How to Plan a Launch Video, and the Post Around It
What I learned from studying 48 of the most-viewed software launches on X
Stripe launched its Link wallet for agents with a 28-second video. It opens on the command a developer would type to install it. Then an agent plans a road trip, asks to spend $180, and a phone shows an approval request that the owner confirms with Face ID. There's no voiceover and no founder talking to camera. By the end you know exactly what the product does and why you can trust it. The launch has more than 5,000 bookmarks.
This year I studied 48 of the most-viewed software launches on X. For each one I pulled frames from the video, read the post and its thread, and read the replies to see what people actually said. Below is how I'd plan a launch now. Every point links to a launch that did it, so you can watch it yourself.
Decide what the video has to prove
Most weak launch videos try to say everything the product does. The strong ones prove one thing.
Stripe's one thing is that an agent can buy for you and you still approve every purchase. Fish Audio's is that its AI voice can hold a real phone call fast enough to sound human, so the video makes a live call to a pizza shop with the response time on screen. Matic's is that you can point at a mess and the robot cleans it, so a kid spills something, points, and the robot goes to clean it up.
Write that claim in one sentence before you write a script. If you can't, the problem is the positioning, and a better video won't fix it.
Decide one more thing up front: what a viewer should do when the video ends. Download the app, join a waitlist, book a demo, reply for access. Qaf ends on App Store and Play Store QR codes. Superblocks ends on "Sign up or book a demo." Pick one action and build the ending around it.
Pick the format that proves the claim fastest
I expected polish to predict success. It didn't. Every format below produced at least one strong launch, and what decided it was whether the format suited the claim.
- Screen recording or UI flow. Best when the product lives on a screen and the result is quick to see. Stripe recreated its whole purchase flow in 28 seconds. Lemon used live macOS demos of voice commands turning into finished tasks.
- Founder to camera with the product overlaid. Best when the audience needs to trust the person. Qaf's founder opens with a greeting to his community, then walks through a phone demo that shows citations for every answer. Liquid keeps its founder on screen for a 60-second pitch with the chat UI beside him.
- Live action in the real world. Best for physical or real-time products. Matic films a family at home. MouthPad needed only 10 seconds to show someone using a touchpad with their tongue.
- Skit or comedy. Best when the pain is relatable. Rippling shows SOC 2 compliance as an IKEA-style manual that reaches "Step 102." Wispr Flow cast small kids in suits running meetings.
- Short film. Best when you already have an audience, or when you can keep the product in the scene. Cofounder is a film of the founder phoning his real grandma, with the product open on a laptop in the field beside him.
- Motion graphics and big type. Best when the product is abstract. Bouncer shows "rage bait" and "AI slop" filtered out of a feed. Meta's TRIBE v2 shows predicted brain activity next to real scans.
What the video should have
The claim in the first three seconds. None of the strong launches opened on a logo animation. Airtable's Hyperagent video had "$10,000,000" filling the screen by the third second. Bouncer opened with "YOUR FEED IS BROKEN." Superblocks 2.0 opened on a crashing chart labelled "$300 billion wiped out from SaaS stocks." Slow openings, like Odessia's misty beach or the two seconds of black before Emergent's founder appears, ask the viewer for patience they may not give you.
The product working, end to end, at least once. This is the part people quote back in the replies. When Sabi launched a brain-computer cap with a manifesto and a static render, the replies asked for proof. One read "Show a demo you cowards!"
Proof a skeptic can check. SubQ put a benchmark chart against named competitors in its video. Fish Audio showed its latency next to ElevenLabs and Cartesia. Edison Scientific put its partner's chief scientific officer on camera. Big claims will draw doubt in the replies either way, but proof gives you something to point to when you answer.
Captions burned into the video. Most people scroll with the sound off, and almost every launch I studied had captions or big text on screen. Rippling's deadpan skit didn't, and one reply admitted they weren't sure what was going on.
A short runtime. The four shortest product videos I studied were Stripe at 28 seconds, Meta at 29, Valar Atomics at 30 and Claude Academy at 34, and all four were among the best on likes per view. Longer can work: Fish Audio ran 2:24 and Koji 2:15. The long videos that worked kept the product on screen. The ones that stalled spent their minutes on a founder talking.
One URL at the end. Claude Academy ends on academy.claude.com. Superblocks ends on superblocks.com. Use a short address someone can remember and type later.
What to leave out
Production that upstages the product. The launches with the most "best launch video ever" replies weren't the ones with the best engagement. Wispr Flow's kids skit got that praise from many people, and it had one like for every 6,000 views. If your replies are about the video, the product has dropped out of the conversation.
Other companies' logos, unless you're ready for them to notice. Vocci opened with the Claude, ChatGPT and Codex logos. It got attention, including a reply from someone on OpenAI's Codex team: "Thanks for the free advertisement!"
A price or claim with nothing to anchor it. Mark II led with "a $159 AI bookmark," and the most-liked replies were jokes about paying $159 for a bookmark.
Write the post that carries the video
The first line of the post usually did more to stop the scroll than the video did. The strong ones take a side:
- "AI is making kids dumber." (Koji)
- "Today, we kill the AI agent and introduce the AI employee." (Lindy)
- "We raised $136M to kill Slack." (PromptQL)
- "Gmail organizes your emails. @extradotemail organizes your life." (Extra)
Then keep the main post short and put the details in a thread of replies. Google's Stitch launch listed five upgrades and gave each one its own reply. Matic used its thread to explain its privacy approach and embed posts from real customers.
If you've raised money, the number can be the hook. Hobbes opened with "We raised $6M to kill the book-a-demo button" and had "$6M RAISED" counting up on screen in the first three seconds.
A word on giveaways. Lemon asked people to "RT + Comment 'Lemon'" for free access, and Fish Audio did the same with "Fish." Both got big reply counts, and most of those replies were just the keyword. That helps reach, but it tells you nothing about whether people want the product.
Where to post it
Choose whose account posts it. About half the launches I studied came from a founder's personal account rather than the company's: Wispr Flow, Lindy, Airtable, Cofounder. A founder account lets the launch sound like a person, and the founder can reply as themselves. A brand account makes sense when the brand already has the bigger audience, as with Claude or Stripe.
Make X the hub. The video, the hook and the thread all live in one post. Everything else points back to it.
Line up the first replies. The launches with lively threads had customers, investors and early users replying with something specific. On Matic, owners described how the robot changed their routine. On Valar Atomics, investors replied under the funding news. Ask your people ahead of time, and ask them to say what they actually use it for.
Cross-post to LinkedIn, rewritten for it. The same claim works, but LinkedIn readers want a few lines of context, so open with the problem in plain words before the video.
Plan follow-up posts. Bouncer launched one day and open-sourced its Chrome extension the next, which gave it a second launch post. Judgment Labs ran an ice cream truck stunt during launch week and posted about it in the thread. Sabi followed up with a WIRED interview. Each one gave people a new reason to look again.
Launch day: stay in the replies
The founders who answered nearly every reply kept their threads alive. Wispr Flow's founder answered with one-line jokes. SubQ's founder answered dozens of technical doubts, like "Any papers? Seems too good to be true," and then published a technical blog post. Skeptical replies are a chance to show your proof in public.
Measure it the morning after
Don't judge the launch by views. Some of the biggest view counts I saw had very little engagement behind them: Superblocks 3.0 shows 19.5 million views and 914 likes. Look at likes and bookmarks per view instead. Across the 43 launches where I had both numbers, the median was about 0.065% likes per view, and the strongest launches were at 0.15% or higher.
Bookmarks deserve their own look. Claude Academy and Airtable's Hyperagent both have more bookmarks than likes, which usually means people plan to come back and try it. After that, the number that matters is in your own analytics: how many viewers did the one thing you planned for at the start.
The checklist
- One claim, written in one sentence, before the script.
- One action you want viewers to take.
- A format that proves the claim fastest.
- The claim on screen in the first three seconds.
- The product working end to end, at least once.
- Proof a skeptic can check.
- Captions burned in, under a minute if you can, and one URL at the end.
- A first line that takes a side, with details in the thread.
- First replies lined up from real users, investors or customers.
- A LinkedIn version and at least one follow-up post planned.
- The morning after, check likes and bookmarks per view, then signups.