
{{first_name | Hey}}, welcome back.
Tavus, Freckle and Promethee all launched the same way inside five days, by putting the thing on camera and letting people judge before any claim arrived. Tavus took 20.7M views for a model it is not selling yet. Promethee, on 4% of that audience, was saved nearly four times as often per viewer.
Freckle sold a kids phone on what it took out, no browser, no app store, no social media, no YouTube, with the price out on day one at $199 plus $25 a month.
Promethee's post carries more bookmarks than likes, the only one of the three whose readers saved it more often than they liked it.
The launch was the call itself
Tavus did not open with a model card. Its thread runs short live calls, and the first plays like footage of two people talking until it becomes clear that one side is generated. Simon Says makes it listen, watch a pair of hands and move at once. A peacock held up mid-story gets worked into the story. It generates every pixel of every frame from one reference photo, chair and shadows included.
The claim, and the three numbers under it
Griffin is "the first model to pass the video Turing test," and the study holding it up is Tavus's own.
The comparison is where it slips. The post says previous systems "have had a pass rate <3%," the research page says "max 2%," and the only previous system Tavus measured came in at 2.4%, above the ceiling its own page asserts. The community note said the same. Publish the comparison you can show the work for.
The scoreboard Tavus borrowed
Tavus also got the claim onto NVIDIA's page, where Griffin-Lite sits under a real person and above every other system listed. Companies send in their own recordings and NVIDIA scores them, which beats grading your own paper. Tavus then calls it "NVIDIA's independent test of face-to-face AI," and NVIDIA's page, read again on 7 October, never describes the benchmark as independent. The arrangement was strong enough without it.
Two early posts carried it past the brand account. Kwindla, who builds real-time video infrastructure, posted an unedited call forty minutes in, and an early tester followed with "48% of people thought this thing was human. after this call, i get why." Seed the demo and let other people make the claim.
The phone that sold what it removed
Freckle leads with one line, "a kids phone built for real-world adventure," and spends the launch on subtraction. Fast Company describes founder Sam Terris's device as having no web browser, app store, social media or YouTube. Three replacements do the selling. A Discovery Camera a kid points at a bug, which names it and drops it into a collection. Quests a parent writes. A map with safe zones, so a kid finds the way and a parent sets the edges.
Kids ask questions all day and an adult is not always there to answer, so the fallback is a phone with everything on it. Freckle points that same question at the thing in front of them. The price went out on day one. Every outlet that covered the launch printed $199 and $25 a month, and the pre-order page splits it into $25 to reserve and $174 when it ships in December. Terris told TechRadar that any other word than phone would have buried the launch in basics: "If we don't call it a phone, all the questions that we're asking are those like, 'Can you make video calls?'"
The video sold the solution, the thread carried the problem
Freckle's film is built like a sixty-second commercial, kids outdoors with the thing in their hands, natural light, no voice-over telling you screen time is bad. The thread underneath does the arguing, that kids have lost most of their physical freedom and that existing kids tech is built around parent anxiety and works like an ankle monitor. One reader takes the feeling and leaves, another stops and finds the case, and the post holds both.
That choice puts all the weight on the film. Nothing ships until December, and Gadget Review found no independent hands-on testing. The warmest reaction came from the cap table, where Alexis Ohanian, whose Seven Seven Six backs it, replied "We're gonna take back childhood!!" Ask the investor closest to your category for the one detail only they can give.
Promethee made the founder the proof
Promethee launched on 4 October with a line a reader can repeat, an app that "turns your work into a multiplayer game," as the launch post puts it. Underneath it is a focus timer with streaks and a weekly leaderboard.
Nicolas Lacombe's thread is where the launch happens. He says he beat cancer at 24 while in law school, failed the bar twice, rebuilt his ability to work by turning it into a game, and finished first in his class. Read on 6 October, that post held 4,248 bookmarks against 3,756 likes, saved more often than it was applauded, and ran 5.12 bookmarks per thousand views against 1.33 for Tavus on an audience 4% the size.
A founder story is not a strategy a CMO can copy. What transfers is the job it does, standing in for proof the product cannot show yet. Promethee has no benchmark and no outside test, so where you have none either, a first-hand account specific enough to be checked is the nearest substitute, and it buys attention while belief waits on the proof.
The screens as the hero, and who else ran it
Nolla Health said on 5 October that it had an agreement with Utah's regulators to let its app issue first-time acne prescriptions, one the state calls "a grant of regulatory mitigation only, not an endorsement or approval," and co-founder Luis Wenus put the product on screen to explain it. The video cuts between him, Dr Zaid Fadul in a lab coat, and full-screen recordings of the app, a skin score, a face scan resolving into treatment options, and a daily check-in. Two minutes, and a viewer knows exactly what a patient does.
Two things travel with those screens. The claim is scoped tightly enough for a regulator to sign. Topical treatments only, Utah adults, mild to moderate acne, with oral medication and severe cases outside the pilot and anything scoring high on the severity scale sent to a physician first. And Zaid Fadul, a physician, is on camera carrying the risk, which keeps an autonomy story from reading as a replacement story.
That shape runs through this whole issue. HeyClicky, which we broke down in September, built its launch on a screen capture that needed no explaining, with the value readable in about three seconds. Tavus used live calls. Nolla used in-app screens with a doctor beside them. Letting the film carry the whole launch is older than any of it. Flowjam's roundup of fourteen launch campaigns has Dollar Shave Club doing it with one 90-second film that pulled around 12,000 orders inside two days, and that section links our own HeyClicky breakdown too.
Run the test on your own launch. If a viewer can tell what your product does with the sound off, the recording is the launch asset and the voice-over is decoration.
4 plays you can run on your next launch
Rent a scoreboard you do not own: Tavus got its claim onto NVIDIA's page, where it reads as a reference and not a brag. Ask your category lead this week which public board, index or test your product could be entered in before the next launch, then describe the arrangement in that board's own words and not a word further.
Ship the film and the argument as two things: Freckle's video shows a kid outdoors and argues nothing. The thread under it makes the case about lost freedom and surveillance. The watcher gets the feeling, the skeptic gets the case, and neither has to leave the post. Brief your team for both, and stop asking one asset to do both jobs.
Take the save rate to the board: Promethee took 5.12 bookmarks per thousand views against 1.33 for a launch 25 times its size. A save is the closest counter to somebody meaning to come back. Put that rate on the board slide with your own last three launches beside it, since a view count mostly reports how hard the platform pushed the post.
Shrink the claim until somebody else signs it: Nolla's announcement covers first-time prescriptions for mild to moderate acne, topical treatments only, Utah adults, with oral medication and severe cases left out. A state regulator signed an agreement covering that version, while saying in the same document that it is not an endorsement. Play 1 borrows another party's scoreboard, and this one borrows their signature, which is the only kind of proof a buyer in a regulated market reads first. Write the smallest version of your claim that a regulator, an auditor or a named customer would sign, and launch that one.
The risk to plan for
The capability is the abuse case, and the precedent is already on the record. A finance employee at engineering firm Arup paid out $25.6m in 2024 after a video call with deepfaked colleagues. Tavus answered the question inside its own announcement and still took the hit in the write-ups, which is the realistic ceiling on what a safety paragraph buys. Ship the disclosure mechanism with the demo.
Tool used
VideoFDB is the board Tavus put Griffin-Lite on, and a row costs one email with your recorded output.
Artificial Analysis runs its own free benchmarks on quality, price and speed across video, image and speech, so your numbers are on a public board whether you enter or not.
Utah's Office of Artificial Intelligence Policy signed Nolla's pilot agreement with the state licensing division, and it has authority to write agreements that "facilitate the deployment of AI in novel ways."
The EU AI Act sandbox tracker shows which member states have a sandbox open, with every one required to run at least one by 2 August 2026.
The FCA's regulatory sandbox is the UK door for anything touching financial services, open all year, and it lets firms "test new products live in the market with real consumers." The counters under every public post gave us every figure in this issue, and bookmarks per thousand views is play 3 on your own last three launches.
Which play are you stealing?
Steal this prompt
Point it at your own launch copy a week before you send. It separates what you measured from what you asserted.
You are a launch auditor. I am the CMO and this is my own launch, a week out.
MATERIAL: [the launch post or thread, the film, every page a reader can reach from them, and any research we are citing, each with the date I captured it] OUR CONTEXT: [what we are shipping, the date, the one claim we most want believed, the price if there is one, and who signed it off]
Get more such prompts in the Prompting Playbook (free for you)
Resources of the week
Positioning by Al Ries and Jack Trout, a three-part Advertising Age series in 1972 before McGraw-Hill published it as a book in 1981, argues that "a company must create a 'position' in the prospect's mind," and Freckle's naming call is that argument in one word.
Andy Raskin's The Greatest Sales Deck I've Ever Seen, free, lays out five steps, and Freckle's thread follows the first four before stopping at the fifth, evidence.
Marketing Examples by Harry Dry is a free weekly run of short teardowns of copy and landing pages that worked, for whoever writes your launch page.
How Brands Grow by Byron Sharp of the Ehrenberg-Bass Institute is the case for settling marketing arguments with evidence, which is the habit behind play 3.
Uncensored CMO, hosted by former CMO and System1 chief customer officer Jon Evans, puts senior marketers on the record about what actually worked, with guests including WPP's former UK president Karen Blackett and Chime's CMO Vineet Mehra.
Other Interesting Launches
PS. Did you see our AI Tools Directory? It’s a curated collection of 250+ useful AI tools for your profession, business domain, or specific use cases.
Stay curious, {{first_name | readers}}

