AI prompts for social media, by the job they do
Prompt libraries usually fail because they are written for a blank page. These are written for material you already have, which is the only kind that produces something you can post.
7 minute read
Why most prompt libraries produce unusable output
The typical social media prompt is a request to invent: write me ten Instagram captions about productivity. What comes back is generic because you gave the model nothing that only you have. Every prompt below is built the other way — it takes something you already produced, and asks the model to do a mechanical operation on it. That single change is what separates output you can post from output you delete. This post is the applied companion to our guide on which prompt techniques actually survive testing; the reasoning behind the shapes lives there, the ready-made prompts live here.
Finding the clips in a long recording
Paste in a transcript with timestamps and ask: "Here is a transcript of a [length] recording. Find every passage that would stand alone as a 30 to 60 second clip. A passage qualifies only if it makes one complete point without needing setup from earlier in the recording. For each, give me the start and end timestamps, the sentence that should open the clip, and one line on why it stands alone. Rank them by how well they work without context. Do not include anything that needs a caveat to make sense." The constraint about standing alone is doing the work — without it you get topic summaries rather than cuttable moments.
Writing hooks that are not lies
"Here is the transcript of a clip: [paste]. Write five opening lines. Each must be true to what is actually said in the clip — do not promise anything the clip does not deliver. Vary the type: one that states the surprising conclusion, one that names the mistake being corrected, one that asks the question the clip answers, one that leads with a specific number if there is one, and one that starts mid-story. Keep each under twelve words." The instruction not to over-promise matters more than any phrasing trick, because the retention damage from a hook that oversells is worse than a soft opening.
One recording, four platforms
"Here is a caption I wrote for [platform]: [paste]. Rewrite it for LinkedIn, X, Instagram and a YouTube description. Keep the substance identical — do not add claims that are not in the original. Change only length, formatting and opening line to fit each platform's conventions. For LinkedIn, lead with the conclusion and use short paragraphs. For X, cut to the single sharpest sentence. For Instagram, front-load the first line because it is what shows before the fold. For YouTube, write it to be read after watching, not before." Ask for one platform at a time if the output starts to blur together.
Mining your comments for the next month of content
"Here are the last [n] comments on my posts: [paste]. Group them into recurring themes. For each theme, tell me how many comments it covers, quote the two most representative, and say whether it is a question I have not answered, disagreement with something I said, or a request for something I have not made. Rank the themes by how many people raised them. Ignore compliments." The instruction to ignore compliments is not modesty — praise crowds out the signal, and the useful material is in the questions and the pushback.
Turning a clip into a carousel
"Here is a clip transcript: [paste]. Turn it into a five-slide carousel. Slide one is the claim, stated flatly with no hype. Slides two to four are one supporting point each, in the order the transcript makes them. Slide five is what the reader should do differently. Maximum fifteen words per slide. Do not add any point that is not in the transcript." The word limit is what makes it usable — without it you get paragraphs that will not fit on a slide, and you end up rewriting anyway.
The two rules that make any of these better
First, always give it your material. Every prompt here starts with a paste, and that is not an accident — the model's job is transformation, not invention, because invention is the part where it sounds like everyone else. Second, always state what it must not do. "Do not add claims that are not in the original" and "ignore compliments" and "do not include anything that needs a caveat" are each carrying more weight than any amount of persona setup. Constraints are the part of a prompt that reliably changes the output.
FAQ
What makes a good AI prompt for social media?
Giving it your own material and stating explicit constraints. Prompts that ask a model to invent content from a topic produce generic output; prompts that ask it to transform a transcript, caption or comment thread you already have produce something only you could post.
Do elaborate personas and role-play prompts help?
Less than most advice claims. The techniques that hold up under testing are about structure, position and examples rather than personas. Our guide to AI prompt techniques covers what has actually been measured.
Can AI pick which parts of my video to clip?
It can shortlist candidates from a transcript, which saves the scrubbing. The quality depends entirely on the constraint you give it — asking for passages that stand alone without setup produces cuttable moments, while asking for highlights produces topic summaries.
Sources
Related pages
Keep reading
More AI →Does AI-assisted content actually perform worse?
Three credible studies asked whether AI content underperforms and got three different answers. The disagreement is the finding — and it tells you exactly where the line is.
8 minute readHow creators and teams actually use AI
The gap between what people use AI for and what they admit to using it for is the whole story — and it is why a written policy beats good intentions.
7 minute readHow to create an Instagram post with AI
The end-to-end workflow, and the one decision that matters more than which tool you pick: which parts of the post you let AI touch.
7 minute readTurn one long video into a week of posts
300 credits for 3 days · no card.
Start free