The product had already launched
Runway Agent went out in May 2026. A 2.0 followed in June. By September it was months old, reviewed, benchmarked and filed away by most of the people who were ever going to care.
Then a 30-second video from an account inside the company did 2,150,049 views. It named no product, made no announcement and came from an account with 60,065 followers, a fifth of the company's own. It said one thing: this ad cost $9 and took 10 minutes to make.
Nothing shipped that day. The whole difference was how it was packaged, and the interesting part is what it beat.
What a launch push actually looks like
Between 1 and 2 July, ten accounts posted about Runway Agent inside a 31-hour window. They shared an opening, close to word for word: I tested Runway Agent with a simple brief. Each one then ran a different demo.
A launch campaign for noise-cancelling headphones. For a premium cold brew brand. For a fitness app.
For a night skincare line. For a smart water bottle.
For a smart notebook. A luxury perfume ad. A cinematic car ad.
That is a well-run campaign by every measure an agency sells on. A roster spanning 24,366 to 271,710 followers, 836,499 between them.
Staggered start times across a day and a half so the timeline stayed warm. One consistent claim, demonstrated ten different ways so nobody saw the same video twice.
What it returned
| Measure | The July push, 10 accounts | One post in September |
|---|---|---|
| Combined following | 836,499 | 60,065 |
| Views | 420,864 | 2,150,049 |
| Median per post | 31,383 | — |
| Best single post | 128,439 | 2,150,049 |
| Reach against following | 0.5x | 36x |
Ten accounts with 836,499 followers between them reached 420,864 people, which is half their combined audience. One account with 60,065 followers reached 36 times its own, and 5.1 times everything the ten produced together.
Read that again as a ratio rather than a total. The push spent fourteen times the audience to get a fifth of the reach.
Why the packaging beat the amplification
The amplification was not the problem. Everything about the July window was competently run, and the same machinery with a different package would have worked.
The problem was the opening. I tested Runway Agent with a simple brief is an ad shape, and a reader recognises it before reaching the demo underneath.
Ten accounts using it inside 31 hours made it more recognisable, not less: the second one a reader saw confirmed what the first one was. A recognisable shape gets skipped on sight.
Compare what the September post led with. A price and a duration, both small enough to be surprising, and checkable against the video in the time it takes to watch it.
No product name to parse, no claim to evaluate, no brief to imagine. The product is the reason the number is possible rather than the subject of the sentence.
The other difference is who sent it. Ten creators saying they tested something reads as a campaign, because it is one. One person saying what a thing cost them reads as a report.
The number that argues the other way
Reach is one measure and the directory holds two more. On one of them the ordering reverses.
| Measure | The July push | September | Median across every launch |
|---|---|---|---|
| Reposts per million views | 245 | 25 | 195 |
| Saves per 10,000 views | 2.9 | 7.5 | 10.1 |
The push earned reposts at roughly the normal rate for a launch. The September post reached seven times as many people and earned an eighth of the rate, well under the median. Saves went the other way, with the single post more than twice the push.
What that says: a short video with a surprising claim is carried by the ranker, because watch time is the thing it can measure. A demo from a creator is carried by people choosing to pass it on, because passing it on says something about the person doing it.
Both are real. Reach is worth exactly what it causes, and these two shapes cause different things.
One caveat on the September figures. That post is days old against months for the July window, and video views land far ahead of reposts, so its endorsement rate should be read again later rather than treated as settled.
What to take from it
- A product gets more than one launch. Runway Agent has had at least three moments since May and the biggest came four months after release, on a day nothing shipped.
- Amplification multiplies the package, it does not replace it. Ten accounts running a weak opening produced a fifth of what one account with a strong one did. Fix the opening before you buy the roster.
- Vary the hook, not only the demo. The July push varied the subject ten ways and kept the opening identical, which is the wrong half to hold constant. The demo is what a reader gets to after the first line decides whether they stay.
- Put a checkable number in the claim. $9 and ten minutes can be verified against a 30-second video. "I tested it with a simple brief" asks the reader to take an interest first.
- Decide which measure you are playing for. Reach and endorsement came apart by a factor of ten across these two efforts. Choosing after the numbers land means grading yourself on whichever one looks better.
Questions we get asked
Can you relaunch a product that already shipped?
Yes, and the reach available later is not capped by how the release went. Runway Agent shipped in May 2026 and its largest moment on X came in September, on a day nothing was released, from a 30-second video that named no product.
The release is a date you do not control. The demonstration is work you can redo whenever it gets better.
Does a creator push beat a single post?
Only if the package is right. Ten accounts with 836,499 followers between them, posting inside a 31-hour window, returned 420,864 views, about half their combined audience.
A single post from an account with 60,065 followers returned 2,150,049. The amplification was competent and the opening was not.
Why did ten creators reach fewer people than one employee?
They shared an opening, close to word for word, so the tenth post a reader saw confirmed the shape of the first. A recognisable ad shape gets skipped before the demo underneath is reached. The single post led with a price and a duration, which a reader can check against the video rather than having to trust.
Does a viral launch post mean it worked?
It means it reached people. Across every launch we measure the median is 195 reposts per million views.
The launch push returned 245, roughly normal. The post that reached seven times as many people returned 25.
A short video with a surprising claim is carried by the ranker on watch time, while a demo is carried by people choosing to pass it on. Decide which the launch is for before running it.