AI and authenticity
How do I make AI-written LinkedIn posts sound like me?
Short answer
Give the AI more of you to work from. Start each post from your own material, such as a voice note, rough notes or a story you often tell, and show it examples of posts you have written.
Ask it to arrange and tighten your words rather than invent new ones. Then read the draft out loud, and rewrite any line you would never say in a meeting.
10.7%
more similarity between stories written with one AI idea than between stories written without, in a study of 300 writers
Why do AI drafts sound like everyone else?
Because a model fills any gap with the most typical wording. In a study of 300 people writing short stories, reported by the University of Exeter, stories written with AI ideas scored higher for novelty one by one, but with a single AI idea they were 10.7 percent more similar to each other. Everyone was pulled towards the same average.
A Cornell study of 118 writers in the US and India found the same drift. With AI suggestions switched on, their writing became more alike, mostly at the expense of the Indian participants' own styles. On LinkedIn the drift shows up as the same hooks, the same lists and the same tidy sign-offs on post after post, and readers learn to spot them.
What makes the biggest difference?
Your own material, and plenty of it. A one-line prompt such as write a post about hiring leaves the model to supply the opinions, examples and phrasing, so it supplies the generic ones. A transcript of you explaining how you hire, including the mistake that changed your mind, hands it your substance and your way of saying things.
Examples come next. Anthropic's prompting guide for Claude calls examples one of the most reliable ways to steer tone and structure, and suggests three to five. OpenAI's guidance makes the same point, that models respond better when shown what you want. A few of your best past posts teach more than any description of your voice.
How do you set it up?
- Collect your own writing. Recent posts you are proud of, transcripts of talks you gave and documents you wrote yourself all count, and the more the tool has, the closer it gets.
- Start each post from something you said. Record a voice note or jot rough notes on one idea, then give that to the tool along with your examples.
- Tell it what to do with the material. Ask it to cut, order and tighten, to keep your phrases, and to leave out anything you didn't say.
- Edit out loud. Read the draft aloud, rewrite anything that doesn't sound like you, then check every claim, name and figure.
Which phrases give AI drafts away?
The usual suspects are openers that promise a lesson, lists of three with a matching rhythm, a question tacked on the end to invite comments, and words you never use in conversation. Keep a short list of them, ask the tool to avoid them, and cut any that slip through. Your own odd turns of phrase are worth more than polish.
Follow-up questions
How many of my own posts should I give it as examples?
A handful is enough to start, ideally posts that did well and still sound like you. Anthropic suggests three to five examples for its models, which is a sensible number for any assistant.
Swap them for newer ones every so often, so the tool learns from how you write now.
Should I just tell the AI to write in my style?
It helps a little, but showing works better than telling. A description such as friendly and direct means something different to everyone, while three real posts show exactly what you mean. Use both if you like, with the examples doing most of the work.
Is it still my post if AI drafted it?
It is if the substance came from you and you approved every word. The ideas, stories and opinions should be yours, with the tool arranging and tightening them.
If a draft says something you have never thought, cut it, however well it reads.
Talk for a few minutes. Post all week.
Hummbird turns your ideas into LinkedIn posts and videos in your voice. Nothing publishes until you approve it.
Start for free7 days free, or 14 with a card