One page, four lines, no software. Every new strategist at our agency writes this brief by hand before they get credentials to anything. This week's kit holds the worksheet, the test I use to separate a real persona from a fake one, and the eight hook shapes:

Hey all, and happy Friday,

We spent the weekend on Rosh Hashanah, and I spent it getting beaten by a toddler in a run of perfectly fair contests.

On Saturday morning, James showed up in a very precious white outfit. Maddie adored it (and as did everyone else who saw James); I was lukewarm on it (but more on that later). We sat through 45 minutes of songs about dinosaurs having dinner at Shabbat (among other things), and James danced through every one. At the end of the dancing, someone passed around red cookies topped with red frosting. All the kids were eating them, so I handed one to James immediately, not really thinking much of it. One important thing I forgot: his motor skills are basically nonexistent, and here I am, an adult who handed a bright red cookie to a toddler in a fancy white outfit. Needless to say, I have since spent quite some time with vinegar, hot water, and Dawn.

And can I just say, Dawn is sorcery. That shit is clearly loaded with terrible chemicals, but man, does it work wonders.

After that, we headed to a pool party in Westchester, where I showed up without a bathing suit even though I knew there was a pool and I knew James is drawn to water like a moth to a flame. I cuffed my pants, parked myself on the edge, and let friends who had actually prepared handle my parenting duties. Needless to say, both Maddie and I were soaked head to toe by the end of the day. That evening we went to an engagement party of our close friends at Other Half in Williamsburg. I hear anecdotally that parents in Portland, Maine take their kids to breweries and just relax. James spent the evening sprinting into off-limits zones and trying to sip from the glass of every adult he saw.

Clearly we are doing something wrong.

Monday, I brought a fellow agency, the Irishman Tris, who owns Webtopia, to Barry's Bootcamp, and afterward we met up with Collin Slattery, who owns Taikun, for soup dumplings. Collin had kindly ordered us Deliriums, which were already sitting on the table. Delirium is a beer I enjoy. It had simply never registered with me that it runs at 8.5%. I am getting into my mid-30s, and after downing two beers expecting a light lager, I was fully cooked, which puts James and me at roughly the same tolerance.

Linus, meanwhile, has been an exceptionally good boy this week. He is the eye of the hurricane surrounding him.

Something shifted on Wednesday. I have spent the last month of your Fridays pitting AI models against one another (stopwatches, blind judges, the full middle school science fair). Then I got on a call with a handful of other creative strategists and heard a sentence leave my mouth that undermined the whole month. Someone asked where AI genuinely pulls its weight at our agency. I answered that our process runs about 10 percent AI at the front, 10 percent AI at the back, and 80 percent human in the middle. Then I said the thing this issue is really about.

No one at our agency gets to use the AI tools until they can do the work without them.

That reads like a rule a guy invents so he can feel like a leader. In practice it is the cheapest quality control I own. (The casting test comes in second. We will get to it.)

The reason this matters more now than it did two years ago is simple. The tools sped up everything surrounding the job, so people who cannot do the job are now shipping bad ads at record pace. So there is no bake-off this week. This issue is for the folks who are newer at this, and for anyone who has ever looked at an empty brief and opened ChatGPT out of pure fear.

Creative strategy has a jargon problem. This entire issue runs on five words, and you likely know four of them already.

A persona is the person the ad speaks to. An angle is the problem you are solving for that person, and what changes once you solve it. A format is the container the ad arrives in: a headline over a photo, a customer video, a comparison chart. A brief is the one-page set of instructions you hand to whoever builds the ad. A hypothesis is a sentence that says "I think this person has this problem, and here is why I believe that."

That is the whole vocabulary. Every ad that has ever performed was a specific person, a specific problem, and a container, and somebody believed in that combination before any proof existed.

The entire test is four lines on one sheet of paper. A new strategist fills them out for a real product with zero tools, and then we discuss it.

The order is the most important thing on the page. Nearly everyone new to this starts at line four. They spot a cool ad format on Twitter (a fake text thread, a singing cartoon, a Pixar dog) and reverse-engineer a reason to build it. That is how brands end up spending real money to justify something they thought was cute.

When lines one through three are right, line four is the easy part. When they are wrong, the best format on earth is a lovely frame around nothing.

The full test: four lines, handwritten, in this exact order. The example is the divorced dad you will meet later in this issue. The format sits last on the page, and that is deliberate.

There is a persona that shows up in briefs over and over: busy professional.

Try to cast that person. Seriously, try. Is it a nurse? A lawyer? A guy running a Jiffy Lube? "Busy professional" covers roughly every adult who has a job, and an ad aimed at every adult who has a job is aimed at nobody.

Now cast this one: a parent whose toddler has turned bedtime into a ninety-minute negotiation. You can picture that person. You know exactly what their kitchen looks like at 7:40 pm. You could write their group chat from memory.

That is the whole test for line one. I call it the casting test. If you handed the sentence to someone who books actors, could they find the person? If the answer is no, you have a category, and nobody can write to a category.

I did not always pass this test myself. Our own dashboard (the tool we built to read ad accounts) spent months cheerfully tagging personas as "busy professional" and angles as "convenience." The labels were technically correct and completely useless. Nobody can brief from "convenience." A couple of weeks ago I rebuilt it so every row now carries the specific person, with the broad bucket demoted to parentheses behind it.

The old labels are still visible in there. They live in the same column as the good ones, which feels a bit like discovering your high school yearbook photo is still in your wallet.

Two rows pulled from the same column of our dashboard. Top: the labels it used to hand us, a persona called Busy Professional and an angle called Convenience. Bottom: what a castable person looks like, plus the Reddit thread it came from. The client name and product are blurred on purpose.

On the call I used the water glass line. A lazy sleep ad says "Tired? Here's the solution." A good one asks "Do you have three half-drunk glasses of water on your nightstand right now?" No marketing meeting produced that second line. A stranger typed it on Reddit at 2 am, and we went and found it.

Once you can fill out the four lines, the next question is which ad to build next. That question used to feel infinite to me. The answer turns out to be exactly three moves, and what you hold still defines each one.

An iteration keeps the format and swaps one other thing. The same person with a new problem, or the same problem pointed at a new person. This is the safest move, because two of your three variables already carry proof.

A reskin holds the person and the problem exactly where they are and swaps only the container. The winning headline becomes a customer video. The customer video becomes a comparison chart. The message already carries its validation. You are just dressing it in a new outfit.

A net new swaps everything. New person, new problem, zero proof. Roughly one in five of the ads we make falls in this bucket, and that one in five is the only source of new customers.

The three columns on our dashboard's main page map to the three moves. Left: build on what works, with 402 proven combinations available to iterate. Middle: 8 winning messages that have never appeared in certain formats. Right: 31 bets on people this account has never talked to. The client name, spend and ad names are blurred.

Here is what one of those rows looks like once you open it. I am showing it because it is the same four-line test, only filled in by the account's own history instead of by a human with a pen.

A single opened row. The equation up top is lines one, two and four of the paper test. The specific person and problem live on the left; the broad bucket is demoted to the parentheses on the right. Underneath: the hook that carried the ad, the ads that proved it, and the untested neighbor worth trying next.

Look at the blue box, the one marked next move. It reads: same person, same problem, one format this message has never lived in. That is a reskin. Then it reads: carry the same problem to a different, neighboring person. That is an iteration. The machine is not doing anything brilliant. It is holding two things steady and moving one, which is the entire job.

The third move is the scary one, so here is one we placed this week.

The account sells a toothbrush. Two of its breakout ads this month talk to the same person, a parent dreading the bedtime brushing fight. So we asked who else lives that exact night.

He was on r/daddit. A divorced dad with 50/50 custody posted about handling bedtime alone on his nights. That post carries 22,333 upvotes. Twenty-two thousand people read "I've been putting him to bed on my own" and clicked the up arrow.

That is line three. That is your reason to believe. Nobody on my team would have offered "divorced dads" in a brainstorm, and if somebody had, I would have asked for the evidence, and there would have been none.

Here is that brief, written by hand. The person: a divorced dad handling bedtime solo on his custody nights. The problem: no second adult exists to tag in when brushing becomes a fight, and he does not want his nights to be the hard ones. The reason to believe: that thread, plus the two ads in this account already proving the bedtime fight is real for parents in general. The format: a plain headline over a photo, because that is the cheapest possible way to learn whether he shows up in the numbers.

AI did not appear once in that paragraph, and it did not need to.

That last choice earns its own section, because most brands run it in reverse.

The standard way to test a new audience is to pay a creator to shoot them a video. That costs a few hundred dollars and two weeks, and if the ad flops you cannot tell whether the person was wrong or the video was.

A static image costs nearly nothing and takes one afternoon. So we test the person with a static first. When a plain headline pointed at a brand new person starts converting, you have learned something big: a market exists there. Only then do you pay for the video.

What a new person looks like when they turn out to be real. One ad, 29 days old, pointed at somebody this account had never addressed. The cost to acquire a customer sits 39 percent under the account average. The box at the bottom is the shopping list: the formats this message has never run in. The dollar figures and the ad itself are blurred.

We call that a breakout. One ad, a small budget, and a cost per customer comfortably under the account average. My favorite part is the box at the bottom of the card. It names every format that person has not seen yet: an explainer video, a customer video, an offer, a comparison. That list becomes next week's briefs. Each one is a reskin, and a reskin is the safe move.

So the scary move only needs to land once. Everything downstream of it becomes the safe move.

This is an odd place to arrive after a month of model bake-offs, and I know it. Either I contain multitudes or I have no consistent principles, depending on the week. Our agency runs on these tools, and I will probably race two more models by October. I am not a reformed man.

AI earns its keep at the two ends of the job. At the front, it reads the 393 ads in an account and tells me which person, problem and container each one used, a task no human wants. It reads ten thousand Reddit comments and pulls up the divorced dad. At the back, it renders the image and proofreads the spelling.

The middle is where it falls apart. Last week I had two frontier models each write 23 headlines from identical data. One of them wrote 18 of its 23 as questions. "Where did the prep go?" "What waits after the shift?" Blind judges gave that set 44 out of 100 on hooks. The model had every piece of the data and still could not tell which person inside it deserved the conversation.

The middle is choosing the person and believing in them, and that is still the job.

If you are new, I think that is good news. The flood of lazy AI ads is real, and it is the softest competition you will ever face. An ad that sounds like its writer has actually met the customer is now rare, and rare things get noticed in a feed.

The kit is deliberately small. It holds the one-page analog brief, in printable form and as a doc. It holds the casting test, with ten personas that fail it and the ten specific versions that pass. It holds the three moves on a single page, with what to hold still for each one. And it holds the eight hook shapes we sort every headline into. That last page will not tell you what to write. It shows you what you keep writing by accident, which for me was questions.

Print the brief. Fill it out for a product you love before you touch a single AI tool. If line one takes you longer than ten minutes, that will be the most useful thing you learn all week.

Every ad is a person, a problem and a container, in that order. You write those down by hand, with a reason to believe, before any tool gets opened. If a casting director could not locate your persona from one sentence, you have a category, not a person. There are exactly three moves: iterate, reskin, net new, and what you hold still defines each one. New people get tested with a cheap static before anyone pays for video. Faces do not get faked here. AI belongs at the two ends of the job, and the middle still belongs to you. Most of this you probably knew already. The piece of paper is the step everybody skips, and for a long time that everybody included me.

Have a great weekend,

Will