Which prompt to choose
Each channel has a different physics: social lives on the scroll and the first three words, the article on depth, the newsletter on intimacy. Using the same prompt for all three is the number-one reason AI texts resemble each other. Choose based on where you publish.
- Social post (Instagram, LinkedIn, Facebook): first prompt, built on a strong opening hook and a clear call to action.
- Blog article: second prompt, which asks for structure, concrete examples, and a non-obvious angle.
- Newsletter: third prompt, which goes for a person-to-person tone and a single idea per email.
How to do it
- Open the AI assistant you prefer.
- Copy the prompt for the right channel.
- Swap in the topic, the audience and — this is the piece almost everyone skips — paste two lines of a past text of yours that you liked, so the AI imitates your voice instead of inventing a generic one.
- Send, read, correct out loud: "shorter," "drop the opening question," "make the ending less salesy."
The operational syntax for a social post:
Act as a social media manager who writes for authentic profiles, not corporate ones.
Write 3 versions of a post on this topic: "why I stopped answering emails after 7 p.m."
Audience: freelancers who overwork and feel guilty about switching off.
For each version: a hook in the first 8 words, a body of 4-5 lines, a closing question that invites comments.
Constraints: no rapid-fire emojis, no poster-style motivational lines. The tone of someone telling a true thing that happened to them.
The operational syntax for a blog article:
Act as a journalist who's an expert in my field.
Write the outline and the first paragraph of an article on: "how to choose an accountant."
Audience: someone registering as self-employed for the first time.
I want a non-obvious angle: start from a common mistake, not from a definition.
Format: title, 5 subheadings, and under each a line on what it'll cover.
Constraints: no opening like "in an increasingly digital world." Start with a fact or a real situation.
The operational syntax for a newsletter:
Act as a newsletter author who writes to a friend, not to a list.
Write an email on: "one thing I learned by getting it wrong this week."
Audience: people who follow me for practical advice on self-employment.
A single idea, developed well. Open with the scene, close with what the reader can do.
Constraints: no "I hope this email finds you well." Get straight to it. At most 250 words.
After the first draft, read it out loud. If you stumble on a sentence or it sounds fake to you, that's where the AI put it on autopilot: flag it and ask "rewrite this sentence the way a real person would say it."
A real example
Luca runs the LinkedIn profile of a small agency. He had to announce a new service and his first by-hand attempt came out brochure-like. He pasted the social prompt, set the audience as "marketing managers at SMEs who don't have time," and added two lines of an old post of his that did well as an example of tone.
The AI produced three versions. The first started with "We've launched a new service" (trashed). The third started with "A customer last week told me something that stuck with me." Luca kept that one, swapped the example for a real one of his, and published it. Time: five minutes instead of half a day of putting it off.
When it does NOT work (and how to fix it)
If all the posts come out with the same rhythm and the same phrases
The AI has a default style of its own and falls back into it if you don't stop it. The remedy is the example: always paste two or three lines of a text of yours that worked and write "imitate this register, not your own." It's the most underrated technique and the most effective: showing an example calibrates the answer better than any adjective about the tone.
If the article is long but says nothing
This happens when you ask "write an article on X" without an angle. The AI fills space with encyclopedia definitions. The way out: impose an angle in the prompt ("start from a common mistake," "tell it from the point of view of someone who gets it wrong"). An angle forces the text to have a point of view, and a point of view always has something to say.
If the newsletter sounds like an advertisement
Add the constraint "no aggressive calls to action, no fake urgency." And cut the last sentence yourself: nine times out of ten the AI closes with one too many commercial invitations. A newsletter that gives value and nothing else sells more than one that asks you to buy on every line.
A tip from someone who actually uses it
Build your "voice prompt" once: a fixed block that describes your tone, your audience, and includes two examples of your best texts. You save it in a note and paste it at the top of every request. You stop re-explaining who you are in every conversation, and consistency from one piece of content to the next stops being a problem.
Frequently asked questions
How much do I have to rewrite by hand after the AI?
Count on 20-30 percent: the opening hook, the concrete example (which has to be yours and true), and the last sentence. The central body the AI does well. If you're rewriting everything, the problem is in the prompt, not the text: give it more context and an example of the tone.
Can I schedule posts directly from the AI?
The AI writes the text, but the publishing is done by you or a scheduling tool (you'll find it in the platform's settings or in dedicated apps). If you can't find the button to schedule, look for "schedule" in the publishing menu: the name changes often, the feature is almost always there.
Do social platforms penalize AI-written content?
No, and here lies the misunderstanding to clear up. The algorithms measure whether people stop, read, comment, and share. They don't have an "AI-written" sensor. A post that generates real reactions gets pushed, who wrote it is irrelevant. What sinks reach is boring or copied content, not the author. The judge stays the reader's reaction.