The problem this solves
Short-form video is decided in the first second, and most teams write hooks by instinct. The evidence of what actually stops a scroll in your category exists - it is sitting in the ads your competitors keep paying to run - but mining it means watching hours of video and transcribing openers by hand. Nobody does that, so hooks get written in a vacuum, and the clip that took a day to produce dies in the feed within half a second.
Production compounds the problem. Even when a strong hook exists on paper, turning it into a finished vertical clip requires scripting, visuals, and editing time that small teams rarely have. So the hook doc grows while the account posts nothing, and the platforms that reward consistent short-form output keep compounding reach for whoever ships. The gap is rarely a lack of ideas; it is the distance between a written opener and a clip that is actually ready to post.
How the mission runs
- Mine the winning hooks. The agent pulls long-running video ads in your category from Ad Libraries - longevity is the tell that a hook is paying for itself - and uses Video Transcripts to extract exactly how each one opens. The output is a pattern map: which opener structures your market demonstrably responds to.
- Translate patterns into your voice. Winning structures get rewritten around your product and tone. A pattern is a skeleton; the agent fleshes each one out with your actual value proposition and vocabulary, so the openers sound like your brand borrowing proven mechanics rather than an imitation of somebody else's ad.
- Write four scroll-stopping openers. You receive four complete hooks, each annotated with the pattern it uses and the evidence behind it. Reviewing four takes minutes in the thread, and your reaction - which two feel most like the brand, which claim lands hardest - decides what moves to production.
- Generate the two strongest as clips. Video Generation produces the two selected hooks as vertical clips built for Reels and TikTok: the opener hits immediately, the visual supports the claim, and the pacing fits the platform. Scripts and visual direction are confirmed in the thread before rendering, so generation spend goes only to approved concepts.
- Deliver the battery. Both finished clips land in the thread ready to download and post, alongside the two unproduced hooks kept in reserve. When you want the next pair produced, or fresh patterns mined after the field rotates, the follow-up mission picks up exactly where this one ended.
The prompt
This is the exact objective the agent receives. Swap the obvious placeholders for your own domain, segment or channel and run it as-is from the console, Slack, or the API.
What comes back
Two finished vertical clips, generated from the strongest of four evidence-backed hooks and formatted for Reels and TikTok, delivered ready to download and post. Alongside them: all four written openers with the pattern and competitive evidence behind each, so the two held in reserve are ready for the next production pass. The hook research itself is preserved in the thread as a reusable map of what opens video in your category.
Make it yours
- Aim the battery at one campaign: mine hooks only from ads selling against your exact use case, and generate both clips around a single product moment.
- Flip the sourcing: transcribe your own best-performing videos alongside competitor ads, and have the agent isolate what your audience specifically rewards before writing new openers.
- Make it a monthly rhythm: fresh transcript mining, four new hooks, two new clips, with last month's reserve hooks promoted or retired based on how the posted pair performed.
Frequently asked questions
Will the clips look like my brand?
Yes. Visual direction is set from your brand kit before rendering - palette, tone, and the kind of imagery your accounts already use - and you approve the script and visual plan in the thread first. The clips borrow proven hook mechanics while looking unmistakably like your feed.
What if neither generated clip feels right?
Say so in the thread and the loop continues. The agent can produce the two reserve hooks instead, revise the visual treatment on the originals, or return to the transcript evidence and write a new set. Rendering happens only after you approve a script, so retries stay controlled.
What format do the clips arrive in?
Vertical, platform-native aspect ratio, short enough to hold completion rates, delivered as downloadable files in the thread. They are ready to upload directly to Reels or TikTok. Nothing is posted by the mission itself; publishing is always your call, from your own accounts.