoid them in what you publish. The editorial position is direct: using AI to think well is fine, using it to write for you destroys your personal brand quietly. The difference shows.
By Sheena de PunkVoice · Edited by Mario Pérez
Why it matters that it shows
When a reader detects a post was made with AI, they do not think "how efficient this profile is". They think "this person does not have the time or the inclination to write what they sign". The association being built is disinterest, not efficiency.
The effect is cumulative. One post spotted occasionally does not sink your reputation. A run of five or six consecutive posts with obvious generalist AI marks puts you in a category that is hard to leave: the automated profile that no longer requires attention.
And there is an additional invisible cost: people stop commenting substantively on your posts because they assume you will not reply with judgement of your own either. The conversations that actually build a personal brand die before they start.
The linguistic marks that give it away
Standard closing connectives. "In summary", "at the end of the day", "all things considered", "ultimately", "without a doubt". ChatGPT overuses these closes because they maximise the sense of rhetorical completeness. A post written natively in colloquial English almost never stacks several of them in the same text.
Long dashes everywhere. Using the em dash to create dramatic pauses mid sentence is a device ChatGPT deploys excessively. A text with more than two or three in two hundred words is usually a signal.
Uniform structure by punctuation. Sentences tend to have the same length and rhythm. A human text alternates more between short and long sentences, a generated one tends to keep a constant cadence.
Categorising filler adjectives. "Deeply meaningful", "truly transformative", "genuinely impactful", "strategically fundamental". Paired adjectives that raise the grandiosity without adding information.
Closing on a rhetorical question. "So what do you think?" at the end of a post about any topic. A typical marketing template device ChatGPT reproduces by default.
Lists with identical structure. Bullets all starting with a verb, all exactly the same number of words. Too much symmetry is a signal.
Narrative patterns that repeat
An opening with an inflated figure or a generic anecdote. "Five years ago I started X", "the other day I had a conversation..." (without the anecdote sounding like concrete real life). ChatGPT reproduces the virality pattern without having any experience behind it.
The "it is not X, it is Y" pivot. Structures like "It is not about doing more, it is about doing better", "it is not talent, it is perseverance". A recognisable rhetorical device ChatGPT inserts automatically to sound profound.
Three short sentences in a row at the end of the post. ChatGPT closes many posts with a rhetorical tricolon (three short sentences finishing off the idea). When it repeats across five consecutive posts from the same profile, it is an obvious signal.
Messages ending by reaffirming your authority. "That is why I will keep betting on...", "I keep choosing X because...". A generic close that gives the feeling of a conclusion without having said anything concrete.
Rewordings of the title inside the post. ChatGPT tends to repeat the title's idea two or three times through the text, paraphrasing it. A human writer trusts that the title is already written.
How to use AI as a thinking companion without diluting yourself
A self test before publishing
If you are unsure whether your post smells of generalist AI, run it through this self test before publishing:
Does it tell a concrete story that is only yours? If the anecdote could apply to any professional in any industry, that is a signal. Specific stories (with detail, context, concrete names) make the difference.
Is there vocabulary of your own? If you used words or expressions you repeat in your day to day, you are on the right track. If everything is phrased in the neutral English of a style manual, be suspicious.
Does the rhythm change between sentences? Short and long sentences alternating, an uneven rhythm. If everything has the same cadence, an AI probably wrote it.
Have you left something unexplained? Human texts assume context, take things for granted, leave the obvious implicit. Generated texts over-explain, almost pedagogically.
Do you believe it when you read it out loud? Read it slowly. If at some point it sounds borrowed, odd or impostorish, that is a signal the human hand has not passed over it enough.
If your post fails two or more of these questions, it is worth rewriting or not publishing.
How to use AI without falling into the marks that give it away
Use AI before writing, not during or after. Ask it to question your idea, give you counterarguments, suggest examples. Then write the whole post yourself with those references.
If the AI writes a sentence you love, rewrite it in your own words. Borrowed sentences almost never fit your voice even when they sound good in the abstract.
Give it your draft to detect logical flaws, not to improve prose. Prose is where AI style slips in. Logical flaws are where AI contributes without contaminating.
Ask for criticism, not applause. If you only ask it to "improve" your text, it will hand you back a standardised version. Ask "what is wrong with this post?" or "where am I being unoriginal?" and you get useful feedback without it rewriting for you.
Never copy whole AI sentences into the final post. Every sentence in the published post should have passed through your head and your fingers.
What is coming: automated detection and new marks
LinkedIn is starting to automatically flag content suspected of being AI generated in some cases. Coverage is still limited, but the direction is clear. In one or two years, it is reasonable to expect visible labelling of posts with a high probability of being generated.
The marks that give it away evolve too. New models correct the obvious earlier patterns but introduce others. Three years ago generated texts had very formal vocabulary. Now they try to sound more colloquial but with a uniform cadence. Within a year, other signals will be the ones giving it away.
The only thing that never ages as a signal of authenticity: your own voice. A text with vocabulary, rhythm, anecdotes and positions clearly yours is distinguishable from a generated one whatever the model. And that voice of yours is the only thing neither LinkedIn nor your audience will confuse with AI.