AI can reduce the mechanical work of turning an idea into a social draft. It cannot supply the experience, evidence, or point of view that makes the draft worth reading.
Generic output usually begins with a generic input: "write a LinkedIn post about leadership." The model has no specific event, audience, constraint, or opinion to work with, so it assembles a plausible summary of familiar advice. The remedy is not a more elaborate role-play prompt. It is better source material and a clearer editorial task.
Begin With an Evidence Pack
Collect what the post is allowed to say before asking for prose:
The event or observation: what actually happened.
The useful detail: a number, quote, decision, mistake, screenshot, or customer question.
Your interpretation: what you think it means and where you may be uncertain.
The reader: who should care and what decision they face.
The constraint: facts, names, confidential details, and phrases the model must not change or invent.
This can be five rough bullets. A transcript excerpt, voice note, or client-safe case summary is even better. The model's job is to organize that material, not replace it.
Generate Angles Before Drafts
One source can support several honest angles. Ask for possible interpretations tied to the supplied evidence: a tactical lesson, a counterintuitive result, a failure analysis, a decision framework, or a question for peers.
Reject angles that require facts the source does not contain. Choose one based on the campaign goal and the reader's context. A product launch may need a concrete use case; an authority campaign may need a strong, defensible opinion; a discussion post may frame a real tradeoff.
Selecting the angle first prevents a common failure where a draft opens with one promise, tells a different story, and ends with an unrelated call to action.
Give the Draft a Clear Contract
Specify the platform, deliverable, audience, purpose, and required evidence. For example:
"Using only the source below, draft one LinkedIn post for independent consultants. Open with the project decision, explain the constraint, include the exact result, and end by asking how readers handle the same tradeoff. Preserve the quoted phrase. Do not add numbers, customer reactions, or hashtags. Flag any claim that the source does not support."
This is more reliable than asking the model to "sound human" or "make it viral." Those phrases describe a wish, not an editable output.
Adapt the Idea for the Platform
Platform adaptation changes the reading experience while preserving the claim.
An X thread needs a concise sequence in which each post contributes something new. A LinkedIn post can carry more professional context and evidence. Telegram can preserve a longer narrative and contextual links. An Instagram carousel needs a visual progression and a caption that adds context. Email needs a subject, preview line, and one primary action.
Do not ask for seven variants in one undifferentiated prompt. Define the contract for each destination or use a repurposing workflow that already stores those rules. Then review every result against the same source.
Edit for Voice and Accuracy
Start with facts: names, dates, chronology, quotations, links, and measurements. Then check the argument. Does the evidence actually support the conclusion? Has an opinion been rewritten as a certainty?
Next, remove language you would never say: ceremonial introductions, generic transitions, inflated adjectives, fake urgency, and repeated summaries. Restore specific words from your source where they carry personality. Read the post aloud; awkward rhythm is easier to hear than to see.
Finally, check the call to action. It should follow naturally from the post rather than switch suddenly from education to a sales request.
Do Not Let AI Lead These Parts
Do not let a model invent first-hand stories, customer quotes, research findings, testimonials, or current-event reactions. It can help structure material you provide, but a plausible detail is not evidence.
Use extra care with medical, legal, financial, employment, and reputation-sensitive claims. Verify changing information with authoritative current sources. Protect confidential client material and personal data before placing it into any external AI service.
Humor and strong opinions also need ownership. The model can offer alternatives, but you should be willing to defend the final words under your own name.
Choose a Tool by Workflow, Not Hype
A general chat model is useful for exploration, unusual transformations, and one-off drafts. A dedicated repurposing tool is useful when you repeatedly convert source material into the same platform set and need saved brand rules, native previews, link handling, and exports.
Test either option with the same three source types: a case, a strong opinion, and a practical guide. Compare factual preservation, editing time, platform fit, and how often the output invents connective details. The best tool is the one that produces reviewable drafts from your evidence, not the one that produces the most text.