The Sale Happens in a DM You Will Never See

Think about the last thing you bought because of somebody else.

Not an ad. A person. Someone sent you a video, or dropped a link in a group chat, or texted you a screenshot with three words attached. Something like "we're doing this." And you did it.

Now go find that moment in a marketing report. You cannot. It happened in a thread nobody logged, between two people who were not asked for permission, at 10:40 on a Tuesday night. The brand that made the video got a number on a dashboard four days later and had no idea it was looking at the receipt for a purchase.

That is where the decision gets made. Not in the feed. In the forward.

The room you are not in

Private sharing is not a niche behavior anymore. It is the default. Around seventy percent of links now move through messaging apps, DMs and email rather than public posts, which means most of your word of mouth arrives at your site with no referrer attached and gets filed under "direct traffic."

Marketers named this dark social over a decade ago and then mostly went back to counting impressions, because impressions are easier to put in a slide.

Meanwhile the thing people trust has not moved an inch. A 2026 survey of a thousand US adults found that friends and family sit at the top of the trust list, ahead of reviews, ahead of creators, ahead of everything a brand can buy. The same survey found 76 percent of people think marketing feels performative rather than genuine, and only 8 percent believe a paid endorsement reflects what the person endorsing it actually thinks.

Read those two findings next to each other and the strategy writes itself. You are not trying to be believed. You are trying to be forwarded by somebody who already is.

Instagram has been saying this out loud

This is not a theory I cooked up in my kitchen. Instagram publishes it.

Across Feed, Reels and Explore, the signals Adam Mosseri has repeatedly confirmed as the heaviest are watch time, likes per reach and sends per reach, and of those, sends are the strongest signal for pushing content to people who do not already follow the account. Not likes. Not comments. Sends. The private ones.

How much stronger? Mosseri has put sends at roughly three to five times the weight of a like when the platform is deciding whether to show your video to a stranger.

He has also said the quiet part directly, which I appreciate, because most platforms make you guess. His advice to creators was to think about creating something people want to send to a friend.

The logic is not complicated. A like costs nothing and says nothing. A send costs social capital. When you forward something, you are attaching your name to it. You are telling one specific person that you thought about them for four seconds.

A share is a recommendation in person. It is the digital version of leaning across the table at a barbecue and saying you have to try this. Nobody sends a video to be polite. The recipient did not opt in, was not scrolling, and is not braced for an ad, and it arrived from someone they trust more than any brand or creator alive. That is the same transaction that has moved product since the invention of the neighbor, just faster and with a paper airplane icon on it. Instagram reads it as the highest quality vote available, because it is.

Which means your kindergarten teacher was running a distribution seminar and nobody in the room realized it. Sharing is caring was never a manners lesson. It was the ranking algorithm, taught on a carpet square to a group of people who could not read yet, roughly twenty years before Meta got around to putting a price on it. She also told you to use your words and to stop talking during the video. Three for three.

Sends per reach is just that vote expressed as a ratio. Sends divided by reach. It travels across account sizes, which is exactly why it is the number worth asking a creator for, and exactly why almost nobody volunteers it.

Meta put a cash register right outside the door

If you want confirmation that the messaging layer is where the value sits, stop reading the algorithm updates and look at where Meta is selling.

Ads Manager treats messaging as inventory now. Messenger inbox, Messenger Stories and sponsored messages are selectable placements. WhatsApp Status carries full screen vertical ads in the Updates tab. Click to Messenger and click to WhatsApp formats exist for one purpose, which is to drop a person straight out of a feed and into a conversation.

The company with the best attention data on the planet built a storefront in the chat apps. That is not a trend piece. That is Meta putting a price on the room.

Now look at the part they did not sell. The private thread itself stays ad free and end to end encrypted, and message content is not used for targeting. Meta will gladly rent you the seat beside the conversation. Nobody rents you a seat inside it. Sharing is caring, and caring turns out to be the only thing in the building that is not for sale.

So the most valuable surface in the building is the one square foot with no auction attached. There is no bid, no budget, no Advantage+ setting that puts your brand into that thread. It gets in exactly one way. Somebody already inside decides to carry you in.

Which is what you are buying when you pay for content, whether anyone framed it that way in the pitch. Not reach. A reason for one person to open a door that has no handle on your side.

Nobody forwards a commercial

So every brand hits the same wall.

Nothing about that mechanism cares how good your product is. The send has to happen before the product is ever considered, and people do not send ads. They send jokes. They send the thing that made them think of their brother. They send the video where the guy clearly did not plan for the dog to walk through the shot.

Entertainment is not the garnish on the strategy. It is the toll you pay to enter the only distribution channel that converts.

The Binet and Field analysis of 996 campaigns in the IPA Databank landed in the same place from a completely different direction. Emotional campaigns produce more brand effects and more business effects than rational ones. And the single most profitable category they identified was fame: campaigns that get talked about report the greatest profit growth, plus improvements across sales, share, loyalty and price sensitivity.

Getting talked about in 1994 meant somebody mentioned your ad at work. Getting talked about in 2026 means somebody hits the paper airplane icon. Same behavior. Better tracking, if you know which number to look at. I dug into why emotional creative outperforms safe creative in more detail in what makes a creative ad work.

Sell the problem you solve, not the spec sheet

This is where most briefs go sideways, so let me be specific about what I do instead.

Say a tire brand hires me. The brief comes over with the good stuff on it: highest rated all terrain tire in its class, three ply sidewall, best in category wet braking. Every word true. Every word unsendable. Nobody in the history of group chats has forwarded a sidewall.

So I do not tell you the tire is good. I show you the problem it solves and then I have the day the tire makes possible. The truck somewhere it had no business being. A hill that looked a lot smaller from the bottom. Mud up the door. Somebody in the passenger seat making a noise I will be hearing about for a year. The tires are in every frame doing exactly what the spec sheet promised, and I never read the promise out loud, because the video already settled it.

The viewer is not thinking three ply sidewall. They are thinking their buddy Dave would lose his mind. Dave gets the video. Dave now knows the tire exists, what it survives, and that a real person put it through that on purpose.

Say what the product solves and show what it feels like when it does. Specs are for the landing page, where somebody who is already sold goes looking for permission. Emotion is what gets them to the landing page in the first place.

It has to sound like the creator, not like your brand deck

This is where a lot of good money dies.

A brand hires a creator with a real audience, then hands over a script written by committee, with the tagline in the first line and three mandatory product mentions. What comes back is technically a creator video and functionally an ad. Nobody sends it. The whole reason you hired that person was the voice, and the brief removed the voice.

Two things make this worse than it used to be.

The first is that audiences have gotten very good at spotting the seam. That same trust survey found a third of consumers actively avoid brands over overly polished branding. Polish used to read as budget. Now it reads as a warning label.

The second is that the platform is on the audience's side here. Instagram's ranking systems now favor original, platform native content and demote reposts and reheated material. Mosseri's own framing for what cuts through is that the content will be more human, more raw, and more real. The same breakdown covers the audition system, where a new post gets shown to a small group of non followers first, and the first two to three seconds decide whether it travels or gets quietly throttled before your own followers see it.

So a video has three seconds to feel like a person. If it opens like an ad, it never gets to the part where it is an ad. That is not a taste argument. That is the distribution mechanism.

Which is also the honest case against generating this stuff synthetically, which I laid out in why hire a real human when AI videos are so much cheaper. Cheap is not the problem. Unsendable is the problem.

Relatable is the mechanism. Somebody has to watch it and think of a specific person in their life. That only happens when the video is about a situation the viewer recognizes, told by a voice they believe, in the rhythm that voice uses.

Entertaining is only half the job

Now the part most creators skip.

Getting sent is not the finish line. It is the handoff. The friend opens that video cold, with no context, no algorithm warming them up, and one question in their head: what is this.

If the answer is "a funny video," the send dies there. Pleasant, worthless. If the answer is "a funny video that told me what the thing is, what it costs, and why I would want it," you have converted a laugh into a consideration.

Which means it has to be entertaining and informative in the same thirty seconds. Not entertaining then informative. Braided. The joke carries the information, and the information gives the joke somewhere to land.

That is the actual craft, and it is what I sell.

I came up in software. Twenty five years, co founded a company and sold it, spent a long stretch running usability testing, which is a formal name for sitting behind glass watching people fail to use a thing while insisting they understand it perfectly. You learn one lesson permanently: what people say they will do and what they do are different data sets, and only one of them pays. A tester would tell me the interface was intuitive and then take ninety seconds to find a button. Same job now, better lighting. I do not ask whether people liked a video. I look at whether they sent it.

Short form and long form are not competitors

They are two jobs.

Short form earns the send and creates the demand. It works cold, on strangers, in three seconds. The system is built for that. There are reportedly 694,000 Reels shared through DMs every minute, which is the market you are competing in.

Long form catches the person after the send, when they have gone looking. Somebody forwarded them a thirty second video about a sauce. Two hours later they are typing "is that sauce actually any good" into YouTube or Google or a chatbot. That is a different moment with a different need, and a thirty second video cannot serve it.

That last surface is the one most brands are sleeping through. When somebody asks an AI tool what to buy, the machine is not watching anything. It is reading. Transcripts, titles, descriptions, chapter markers. A thirty second clip does not contain enough spoken language for a model to pull a confident answer out of, which is why an analysis of more than 100 million AI citations found that 94 percent of YouTube citations went to long form video, with Shorts taking under 6 percent.

The same study turned up something that should change how you value these videos entirely. Roughly 41 percent of the cited videos had fewer than a thousand views at the time they were cited, and the correlation between view count and citation frequency was effectively zero. The models are not reading your popularity. They are reading your transcript. Say the product name out loud, answer the question a buyer would type, break the video into chapters, and a small video can end up quoted inside an answer delivered to somebody who has never heard of you.

That is the loop closing. Short form gets you sent. The send creates the question. Long form is the thing sitting there when the question gets asked, and it no longer matters much whether it gets asked to a friend, a search bar or a model.

Worth knowing that the platforms draw this line too. Instagram will host a very long Reel, but the recommendation system stops pushing anything past roughly three minutes out to non followers. Long content mostly lands with people who already know you. Which is exactly the job description. Long form is not how the feed finds you. It is what is waiting when somebody comes looking, human or otherwise.

I have written the search side of this at length already, so I will not repeat it here: Instagram is a search engine now covers what gets indexed, and how to be the AI answer with video covers the surfaces where those answers get assembled.

The point for this piece is narrower. One format earns the recommendation. The other survives the research. Attribution data suggests 55 percent of paid social conversions need three or more touches to close, so running one and not the other means paying for the introduction and then leaving the room.

The honest part about the long video nobody watched

Now let me argue against my own team for a second.

A long video with low view counts is not automatically a failure. It is also not automatically a success, and the creator industry loves to pretend otherwise. "It's a slow burn, it's for SEO and AEO, it's building authority" is a real defence and it is also the most convenient excuse ever invented, because it can be deployed for anything and disproven by nothing.

So make it falsifiable. A long video gets judged on a different scoreboard, and the scoreboard has numbers on it:

• Search impressions and what queries it is surfacing for
• Average watch time and where the drop off happens
• Whether it turns up when you ask the AI tools what to buy in your category
• Whether it is the page somebody was on before they converted

If those move, the video worked, regardless of the view count. If none of them move over a reasonable window, the video did not quietly work. It did not work. Kill it, learn from it, make a different one.

I would rather tell a client that in month two than build a story around it for a year. The measurement framing I use is on my measurement page, and the difference between real evidence and impressive noise is the whole subject of how to vet a UGC creator portfolio before you pay.

The auto DM is the third format, and almost nobody briefs it

Short form earns the send. Long form catches the research. There is a third piece, and it sits in the exact room this whole article is about.

Meta will not let you cold DM anyone. That was the point of the cash register section. But it does leave one door unlocked, and the person has to open it themselves. When somebody comments on a post, replies to a Story, or sends a DM, that action opens a 24 hour messaging window, and inside it you can send links, offers, questions and follow ups automatically. The clock resets every time they reply. Once it closes, a real person on your team can keep the thread alive for up to seven days, but the automation is finished.

Read that twice, because it is the argument of this entire post written into Meta's own policy. The handle only exists on their side of the door. Content is what makes them reach for it.

Tools like ManyChat, an official Meta partner, built the comment to DM play on top of that window. Somebody watches the tire video, comments a keyword, and forty seconds later they have the link, the sizes, the price and one question that quietly qualifies them. Not a landing page they have to go hunt down. A message sitting in the same inbox where their buddy sent them the video in the first place.

The rules are tight and 2026 made them tighter. One automated DM per user per 24 hours from a comment or Story trigger, which kills the old habit of hitting the same person from three different posts. Automated experiences have to disclose that they are automated. Anything that asks for your password or runs as a browser extension will end your account rather than your sales problem.

Now the part that matters for what you are buying.

This is the line between UGC and a content system. UGC is an asset. You commission it, it gets delivered, you put it somewhere, and whatever intent it creates has to survive the trip to your website unaccompanied. A creator running a DM flow behind the same video is catching that intent while it is still warm, inside the app, in the window Meta hands you, in a thread the person opened on purpose.

Same video. One version ends at a view. The other ends in a conversation.

And it fails for exactly the same reason everything else in this post fails. The flow does nothing if nobody comments. Automation does not manufacture intent, it catches it. If the video was boring, the trigger never fires, and you have bought a very sophisticated funnel with nothing walking into the top of it.

One round is not a test

Sendability is testable, cheaply, before anybody spends a dollar on media.

Instagram now offers Trial Reels, which show a video only to non followers first. That is a free cold audience read on your hook. You can run two versions of the same idea, same information, different opening three seconds, and find out which one strangers stay for before it ever touches your real feed.

The way I work it: post organically, small and cheap. Read sends per reach and completion, not likes. Adjust the hook or the framing. Run it again and try to beat my own winner. Only then does paid budget get involved, and only behind the version that already proved it travels.

One round is not testing. It is a single data point wearing a costume. A creator who posts once, gets a decent number and stops has left the better version of that idea sitting on the table, and you paid for the worse one.

This is also why the roster matters more than any single video. Creator made assets used as paid ads tend to outperform brand made ads by two to three times, and that gap widens with familiarity, because the audience stops treating the creator as a billboard and starts treating them as a person with opinions. I made the full trust and repetition argument in four million views a month, and why that is the least important thing I sell, so I will leave it there.

What a real system looks like

Both formats, one plan, running against a stated goal.

Short form does the recruiting. Three to five pieces a month minimum, built hook first, tested cold before amplification, written so a specific type of person thinks of a specific friend. Long form does the closing. One or two pieces that answer the exact questions a shopper types after they get sent something, structured so a machine can quote them and a human can skim them. The DM flow does the catching. A trigger on anything designed to pull a comment, a reply written in the same voice as the video, and a human watching the thread once the automated window shuts.

The connective tissue is a real brief. Who is this for, what does it need to make them feel, what does the friend need to understand in the first ten seconds after they open it, and what number tells us it worked. Then an honest read at thirty days, including the parts that did not land.

I run about four million views a month across five platforms, and that number is close to the least useful thing on my dashboard. It tells me almost nothing about whether a client sold anything. Sends per reach tells me whether the video entered the room where the decision gets made. Those are not the same metric and only one of them is worth paying for.

I have also turned down products I would not use. Not as a moral flex. If I do not believe it, the audience hears it in my voice, the video does not get sent, and everybody loses money. The comparison of how to buy this work at all, agency versus marketplace versus hiring direct, is broken down here, and the underlying data on whether any of it works is in does UGC actually work.

Questions worth asking any creator

Take these to whoever you are about to hire, including me.

1. What is your sends per reach on branded work, and how does it compare to your organic posts?
2. Show me a video where the information and the joke are doing the same work. Which line is the send trigger?
3. If I hand you a list of product specs, what do you do with it? Ask them to walk you through the video they would build instead.
4. How do you test a hook before we amplify anything, and what did your last test change?
5. What happens in the first hour after somebody comments? Walk me through the DM flow and who writes it.
6. If we run long form, what specifically are you measuring, and what result would make you tell me it failed?
7. Whose voice ends up in the final cut, mine or yours, and what happens when my legal team wants a line that breaks it?
8. What have you turned down, and why?

Anyone who answers question six with a view count has told you what you needed to know.

You can see the work at notapourdecision.com/work, and if you want the honest read on whether your category is even sendable, start a conversation. I will tell you before you pay me.

The brands winning right now are not the loudest ones. They are the ones sitting in a group chat they were never invited to.