There's a specific kind of LinkedIn post you've definitely seen. It starts with a single sentence on its own line. Then another. Then a pseudo-profound observation about entrepreneurship or resilience. Then a list of bullet points that could apply to any business in any industry at any point in history. You scroll past it. Everyone scrolls past it. And yet it keeps getting posted, because someone, somewhere, discovered that a particular AI prompt produces it reliably and confuses output volume for content strategy.
This is the generic AI social media problem. It is not caused by AI itself — it is caused by using AI as a replacement for thinking rather than as an accelerant for it. The distinction matters enormously for how you approach your workflow, and the gap between those two approaches is the difference between social media that builds an audience and social media that fills a queue.
Training AI on Your Voice Before You Use It
The single most important thing you can do before using AI for any social media content is give it an accurate representation of how you actually write and communicate. Most people skip this step entirely and then wonder why everything AI produces sounds generic. It sounds generic because you gave it a generic brief.
Building a voice document
Before you write a single prompt asking for social content, create a voice document. This doesn't need to be formal — a simple text file works. It should contain:
- Five to ten examples of your best past posts — ones that got real engagement, felt authentically like you, and said something worth saying. These are the ground truth of your voice.
- Words and phrases you use often — the specific vocabulary that is yours. If you always say "genuinely" instead of "truly," that belongs in here. If you habitually use short sentences for emphasis, note it explicitly.
- Words and phrases you never use — the ones that make you wince. "Exciting journey." "Game-changer." "Thrilled to announce." List them. A good AI will avoid them if you tell it to.
- Your perspective on your subject area — not what the industry consensus is, but where you differ from it. What's your contrarian take? What do you say that most people in your space don't? This is what makes your content worth following.
Include this document in every prompt session where you're generating social content. Paste it in. Make it the first thing Claude or ChatGPT reads before you ask for anything. The output quality will improve immediately and substantially.
Shortcut: Paste five of your best existing posts and say: "Analyse the voice, tone, sentence structure and perspective in these posts. Use this analysis as the style guide for everything you write for me today." This takes 30 seconds and dramatically changes the quality of what follows.
Platform-Specific Prompting: Why One Brief Doesn't Work Everywhere
Every platform has its own grammar — not just formatting conventions, but the kind of content that earns attention and the kind that earns dismissal. AI doesn't know which platform you're writing for unless you tell it explicitly, and "write me a social post" will produce something that fits nowhere perfectly.
LinkedIn rewards specificity and professional perspective. The posts that perform are the ones that share a concrete experience, a specific lesson, or a point of view that other professionals in the field would find worth discussing. What doesn't work: motivational platitudes, personal branding fluff, and anything that reads like it was written to perform engagement rather than share genuine insight.
When prompting for LinkedIn, include: your profession, your target audience's role and seniority, the specific experience or lesson you want to convey, and the action or reaction you want the reader to have. "Write a LinkedIn post for a web designer targeting marketing managers at SaaS companies, sharing the observation that most SaaS homepages bury their value proposition below the fold. Tone: direct, a little blunt, no inspirational framing."
Instagram is visual-first, which means the caption is secondary to the image or video — but it still needs to do specific work. Captions should either deepen the context of the visual, invite genuine interaction, or tell a short story that adds a layer the image alone can't. What doesn't work: long paragraphs, corporate tone, and calls-to-action that feel transactional.
When prompting for Instagram captions, describe the visual first and ask AI to write copy that complements rather than simply restates it. "The image shows a website redesign before/after. Write a caption that focuses on the story behind the decision to redesign, not on the features we changed." Let the brief lead with the story angle.
X (Twitter)
X rewards opinions and takes — the more specific and disagreement-provoking, the better for reach. It also rewards genuine expertise communicated concisely. What AI tends to produce for X without specific guidance is either too safe (no real angle, no controversy) or artificially pithy in a way that reads as performative.
When prompting for X, push for the specific claim and the specific reasoning. "Write a tweet arguing that most small business websites are over-designed. Keep it under 200 characters. It should prompt disagreement from web designers and agreement from business owners." The platform-specific conflict you're writing for is part of the brief.
The platform isn't just a distribution channel. It's a context that determines what your content means. A post that works on LinkedIn will often actively damage your brand on X, and vice versa. Brief your AI accordingly.
What to Never Automate — and a Batch Creation Workflow
Some content categories should never be AI-generated, no matter how good your prompting practice is. Not because AI can't produce the words, but because the value of the content comes specifically from it being you, in the moment, without an intermediary.
- Real-time responses to trending conversations. Your takes on breaking news, industry developments, or community conversations need to be yours. Audiences can sense when an account is running on a delay and AI intermediation makes the delay obvious.
- Personal stories. AI cannot tell your story. It can help you structure it after you've written a rough draft, but the first-person specific narrative — the client who said the thing, the project that went wrong, the moment you changed your mind — has to come from you.
- Comment and reply engagement. The conversations that build community happen in the comments. Automated or AI-drafted replies kill trust faster than anything else. Write these yourself.
- Your opinion on contested topics. If you're sharing a genuine point of view on something where reasonable people disagree, it needs to be your actual view. AI-generated opinion posts are identifiable by their careful balance and refusal to commit. Audiences don't follow brands for balance — they follow for perspective.
A batch creation workflow that actually works
The highest-leverage application of AI in social media is batch creation: producing a week or month's worth of core content in a single session, then editing and scheduling it. Here is a workflow that preserves voice while extracting real time savings.
- Choose your content pillars for the batch — the two or three themes you'll post about this week. Don't let AI choose these. They should come from your actual business priorities and what you genuinely have something to say about.
- Write one rough post per pillar yourself — rough is fine, even notes or bullet points. This is the raw material. It contains your actual perspective.
- Give AI your voice document, your rough notes, and the platform you're writing for. Ask it to develop the rough notes into fully formed posts while maintaining the specific language and perspective in the voice document.
- Edit every output before scheduling. This edit pass is not optional. Budget 10 minutes per post. You're looking for: anything that sounds generic, any phrase that isn't yours, any loss of the specific point of view you had in your rough notes.
- Now ask AI to repurpose each post for the other platforms you need, with a specific brief for each. A LinkedIn post becomes a shorter, more opinionated X thread. An Instagram caption gets written from the same story with a different hook.
Engagement versus reach is worth addressing here: AI-optimised content tends to optimise for reach, because reach is the metric most associated with visible performance. But engagement — comments, saves, shares from your specific target audience — is almost always more valuable for business outcomes than raw impressions. Brief your AI explicitly for engagement by including the specific audience whose reaction you care about, and by asking for a hook that prompts them to respond rather than one that prompts them to scroll.
The brands using AI well on social media are not posting more. They're posting more thoughtfully, with less of the preparatory labour that used to consume their creative time. The AI handles the heavy lifting of structuring, drafting, and reformatting. The human handles the parts that can't be faked: the real perspective, the specific story, and the judgment about what's actually worth saying today. That division of labour produces content that sounds like a person — because it is.