Prompts to vibe-code your own marketing tools. Paste one into your AI editor and build it in an afternoon. A growing list.
Short-form clip finder
Video
Build a web app that turns a long video into ready-to-post short clips.
Input: a YouTube URL or an uploaded MP4.
Steps: pull the transcript (Whisper, or the YouTube transcript API). Then have an LLM score every 20 to 60 second window for hook strength, payoff, and whether it stands on its own. Return the top 5 windows with a title, the start and end timestamps, and a one line reason.
Output: render each pick as a card with an inline player trimmed to the window, a generated caption, and an Export button that burns in captions and crops to 9:16 with ffmpeg.
Stack: Next.js, a transcription API, an LLM for scoring and captions, ffmpeg for export. Keep it single user and local-first.
UGC ad script writer
Ads
Build a tool that writes UGC style ad scripts from a product URL.
Input: a product page URL, plus optional brand voice notes.
Steps: scrape the page for the product name, key benefits, price, and any proof. Then generate 10 scripts, each 20 to 30 seconds, in a different angle: problem-solution, before-after, founder story, honest review, unboxing, comparison, myth-busting, day-in-the-life, social proof, and urgency. Each script has a hook (first 3 seconds), the body, a call to action, and a simple shot list.
Output: a table you can sort by angle, with a Copy button per script and an Export all to CSV.
Stack: Next.js, a scraping step, one LLM call per batch. Let the user paste brand voice notes to steer the tone.
Competitor ad board
Research
Build a dashboard to study a competitor’s ads.
Input: a brand name or domain.
Steps: pull their currently running ads, starting from the Meta Ad Library, which is public. For each ad, capture the creative, the copy, the format, and how long it has been running. Have an LLM tag every ad by hook type, offer, and emotional angle.
Output: a filterable grid. Let the user filter by format and hook, sort by run length (a longer run usually means it is working), and save ads to a swipe file.
Stack: Next.js, a scraper or an Apify actor for the ad library, an LLM for tagging, and a small database for the swipe file.
X reply opportunity finder
Social
Build a tool that finds good reply opportunities on X.
Input: a few topics or accounts to watch, and a short description of what I do.
Steps: poll recent posts on those topics. Score each for relevance to me, how much attention it is getting, and whether a helpful, non-salesy reply could genuinely add value. Drop anything that looks like ragebait.
Output: a feed of the best opportunities, each with the post, why it is a fit, and a suggested reply draft in my voice that I can edit before posting. Never auto-post.
Stack: Next.js, the X API or a scraping fallback, an LLM for scoring and drafting. Refresh on a schedule.
AEO visibility checker
AEO
Build a tool that checks whether AI answers mention my product.
Input: my product name, my category, and a list of buyer questions.
Steps: for each question, query the major answer engines (ChatGPT, Perplexity, Gemini) through their APIs. Detect whether my brand is mentioned, in what position, and which sources they cite. Compare against two competitors I name.
Output: a scoreboard showing my mention rate per engine, the questions where I am missing, and the sources getting cited that I am not on yet.
Stack: Next.js, the answer-engine APIs, an LLM to parse mentions and sources. Re-run weekly and chart the trend.
