Writing with AI without sounding like AI

Many people who read LinkedIn every day can spot a post written by a language model and published without much editing. They may not be able to say why, but the feeling is there: the post is smooth, balanced, a little grand, and says very little. Once readers notice it, they stop reading, and they remember the name.

AI can help you write good posts. The trick is knowing what gives it away and fixing those things in a way that lasts.

Words that give it away

Some words turn up far more often in generated text than in the way people talk about their work. One of them in a post is not a problem. Three in a paragraph is a signal.

• Grand verbs where a plain one would do: 'navigate' for handle, 'foster' for build, 'transform' for change.
• Business fog: 'ecosystem', 'paradigm', 'holistic', 'journey', 'at scale'.
• Intensifiers that add nothing: 'incredibly', 'deeply', 'profoundly'.
• Openers that announce instead of say: 'In today's fast-paced world', 'Let's dive in', 'Here's the thing'.
• Closers that ask for engagement without earning it: 'What do you think? Let me know in the comments.'

Sentence shapes that give it away

Word choice is the easy part. The shapes of the sentences are harder to see and more telling.

• The reversal. A sentence that denies one thing in order to announce another. Used once it can be sharp; used in every post it becomes a tic.
• The rule of three everywhere. Every list has three items, every sentence has three adjectives.
• The one-line paragraph for drama, repeated until nothing feels dramatic.
• The tidy summary at the end that restates the post in softer words.
• Balanced hedging. Every claim followed by 'of course, every situation is different'.
• Long dashes used as the main punctuation mark, where a comma or a full stop would do.

The deeper problem: nothing specific

The surest sign of an unedited AI post is that it could have been written about any company. No names, no numbers, no details you could check. A model with no facts to work from fills the gap with general statements, and general statements are what readers scroll past.

The fix is to give the model facts before it writes. Your client's industry, the number that changed, the question someone asked on a call. Specific input produces specific output far more reliably than any instruction about style.

Set voice rules once, in writing

Editing every post by hand works for a week. What lasts is a short written document of rules that goes into every request. Keep it plain and concrete.

1. List the words you never want to see. Include the ones above and any you personally dislike.
2. Describe the sentence length you want. For example: short sentences, one idea each, no sentence over 25 words.
3. Name the punctuation rules. For example: no long dashes, no exclamation marks, no emoji.
4. Write two or three example sentences in the voice you want, taken from things you have written and liked.
5. Say who the reader is. A founder of a small software company reads differently from a hospital procurement lead.
6. Every time you fix the same thing in a draft twice, add a rule so you never fix it a third time.

The last step matters most. The rules document should grow from your corrections, so the tool gets closer to your voice every week instead of making the same mistakes forever.

Read it out loud

The oldest editing test still works. Read the draft aloud. Anything you would not say to a client across a table, cut or rewrite. If a sentence makes you wince, a reader will feel the same thing without knowing why.

Be honest about who wrote it

A post that says 'I spent the weekend thinking about this' when a tool wrote it in four seconds is a small lie, and small lies add up. One answer is to write company-voice posts in the third person, naming the company and the source of each fact. It reads well, and every claim in it is one somebody can stand behind.

How GoodSocials handles this

In GoodSocials a revise comment rewrites the card and updates your voice principle, so the correction carries to every future post. Posts are written in the third person by rule, and the quality gate holds drafts that make unsupported claims or say nothing.