# AI Prompts for Social Media: 20 Briefs You Can Actually Use [Read the original article]() By Garry · Founder Published: 2026-09-25T10:40:24.062Z Updated: 2026-09-25T10:40:27Z 9 min read Categories: Content creation Use 20 practical AI briefs for ideas, carousels, captions, repurposing, and review, with source requirements and clear boundaries against invented claims. ![Organized source cards, a pencil, and a blank editorial brief assembled on an ivory workspace.]() ## Key takeaways - Supply approved facts, audience context, a small deliverable, and explicit limits before asking for a draft. - Separate planning from writing so an unclear premise is resolved before polished copy is produced. - Use prompts to identify unsupported claims and missing context, while keeping a human responsible for source checking. - Save prompts with useful input examples, acceptable outputs, and recurring mistakes rather than relying on wording alone. A useful AI prompt for social media looks more like an editorial brief than a request to “make this go viral.” It names the reader, supplies reliable source material, defines a small deliverable, and explains what the draft must not invent. The twenty prompts below cover research organization, ideation, writing, repurposing, and review. Replace bracketed material with your own information. Treat every response as a draft that needs checking; none of these prompts guarantees reach, sales, or factual accuracy. ## Prepare a source pack before asking for copy Give the model a short collection of approved facts, audience questions, product details, and links you have actually checked. Separate facts from hypotheses. If a claim has no source, mark it as an assumption rather than inviting the model to fill the gap. Include a compact [brand voice guide](): a few writing rules and examples are more useful than a dozen adjectives. Specify the channel, format, audience situation, and intended next step. Remove private customer information unless your organization has explicitly approved the tool and that use of the data. Google's [guidance on generative AI content]() emphasizes accuracy, quality, and relevance. Although it concerns website content, the editorial principle applies here: generation speed is useful only when the result genuinely helps the intended reader. ## Prompts for finding a useful angle ### 1\. Turn recurring questions into a topic shortlist > Using only these anonymized audience questions, group them by the task each person is trying to complete. Suggest six social post topics. For each, give the reader situation, the question answered, and the source material needed. Flag gaps instead of inventing evidence. Questions: \[paste\]. Use the groups to improve your [content pillars](). Keep the audience's language where it makes the problem clearer. ### 2\. Find a specific angle inside a broad topic > Our topic is \[topic\], and our audience is \[audience\]. Produce five narrow angles that each answer one practical question. Explain what the reader should be able to do afterward. Exclude generic inspiration, unsupported trends, and claims outside these approved facts: \[facts\]. Select one angle before requesting a complete post. Combining all five usually creates an unfocused draft. ### 3\. Challenge an idea before drafting > Review this proposed post idea: \[idea\]. Identify who would find it useful, what is vague, what proof is missing, and what a reader could misunderstand. Suggest one narrower version. Do not write the post yet, and do not assume our audience agrees with the premise. This is useful when an idea sounds attractive in a meeting but lacks a concrete teaching point. ### 4\. Generate examples without fabricating a case study > Explain \[concept\] using three clearly labeled hypothetical examples for \[audience\]. Keep the situations plausible and avoid invented customer names, testimonials, results, or precise performance improvements. For each example, explain the decision it illustrates and the limits of the analogy. Retain the hypothetical label in the published version. A realistic story is still an invented example. ### 5\. Build a balanced content mix > Propose a two-week content outline using these pillars and available sources: \[list\]. Our realistic capacity is \[capacity\]. Include educational, decision-support, and product-explanation posts. Give each a distinct audience need and source. Leave space for current questions; do not prescribe a universal posting frequency. Review the outline against actual production time. A calendar that assumes unavailable people or assets is not ready to use. Before moving into drafting, choose one of these planning prompts and inspect its output. Ask whether the proposed reader, source, and outcome are specific enough to recognize in a finished post. If the model returns broad themes such as growth or engagement, narrow the input. A useful planning result should make the next writing decision easier, rather than create another list that needs a separate strategy meeting. ![Four input cards labeled Sources, Audience, Task, and Constraints forming a prompt brief.]() Specific inputs make the requested output easier to evaluate. ## Prompts for writing a clear first draft ### 6\. Create a carousel with one coherent promise > Draft a \[slide count\]-slide carousel answering \[question\] for \[audience\]. Use these verified notes: \[notes\]. Give each slide a short heading and one concise point. Include a concrete example before the final takeaway. Flag unsupported claims. End with \[appropriate action\], without promising performance results. Check that the slide sequence teaches a complete idea. The [Instagram carousel workflow]() can guide the subsequent editing and design. ### 7\. Write hooks that match the content > Write eight opening lines for this post outline: \[outline\]. Vary the approach across a specific problem, a useful contrast, and a direct promise. Every hook must be fulfilled by the body. Avoid fearmongering, false urgency, exaggerated results, and claims that the reader is doing everything wrong. Choose a hook based on accuracy and fit, not simply how dramatic it sounds in isolation. ### 8\. Draft a caption that adds context > Write a caption for this carousel: \[slide text\]. Add context the slides do not already repeat, explain one practical implication, and end with \[action\]. Follow these voice rules: \[rules\]. Keep the opening specific and use readable paragraphs. Do not add facts, features, or statistics. Read the caption with the visuals. Together they should feel complete without saying every sentence twice. ### 9\. Explain a product capability through a user task > Using this verified capability description \[description\], write a social post for a person trying to \[task\]. Explain the problem, the relevant capability, and one limitation or prerequisite. Use plain language. Do not claim integrations, availability, pricing, or automation beyond the supplied description. Have someone who knows the product check the draft. Fluent wording can conceal a technically incorrect implication. ### 10\. Offer calls to action at different commitment levels > Suggest six calls to action for this post and objective: \[post and objective\]. Include options for learning, saving, discussing, and visiting the supplied resource. Explain when each fits. Avoid fake scarcity, pressure, and requests unrelated to the reader's next useful step. Choose one primary action. Several competing requests can make an otherwise clear post feel indecisive. ## Prompts for adapting useful source material ### 11\. Extract social ideas from an article > From this article \[text\], identify five standalone social ideas. For each, provide the exact supporting passage, a proposed format, and what context must remain. Do not introduce new statistics or turn a qualified claim into an absolute. Suggest where a link to the full article is necessary. Check the quoted supporting passages against the original. Use the [blog repurposing guide]() to organize the resulting assets. ### 12\. Turn a webinar transcript into a useful sequence > Review this transcript excerpt with timestamps: \[excerpt\]. Propose four social posts that each teach one complete point. Include the supporting timestamp, any caveat, and whether a speaker quotation needs confirmation. Do not fabricate quotations, visual demonstrations, or claims from sections not provided. A transcript may contain errors. Listen to the relevant recording before attributing a quotation or publishing a clip. ### 13\. Adapt a message for another channel > Adapt this approved message \[message\] for \[destination channel and format\]. Preserve the facts and intended audience action. Change the opening, structure, and context where needed. Note any link-placement or visual requirement for a human to verify. Do not invent platform specifications. The [cross-posting guide]() explains why adaptation is more than pasting the same caption into several boxes. ### 14\. Refresh an evergreen post responsibly > Compare this older post \[text\] with these current approved facts \[facts\]. Identify statements that remain valid, statements that need updating, and claims you cannot verify. Propose a revised angle only after that comparison. Do not imply that changing the date alone makes the advice current. Save the change notes in your [evergreen content library]() so the next review starts with useful context. ### 15\. Create an image brief that teaches the point > Propose three distinct visual concepts for this post: \[post\]. Each concept should explain one idea or decision through a simple diagram or concrete illustration. Specify composition, exact short labels if needed, and descriptive alt text. Do not fabricate a product interface, customer photograph, or analytics screenshot. Review whether the visual adds information. A decorative image can support the mood, but should not pretend to be evidence. ![A verified source passage branching into three different deliverables: carousel, caption, and diagram.]() One supported idea can become several formats while preserving its factual core. ## Prompts for reviewing before publication ### 16\. Audit claims against supplied evidence > List every factual or product claim in this draft \[draft\]. For each, identify the supporting item in this source pack \[sources\], or mark it unverified. Preserve important qualifiers. Highlight claims that sound stronger than the evidence. Do not search your memory for a replacement statistic. This creates a review list, not a guarantee. A person still needs to inspect the sources and confirm what they support. ### 17\. Edit for a specific voice without changing facts > Edit this draft \[draft\] using these voice rules and examples \[guide\]. Preserve factual meaning, limitations, and links. Replace vague praise and jargon with concrete language. After the edit, list the three most important changes and flag any sentence whose meaning was uncertain. Compare the before and after versions. A smoother sentence is not an improvement if it loses a necessary qualification. ### 18\. Review readability and accessibility > Review this caption and visual text \[content\] for unfamiliar terms, long sentences, unclear reading order, and meaning conveyed only through color. Suggest specific edits. Draft descriptive alt text for the supplied visual description \[description\]. Do not claim to have inspected an image that was not provided. Follow [W3C's writing accessibility guidance]() when checking headings, descriptive links, and clear instructions. ### 19\. Prepare careful replies to common comments > Using these approved answers and response boundaries \[notes\], draft replies to \[anonymized comment examples\]. Acknowledge the question, answer what is supported, and identify when a human should follow up. Do not promise refunds, deadlines, account fixes, or exceptions outside the supplied policy. Keep replies specific to the actual conversation. Sensitive or unusual situations deserve a person's judgment rather than automatic reuse. ### 20\. Turn a performance review into a testable hypothesis > Review these anonymized post metrics and definitions \[data\]. Separate observations from possible explanations. Propose three small content experiments with a clear change and a consistent measurement window. Identify missing information. Do not infer causation, statistical significance, or customer intent from the numbers alone. Use your [engagement rate definitions]() in the source pack so the model does not compare incompatible percentages. ## Use a two-pass workflow instead of one giant request First ask for structure, missing evidence, and a proposed angle. Resolve the important gaps before asking for polished copy. This reduces the chance that you spend time editing a smooth draft built on an unclear premise. In the second pass, provide the approved structure and request the final format. Keep the source pack attached or clearly available in the conversation. If the draft introduces a new claim, return to the evidence rather than asking the model to sound more confident. A human review should check the actual destination link, product behavior, quotation, and image. The [FTC's advertising and marketing guidance]() is a useful reminder that marketing claims need appropriate support. AI assistance does not change the responsibility for what a business publishes. ## Save prompts with the examples that make them work A reusable prompt library should include the prompt, a suitable source pack, an example of an acceptable result, and notes about common mistakes. Record the model or tool used when it helps reproduce a workflow, but do not assume the same wording will always produce the same output. Keep prompts narrow enough to evaluate. “Make our social strategy better” gives the reviewer little to check. “Extract four supported audience questions from these notes” has a clearer boundary and a more inspectable result. Caroush's AI carousel and social post creation can be part of a content production routine. Apply the same evidence, voice, and review standards regardless of which tool produces the draft. The strongest prompt is the one that helps your team produce a useful, accurate post and recognize when the material is not ready yet. ## Sources - [Google Search Central: Guidance on using generative AI content]() - [W3C Web Accessibility Initiative: Writing for web accessibility]() - [Federal Trade Commission: Advertising and marketing guidance]() ## Frequently asked questions ### What should a social media AI prompt include? Include the audience situation, source material, channel and format, intended outcome, voice rules, and limits on unsupported claims. Define a deliverable a person can inspect. ### Can these prompts guarantee accurate content? No. A model can misunderstand sources, lose qualifiers, or introduce unsupported details. Verify factual claims, quotations, links, product behavior, and visual descriptions before publication. ### Should I ask AI to write an entire content strategy at once? Smaller tasks are easier to evaluate. Start with audience questions, evidence gaps, and a proposed angle, then request a draft after resolving the important decisions. ### Can I use AI to create customer examples? You can ask for clearly labeled hypothetical examples. Do not present invented names, testimonials, results, or performance improvements as real customer evidence. ## About the author Garry Gaurav Sapkota builds Caroush, a workspace for creating, scheduling, and publishing social content. - [https://x.com/gauravsapkotanp]()