Hey all, happy Friday,

Last weekend we went out to my dad and stepmom's place on Long Island. Saturday was a little cloudy, but Sunday turned into one of those perfect clear beach days. We dipped James's feet (and by feet I mean his whole body) in the ocean a few times, and he went absolutely nuts. This continued until he started getting cold, and even then we had to drag him away as if he were answering a distant siren's call.

Then he rolled around in the sand, eating fistfuls of it in between. He also fell hard for the big orange bucket I used to play with as a kid, and the bucket did what Calypso does: it simply would not release him. We eventually pried him loose.

Tuesday night we spent the evening in the ER because James somehow dislocated his toe. Credit where it's due to the children's ER near us, which was kind of unbelievable. James was seen immediately. A nurse asked, "Does James like trucks?" and produced a fire truck with light-up flashers and sirens.

And then, somewhere between arriving and the X-rays, James did something to his own foot and popped the toe back into place.

Homer's heroes get recognized by the scars on their legs. Ours will be known by the second toe.

With toe-gate behind us, Maddie and I finally got to use the tickets we bought last month for The Odyssey in 70mm IMAX. I get genuinely stubborn about tickets for these things. Years of Phish on-sales taught me when to keep refreshing and when to cut my losses, and I paid for that education in an embarrassing number of hours spent staring at queues.

This time the stubbornness paid off. We ended up in the fourth row, which for a normal movie is punishment. On that screen it meant getting swallowed whole.

It was incredible, and worth the seven years I mentally spent to get us into Lincoln Square.

Somewhere in that fourth row, a work thought snuck in, the way work thoughts do. The Odyssey has been re-released in new formats for about three thousand years, and it still sells out. Hold that thought.

Before I show you the ad that grew three new bodies this week, I have to own something. If you replied to last week's issue, your reply bounced.

The newsletter moved to a new sending domain, and the reply-to address was still pointing at the old dead one, which is a spectacular own goal for a guy whose entire newsletter is about building reliable systems. It's fixed now. This address works, and I read everything.

If you replied last week, especially if you sent "September" about the consulting spots, please send it again, because there is no record of who tried. I'm sorry, and I'd love to make it up to you by actually receiving your email this time.

Here's a mistake I made for years: treating a winning ad as an asset. I did it most memorably with an ad that quietly printed money for three straight months, during which I read exactly one sentence of what it was telling me ("spend more on this") and never once sat down and actually read the thing. Past me thought scaling a winner meant pushing the budget slider and going back to brainstorming new concepts from zero.

But a winning ad is not really an asset. It's a discovery. Somewhere in your audience, a specific kind of person holds a specific belief, and your ad said that belief back to them well enough that they paid you. That's the finding, and it's the most expensive research you own, because you paid for it in real media spend against a live auction. The layout it happened to be wearing that day was just the first outfit. This week I got to run the better version of my three-month failure on a client's ad, and I want to walk you through the whole thing.

It's a toothbrush ad.

The surface read: ADHD adults, and the angle is mental load, not hygiene. The headline does the work ("Brushing with ADHD, minus the mental load"), the crossed-out words name the pain, and the proof stack is three checkmarks. That much anyone gets in five seconds.

The deeper read is in the margins. Look at the doodles. "You snoozed 6 times." "Streak broken." "I'll do it later." "Did I brush?" They read like background scribbles, almost set dressing. Except every one of them is a moment from the audience's actual life, specific enough that the right person feels seen by a doodle.

And that's the trick this whole issue turns on: every detail inside a winner is a hypothesis that already passed testing. The audience approved the headline, sure. But they also approved the doodles, the strikethrough gag, the glow, the navy. Each of those details rode along on real spend and survived. Which means each one is allowed to grow up and become the hero of its own ad.

I call this promoting the doodle.

So the winner gets multiplied. My team calls this reskinning, and the word is precise. Iteration holds the persona and changes the angle, or holds the angle and changes the persona. Reskinning holds everything, the persona, the angle, the proof, and changes only the format. Our working split across a month of creative is 60 percent iteration, 20 percent reskinning winners, 20 percent net-new personas, and this issue is that middle twenty.

Into what format, though? This is where last issue's swipe file stops being research and starts paying for itself.

The combo stays constant: same persona, same angle, same proof stack. What changes is the format, and two inputs pick the lane: the account's own history (what this brand has already tested) and the digest (what the market is winning with right now). Formats that rank well out there and sit at zero tests in the account are the free lanes. I found three. My gut would not have found them. It notices headlines. The spreadsheet notices absence.

The file's folders then supply the exemplars. For each lane I opened real survivors from the file and pulled the structure. One was a real estate app's ad, built around a giant floating notification pill. Another was a menswear brand's ad styled as a toy-kit box, inventory grid and all. The third was a wellness brand's ad that's just an oversized search bar with the customer's actual query typed into it. Three brands, none of them remotely dental, all holding a composition worth stealing. TAKE the structure, IGNORE the brand, the scene, the words. Same rule as last issue.

Then each format got matched to the detail it could promote, and this is the part I'd frame and hang on the wall. The "you snoozed 6 times" scribble became the notification ad. The little checklist doodle became the kit ad. The third doodle I'm not going to spoil, because it grew up into my favorite of the three, and its ad explains itself on sight.

Here's what shipped, all three in formats the account had never run.

And here's the one I wouldn't spoil.

Notice what did NOT carry over: the copy. Not one line is reused. The notification ad, the kit ad, and the search ad each got fresh copy written into their format's shape, because a search bar carries a lowercase typed query and a kit box carries inventory labels, and one properly formal marketing sentence would kill either. The funnel inherits the combo. It never inherits the words. (Issue six covered copy shape in depth; this is that rule earning its keep.)

Honesty about production, since I show the failures too: GPT Image 2 fought me on two of the three. It kept obeying gravity instead of the brief, sliding the CTA button down into the zone where feed UI covers it, and it took me several re-rolls and one genuinely unhinged "why the fuck is it doing this" monologue at my monitor before I remembered that coordinates are a thing you can just specify. Total wall-clock for all three, including the arguing: two hours and fifty minutes. The arguing used to be the whole week.

Everything above is one prompt now. It wants two things to exist first: a winner you can defend with evidence, and the swipe file from issue six (the starter version works fine). Paste it into Claude with your dashboard folder, attach the winning ad image, and go.

The credentials check in phase 1 is load-bearing. Skip it and you're multiplying a guess, not a winner. The detail inventory in phase 2 is where the margin scribbles get their due. And phase 5's never-reuse-a-line rule is what keeps three formats from being one ad in three costumes.

Honest limits, as always.

Multiplying tells you nothing about WHY the winner won. The genome read is an educated guess from the artifact. Your account data knows which ad won; only testing the offspring tells you which trait actually mattered. Sometimes all three work and you've struck something real. Sometimes none do, and the winner was carried by something the pixels never show, placement or audience or plain luck. That is the limit of reading pixels instead of running the test.

The thumbnail trap is still live. Three formats can converge back into sameness if the same product shot sits dead center in all three at the same scale. Check the offspring as a grid at thumbnail size before shipping, the same discipline as last issue.

Reskinning doesn't replace discovery. It mines a proven vein deeper, and it will never open a new addressable market, because the new persona and the new angle come from research, not from the file (that's issue two's territory). Net-new personas hit less often and matter far more when they do, which is why they keep their twenty percent of the mix even though reskinning feels better week to week. A team that only reskins eventually runs out of winners to reskin.

And fatigue is not defeated, only outrun. The offspring inherit the combo, which means they eventually inherit the combo's fatigue too. When the vein cools, no format will save it, and the answer is back to discovery. (Though the ceiling on a truly great combo is higher than you'd think. The Odyssey has been reskinning for three thousand years and I just paid IMAX prices for the fourth row.)

New game this week, and it's a trade. Reply with your winner, the actual image, and I'll reply with the three formats I'd multiply it into and which detail I'd promote. This is not wizardry, it's a checklist and a very large folder, which is exactly why I can run it for strangers on the internet. First come first served until my inbox teaches me a lesson about open-ended offers. (Replies also, genuinely, help the new sending domain learn that this newsletter is not suspicious.)

And since replies actually work again: September consulting is still open, and this is the last issue before September starts. Same shape as before, I work directly with you or your team to stand up this exact system in your account, with your brands, in time for Q4. The math hasn't changed: the brands that walk into November with a creative engine built it in September, and the ones that didn't get to learn it live, at November CPMs. Reply with "September" and I'll send details.

James's toe is fine, by the way. He remains the only member of this family who fixed his own problem in under thirty seconds this week. The rest of us needed re-rolls.

Have a great weekend,

Will

P.S. The immortal-ad collection from last issue is real and growing, and the reply bounce means some of your immortals died in transit. Send them again. The museum has room, and this time the door is actually open.