20 AI prompts for better social media posts

A useful social media prompt behaves like a good assignment. It names the reader, the job, the source material, the boundaries and the expected output. The 20 prompts below are designed to improve human thinking and editing, not ask a model to invent expertise.

Replace the bracketed fields. Keep source notes and sensitive data within your organisation's approved tools and policy.

Prompts for finding the point

1. Extract a defensible idea from notes

Read the source notes below. Identify up to five claims that are directly supported. For each, quote the supporting note, explain which reader would care and flag any missing evidence. Do not write a post yet. Source notes: [notes]

2. Find the decision behind an update

The update is [event]. Ask me seven questions that uncover the decision, trade-off, constraint and changed practice behind it. Do not suggest inspirational lessons.

3. Separate facts from interpretation

Divide these notes into verified facts, the author's interpretation, assumptions and claims requiring a source. Return a table. Do not complete gaps. [notes]

4. Find useful disagreement

The common advice is [advice]. Using only the evidence below, identify the conditions where it works, where it fails and the strongest fair objection to my view. [evidence]

Prompts for audience and relevance

5. Clarify the intended reader

Compare how this idea matters to [reader A] and [reader B]. List their different decisions, vocabulary and likely objections. Recommend one primary reader for a single post.

6. Remove background the reader does not need

Mark each paragraph as essential context, useful evidence or removable background for [reader]. Explain any cut in one sentence. [draft]

7. Check professional value

After reading this draft, state the decision a reader can make better. If there is none, say so and ask for the missing evidence. [draft]

Prompts for structure

8. Create three honest openings

Create three openings for this approved point: direct, evidence-led and question-led. Use no suspense, exaggerated claim or invented result. Keep each under 35 words. [point]

9. Choose a post structure

Recommend one structure from observation, argument, process, comparison or short story. Explain why it fits the source. Return an outline only. [source]

10. Build a carousel concept

Turn the approved material into a maximum of eight carousel slides. Give each slide one distinct job. Include the source or limitation on the relevant slide. Do not stretch one idea across multiple slides. [material]

11. Condense without deleting evidence

Reduce this draft to [length]. Preserve the central claim, one concrete example, every necessary qualification and source links. List anything important you had to remove. [draft]

Prompts for voice

12. Compare with approved examples

Compare the draft with the approved examples. Identify differences in directness, vocabulary, sentence rhythm, evidence and calls to action. Recommend edits without copying phrases from the examples. [draft and examples]

13. Remove generic AI patterns

Flag broad openings, inflated adjectives, repeated three-part lists, unsupported certainty, abstract conclusions and contrast formulas. Explain the problem and propose a more specific revision using only facts already present. [draft]

14. Adjust expertise level

Rewrite for [audience] who already understands [known concepts]. Remove basic definitions they do not need, define unfamiliar terms and retain the evidence. [draft]

Prompts for verification and risk

15. Create a fact-check table

Extract every name, date, number, quotation, product capability and policy claim. For each, show the supporting source from the provided material or mark SOURCE NEEDED. Do not browse or infer. [draft and sources]

16. Run a confidentiality check

Identify details that may reveal a customer, employee, private project, personal data or unpublished strategy. Suggest a redaction or aggregation. Do not decide whether publication is legally safe.

17. Find misleading implications

List claims that are literally true but may create an inaccurate overall impression because context is missing. Suggest the smallest clarification needed. [draft]

Prompts for adaptation

18. Adapt rather than copy

Adapt this approved LinkedIn post for [platform]. Preserve the core claim and evidence. Change the opening, structure, length and call to action for the platform. Explain the five most important changes. [post]

19. Create a follow-up, not a repeat

From the comments and original post, identify unanswered questions, disagreements and examples. Propose five follow-up posts with a new search intent. Do not restate the original. [post and comments]

20. Build an editorial review

Review this draft for accuracy, usefulness, specificity, voice, accessibility, platform fit and commercial pressure. Score each from one to five, quote the evidence and give the three edits with highest impact. [draft]

Combine prompts into a controlled workflow

Do not run all 20 prompts for every caption. Use the smallest set that matches the risk.

For a researched LinkedIn post:

1. Run prompt 3 to separate claims.
2. Use prompt 8 for openings.
3. Draft or edit the post.
4. Run prompt 15 for fact checking.
5. Use prompt 20 for final review.

For a platform adaptation, start with prompt 18, then check voice with prompt 12 and risk with prompt 17.

SignalGenie can maintain content and profile context inside a repeatable workflow, but the source owner should still approve the final draft. The brand voice profile gives these prompts stronger, more consistent constraints.

Use the AI slop checklist to reject generic output and the LinkedIn AI workflow to test prompts against real source material.

Frequently asked questions

Should prompts tell AI to act as an expert?

Only a role label is rarely enough. Give the model sources, boundaries, audience and output criteria. Do not use a fictional role to justify invented expertise.

Are longer prompts always better?

No. Add information that changes the task. Long, conflicting instructions can make review harder. Stable voice rules can live in a reusable profile.

Can these prompts stop hallucinations?

They reduce avoidable gaps but do not guarantee accuracy. Verify output against primary sources and retain human approval.

Should I ask AI to make a post viral?

No prompt can guarantee distribution. Ask for clarity, relevance and evidence. Avoid manufactured drama and engagement bait.

Sources and further reading

• Prompt engineering, OpenAI.
• Writing with ChatGPT, OpenAI Academy.
• Safety best practices and human review, OpenAI.