Hey all, happy Friday,

Housekeeping at the top, because it matters more than my weekend: we're still hiring creative strategists. We're looking for people who can do the creative research analog style, on their own, and then amplify themselves with AI, not use AI to do all of the work for them.

If that's you, reply to this email, and we'll set up a call if it makes sense.

Okay. Newport Folk Festival, year three in a row, and it remains the best-mooded event in live music. Fruit Bats were incredible. The Lumineers came out on a small stage we happened to be standing near, which is the kind of thing that just happens at Newport. The lobster rolls hit. Perfect weather. And the thing I forget every year until I'm back: everyone there is simply happy to be there and be alive, and it turns out that's contagious.

We went with two of Maddie's friends and left James home, which settles the festival-debut debate from the last issue, and settles it correctly. Papa and Grandma watched him and fed him chocolate ice cream in quantities I would describe as "grandparental." The one bruise on the weekend: Dave Matthews came out Sunday night, and we had two-day passes for Friday and Saturday. Dave was my first concert ever, middle school, the gateway drug to the entire jam-band problem I have today. More on this in a second.

James got his first haircut Monday. He had a lot of very long, very curly hair, and for about an hour afterward it sat flat, and we all quietly panicked, and then it bounced back, and now he looks like an actual little boy. He took the haircut sitting in a tiny race car. I had to miss it, so Maddie ran haircut duty, and I received live updates at my desk like a man following a playoff game he has money on (I think most dudes do that these days. Idk. Insert other cultural ref here lol).

Also, Phish at MSG on Wednesday, which is why I'm writing this sentence on Thursday at partial capacity. Five nights at the Garden, their 92nd through their 96th shows there, playing 90s Phish. My oldest sister is the one who got me into them (she's nine years older, so 90s Phish is her native language; I'm a tourist), and it was an absolute blast. Late bedtime, early alarm. Thriving. Luckily, past me wrote most of this issue on Wednesday afternoon, and I'd like to publicly thank him.

Other James updates: fully walking. Word count way up; we told the doctor 30 to 40 words, which may be generous accounting. (Are "dada" and "daddy" different words? This is the alts debate all over again. Is a new hook a new ad, or the same ad with a haircut? Nobody knows. The doctor wrote down 40.) He says "bye-bye Lili (pronounced lie-lie)" to Linus every morning before daycare, which is so cute. And he's invented a game where, when he's with me, he points at the other room chanting "mama, mommy, mama," and when he's with Maddie, it's "daddy, dada, dada." He knows exactly which of us is which. He simply requests whoever isn't in the room. I don't know what mental game he's playing, but I don't like it.

Last thing: we leave for France next at the end of next week, and we're bringing James. First real vacation in a long time, which means no newsletter for two Fridays. I know. Devastating. If something truly unhinged ships while I'm gone, I reserve the right to file an emergency issue from a café. Otherwise, see you in three.

Keep Dave Matthews in mind, because that's the whole issue in one image: the moment I'd been waiting twenty years for showed up on Sunday, and my pass expired Saturday.

This is part five, and it looks the other way.

Every issue so far reads standing signals. Issue one graded every ad you've ever run. Issue two found the angles you haven't, by mining a year of simmering Reddit demand. Last issue watched your winners die in real time and told you which half to save. Good system, and all of it has one thing in common: your ad history and a year-old thread hold still while you study them. The system has been building a very good rearview mirror, and I've been calling it navigation.

Here's what none of it can see: the internet, this week. The sound that's peaking right now. The thread your exact buyer upvoted 2,500 times on Tuesday. The angle your biggest competitor just put fifteen fresh ads behind. The stuff that makes an ad feel like it was made this morning instead of last quarter.

The industry term for this is "reactive creative," and the industry system for it is an intern with a group chat and vibes. I wanted the same thing I built for fatigue: a monitor. Something that watches while I sleep and hands me a short list with receipts.

So I built one. It took a day, if you don't count the two times it broke in ways that are funnier than they are instructive, which means it took a day and a half and some dignity. The first list it handed me was good enough that I briefed off it the same afternoon, and I'm still deciding how much of that was the monitor and how much was luck.

One honest detail about hour one of that day, because it doubles as this issue's sponsor, and my house rule on sponsors is simple: I don't run ads for anything I don't personally use every day. The monitor started as twenty minutes of talking. I opened Wispr Flow, held down the hotkey, and rambled: what I wanted the thing to do, the trend methodologies I know are proven, every edge case I could think of, no structure, no typing, no idea where I was going until I heard myself get there. Flow turned the ramble into clean text, Claude and I went back and forth from it, and that transcript is a real part of why the build took a day and not a week. Same workflow when I'm editing ads, same workflow for the outline of the issue you're reading. Talking is fast enough that the edge cases actually make it in. Typing quietly talks you out of them. If you're not dictating your prompts yet, you're behind.

Power users figured this out early: speaking a prompt gives you 10x more context in half the time. You include the edge cases, the examples, the tone you want — because talking is fast enough that you don't skip them.

Wispr Flow captures everything you say and turns it into clean, structured text for any AI tool. Speak messy. Get polished input. Paste into ChatGPT, Claude, Cursor, or wherever you work.

89% of messages sent with zero edits. 4x faster than typing. Works system-wide on Mac, Windows, and iPhone.

Start flowing free

Anyway. Let's get into it.

(Quick catch-up if you're new: issue three packages the dashboard everything runs on, two paste-ready prompts, no keys. Issue four is the fatigue monitor. Today's prompt bolts onto that same dashboard, so start there if you're missing it.)

First discovery, ten minutes in: TikTok quietly killed its public trending page this summer. The page is just gone. The go-to TikTok trends actor on Apify, the one thousands of tools and agencies quietly sit on top of, now returns a maintenance note that says, in effect, "TikTok has disabled the whole page on Trends." That was break number one.

Let that breathe for a second. Every "what's trending on TikTok" roundup in your feed, every agency slide with this week's sounds on it, is now sourced from either TikTok's ad-platform data or someone's vibes. Mostly vibes. The screenshot economy runs about two weeks behind the actual culture, which in trend-time is a fossil record.

The good news: the signal just moved. Three places still tell the truth, and all three are public:

The Explore feed. TikTok's category pages (Food, Fitness, Comedy, Daily Life) still serve what's actually getting pushed. This is where the raw videos live: the sounds, the formats, the captions. (The maintenance note itself points you here, and the actor I landed on, clockworks' Explore scraper, charges about a dollar per thousand videos.)

Creative Center. TikTok's own tool for advertisers publishes rising hashtags with view counts and post counts, by country, by week. It's the one trend surface TikTok still maintains, because it's the one that sells ads.

Everyone else's launch behavior. More on this one in a minute, because it's the source nobody thinks of as a trend feed, and it might be the best one.

Second discovery, and this one cost me an embarrassing hour: the Explore feed is not a trend feed. It's a greatest-hits feed with a trend feed's name on the door.

Pull the top of any category and you'll get videos with twenty million views, and a third of them are three months old. Evergreen viral. Still excellent, still getting served, and completely useless to you. Build an ad on a wave that crested in April and you're covering a song that already left the charts.

The fix is one division. For every video: views divided by days since posted. Views per day. That's it, that's the whole velocity model. A video with 25 million views posted eight weeks ago is coasting at 450k a day on reputation. A video with 3 million views posted four days ago is doing 750k a day, and it earned every single one of them this week. The second one is the trend.

(I also make the monitor throw out anything older than about six weeks and anything under 100k views, because below that the "trend" is just a video someone liked. Guardrails, same philosophy as the fatigue gates: below that floor the monitor is just horoscoping.)

Sort by velocity instead of size and the list transforms. The twenty-million-view monuments drop out, and what's left is genuinely this week: the format that's suddenly everywhere, the audio that's spreading, the caption style that every fourth video is copying. That list is worth money. The other list is worth a LinkedIn carousel. (I say this as a man with drafts.)

And the best column on that list is the one TikTok won't publish.

Here's the part I'm most pleased with, mostly because it started as a failure.

Sounds are the connective tissue of TikTok. Ride the right audio in its window and TikTok rewards you for it; use it two weeks late and you're the brand showing up in last year's Halloween costume. So I wanted a trending-sounds feed, and here's the thing: you can't reliably get one. TikTok gates its sound rankings harder than anything else it publishes.

Break number two, and my favorite: the actor I pointed at Creative Center's sound rankings, automation-lab's trends scraper, ground for three minutes and then quit with the single most dignified failure message I've ever received from software. It refused to make up sound trends rather than return fake data. Respect. Most software would have just handed me something. (I have, at points in my career, been the software.)

Then the obvious thing occurred to me, several hours later than it should have. I was already pulling the top accelerating videos. Every one of them carries its audio metadata: track name, artist, a music ID. I don't need TikTok to tell me which sounds are trending. I can count.

If the same track shows up under multiple unrelated fast-rising videos, that IS the trending sound. Different creators, different topics, same audio, all accelerating at once: that's a wave. And when the reused audio is some creator's "original sound," it's even better, because that's a meme audio being born, and those windows are the shortest and the most valuable.

The satisfying part: the derived feed is built entirely from data I was already pulling. Zero extra cost. The gated front door had an open side entrance the whole time, and the side entrance was arithmetic.

Facebook was on my source list, and Facebook has no trends page at all. What Facebook has is better, and it's the source I'd keep if I could only keep one: the Ad Library. Public, searchable, no login, every active ad from every advertiser on Meta, with launch dates.

When you read it as a trend feed, it changes character. When a competitor launches one new ad, that's noise. When they launch fifteen in a week, that's not an ad buy, that's a memo. (Count distinct creatives, not rows. The Ad Library will happily count the same creative nine times across placements.) Someone in their building looked at data you can't see, drew a conclusion, got a budget approved, and shipped. You're reading the output of their strategy meeting three weeks after it happened, for free.

The monitor sweeps the category weekly: my brand's competitors by name, plus the category keyword. It groups fresh launches by advertiser and reads the copy. You learn two things: where the category just put its money, and, more useful, where it conspicuously didn't. A crowded bet next to an empty lane is the clearest brief there is: everyone just bet on flavor, nobody bet on the use case, the use case is open.

I'll show you a live one from the first sweep in a minute, because it found exactly this shape.

So the sweep runs: accelerating videos with their sounds, rising hashtags, competitor launches, plus the Reddit and search lanes the system already had from earlier issues. Seventy-seven signals on the first run, since we count things around here. Which creates the real problem.

Raw trends are useless. A trend report is a list of things that are popular, and popular does not mean yours. You can blow this in both directions. Chase everything: you become the brand doing a dead meme in a suit, and everyone can smell it. Chase nothing: your creative calcifies into evergreen, and the fatigue monitor from last issue gets to watch it die.

What's missing is the bridge. For every trend, one honest question: does this connect to a buyer I already understand? Not "could a copywriter connect them," a copywriter can connect anything, that's the disease. The test I gave the monitor is meaner: say the bridge out loud in one sentence. If the sentence needs the word "somehow," kill the card. The sober-milestone thread connects to an alcohol-free beer brand with no somehow in the sentence. The viral watermelon-cutting video does not, and no amount of workshopping will put a somehow-free sentence under it.

Everything that survives becomes a moment card. One card is one ad you could brief today: the moment with its receipts, the persona it maps to (from your dashboard, the one the whole system runs on), the format, the sound if there is one, the hook direction. And one more field, and the more I use it the more I think it's the point of the entire build: an expiry date.

Every card dies. Sounds get about ten days. Hashtag curves get maybe three weeks. A community's emotional peak gets a week or two. (Those numbers aren't from a study, they're starting bids from a few months of watching these curves by hand. Tighten them every time one burns you.) After that the card gets deleted, not argued about. A trend recommendation without an expiry date is a backlog item. Ship this week or don't ship: there is no "next sprint" for a wave.

The pilot brand is an alcohol-free craft beer client, name withheld as always, and honestly the perfect test: a brand whose entire category is a cultural conversation. First live run, five cards. Two are worth showing you.

Card one: the milestone wave. The monitor found three separate sobriety-milestone posts peaking in the brand's audience subreddits in the same seven-day window. "I am now 2 months sober," 2,562 upvotes and climbing. A one-year post right behind it on the same front page. A ten-month post behind that. Same emotional shape, same week, all at once. The read: it's late July, which means the Memorial Day quitters are hitting two months, last October's sober-curious class is hitting ten, and the people who quit last summer are rounding one year, publicly, at once. Summer is a stacked finish line.

The card mapped it to the brand's core buyer, specced a milestone-number static (the number huge, the can as the trophy, the community's own understated phrasing for the copy), and put roughly ten days on it, because the milestone window closes when the cohort's round numbers pass. That ad got briefed the same afternoon. The community had already written the copy in the thread titles. Issue two and three readers will recognize the method.

Card two: the empty lane. The Ad Library sweep found the category leader had launched fifteen distinct new creatives inside a week, and the monitor read the copy across all fifteen: flavor, summer, refreshment, flavor again. A fifteen-ad memo that says "we are betting on taste." The other lanes showed the brand's endurance-athlete buyers deep in heat-training threads, asking how to keep running through a heat wave, and the card wrote itself: the entire category just bet its July on flavor, and the athletic-recovery lane is standing empty during the exact weeks the athletes are loudest. Fifteen data points about their strategy, zero dollars of ours, one open lane with a date on it.

That's the shape I was hoping for when I built this: a specific open lane, with receipts from two independent sources and a deadline, instead of a meme and a dare. I'd love to tell you I knew the first run would find one. I did not.

The other three cards were fine, nothing I'd show you, and one of them I killed myself despite the monitor's high confidence, which felt great, like overruling a very smart junior who will not remember this. The somehow-test survives human review. And everything above, the sweep, the bridge, the cards, the expiry dates, is one prompt. Here it is.

Paste this into Claude in the same folder as your dashboard (the index.html from issue three). Web search on. Meta Ads connector optional but worth it, the Ad Library lane is the best lane. Run it Monday, brief by Wednesday, because everything it hands you is dated. Treat it like a tide table.

Three design choices in there, briefly, because each one exists because something went wrong. Velocity beats size is non-negotiable or you'll ship monuments. The somehow-test is the entire quality bar; the model will bridge anything if you let it, because bridging is what models do. And zero-cards-is-valid is what keeps you from becoming the brand that does trends because the report said to. The report working is the report occasionally saying "not this week."

Honest limits, the usual drill.

The paste version above runs on web search, and web search is a keyhole. The production version of this runs on scrapers (Apify actors, a couple dollars a month) pulling the Explore feed and the Ad Library directly, and it sees maybe five times what the prompt version sees. The prompt version still catches the big waves. It'll miss small fast ones. If this issue does its numbers I'll write up the full scraper build, cron jobs and all.

Sounds are the weakest lane and the fastest one. The derivation trick works, but by the time a sound is countable across multiple big videos, you're mid-wave, not early. Mid-wave is still fine for a brand. Early is for creators, whose cost of shipping is an afternoon.

It cannot make your org faster, and speed is the actual constraint. A moment card with nine days on it is worthless to a brand that needs three approval layers and a legal pass. If that's you, the honest move is to fix that before building any of this, or accept that your lane is the three-week hashtag curves and let the ten-day sounds go.

And the standing one: this is a diagnosis machine, not a taste machine. It will hand you an open lane and a peaking thread. Whether the ad you make in that lane is any good is still, stubbornly, the human's problem. Issue one called the forward bet the human job. I keep trying to automate that clause away, and it keeps surviving the replay.

Reply and tell me the most embarrassing trend ad you've ever seen a brand ship. The dead meme in a suit. The sound used three weeks late. I'm collecting these next to the zombie ads and the weird subreddits, and at this point the dataset is less "market research" and more "evidence."

Have a great weekend,

Will

P.S. If you spot a family in Paris where the toddler is pointing directly at his mother and chanting "daddy, daddy, dad," come say hi. He knows which one of us is which. That's the part that scares me.