Every few months a founder tells me they are moving budget to YouTube because the CPM is half what they pay on Meta. Six weeks later the line is switched off and YouTube joins the list of channels that did not work for them. The CPM was the reason they started and the ROAS column was the reason they stopped, and neither number was ever going to answer the question they were actually asking.
YouTube is priced like a reach channel, sold like a performance channel, and only behaves like either one if the rest of your business is in shape. Those three facts do not agree with each other, and the gap between them is where most DTC budgets quietly disappear.
What YouTube is actually for in a DTC media mix
Google Search harvests demand that already exists. Somebody wants magnesium, they type magnesium, you pay to be in front of them. Meta does a bit of both, which is why it feels like the only channel that works until it gets expensive. YouTube sits closest to television with a checkout attached. It is where you convince somebody who was not looking for you, which is a slower and more valuable job than catching somebody who was.
That distinction has a practical consequence most brands skip past. If YouTube is doing its job, your blended efficiency and your branded search volume move before your YouTube ROAS does. A channel that creates demand gets robbed by the channels that capture it, and in a last-click report the thief looks like the hero. Start the test knowing that, or do not start it.
Cheap impressions are not a strategy. If you cannot say what demand you are trying to create and who you expect to create it in, the CPM is irrelevant.
The four formats worth your budget
YouTube sells more ad formats than any DTC brand needs. Four of them do real work, and they do different jobs. Running all four at once on a small budget is the fastest way to learn nothing.
Skippable in-stream
The closest thing to a TV slot that a DTC brand can buy by the view.
Cost per view sits around one to three cents, with a 2026 cross network average near $0.024, and you only pay when somebody stays. The sweet spot for length is thirty to ninety seconds, which is long enough to demonstrate a product properly and explain why it exists. The entire campaign lives or dies in the first five seconds, because that is when the skip button appears. This is the format that makes YouTube worth having, and it is also the one that punishes recycled Meta creative hardest.
Demand Gen
The sensible starting point for most DTC brands.
Demand Gen combines video and image creative with social style audience building across YouTube, Discover and Gmail. It accepts assets closest to what you already run on Meta, it reaches a readable result fastest, and it needs the least new production to get going. The trade off is that it blends surfaces, so a good result tells you that Google inventory works for you without telling you that YouTube specifically does. Treat it as the workhorse, not as the whole test.
YouTube Shorts
The cheapest inventory on the platform and the closest to TikTok.
Shorts CPMs run around $4.85, lower than any other YouTube surface. Vertical, fifteen to thirty seconds, a hook window closer to two seconds than five, and most of the audience watching with the sound off, so text on screen is not optional. If you already have a working TikTok library, this is where it goes with the least rework. If you do not, Shorts is a poor place to learn, because cheap inventory hides weak creative for longer than expensive inventory does.
Video Reach Campaigns
Maximum unique reach across every surface from one campaign.
Video Reach buys bumper, skippable in-stream, non-skippable, in-feed and Shorts together and optimises for reaching the most distinct people for the money. It is the right tool for a product launch, a seasonal push or a moment you need a lot of people to hear about in a short window. It is the wrong tool for a standing acquisition line, because reach efficiency and customer acquisition are not the same objective and it will happily deliver one while you are paying for the other.
Your Meta creative will not work here
On Meta you are interrupting a feed the viewer already controls. The job of the first second is to stop a thumb that is moving anyway, so the winning openings are pattern breaks, captions and fast cuts. On skippable in-stream you are standing between somebody and the video they deliberately chose to watch. They are not scrolling, they are waiting, and they are mildly annoyed with you. The job of the first five seconds is to make staying feel like a better use of time than skipping.
That is why the same fifteen second Meta cut that produces a 3x on paid social produces nothing on in-stream. It was engineered for a different emotional starting point. The fix is usually not a new film, it is a new opening and a longer edit. Formats that die on Meta because they are too slow, such as a founder explaining why the product exists, a real demonstration, or a properly earned before and after, are exactly the formats YouTube pays for.
The working rule is one asset per surface, not one asset resized three times. Shorts gets the vertical TikTok style cut. In-stream gets the thirty to ninety second version with a genuine opening. Demand Gen gets whichever of the two tests better. Three assets, one shoot, and you will learn more in eight weeks than most brands learn in a year of resizing.
The measurement problem, stated plainly
Two things are true at the same time, which is why operators argue about this channel forever. Platform reported YouTube performance overstates the channel, because view-through windows are generous and will claim conversions that were always going to happen. Last-click reporting understates the channel, because the customer YouTube convinced goes on to search your brand name or open an email, and the credit lands there instead.
Meta and Google both over-attribute by somewhere between twenty and sixty percent depending on the account, so a campaign showing 3x in the interface can be delivering closer to 1.5x in reality. Reading either report on its own gets you a confident answer to the wrong question. This is the same disease covered in more depth in our piece on attribution and post-purchase surveys, and YouTube is where it hurts most.
The honest measurement stack is short. Run a holdout or conversion lift study, or a geo split where one region gets the spend and a matched region does not. Add a post-purchase survey so customers can tell you where they heard about you. Then watch blended marketing efficiency, branded search volume and new customer count across the whole test window rather than the YouTube ROAS column. Independent incrementality testing became the most trusted measurement method among senior decision makers in 2026 for exactly this reason, roughly twenty points ahead of media mix modelling and close to double in-platform reporting.
When YouTube is the wrong channel for you
I turn brands away from this channel more often than I put them on it. If your total paid media budget is small, the eight to twelve weeks YouTube needs to read properly will cost you more in lost Meta learning than the channel can return. If nobody can produce video beyond product photography, you will spend the budget proving that a slideshow does not hold attention for thirty seconds. Neither of those is a YouTube problem.
The bigger disqualifier is retention. Demand creation on a leaky bucket is the most expensive mistake in DTC, because you pay full price to introduce people to a business that will not keep them. If your repeat purchase rate is weak and your flows are half built, fix that first. YouTube will still be there, it will still be cheaper than Meta on a CPM basis, and the customers it brings will be worth several times more once you can hold onto them.
What this looks like in practice
A test that can actually answer the question looks like this. Eight to twelve weeks, a budget floor you have already agreed you will not pull early, and Meta held flat for the duration so you are not reading two changes at once. Demand Gen as the workhorse, one skippable in-stream line carrying a real long form asset, and Shorts fed from the TikTok library if one exists. Existing customers excluded unless the campaign is deliberately a win-back, and a frequency cap set before launch rather than after somebody complains.
Write the kill criteria down on day one. Not a ROAS number, because you have already agreed that number is not trustworthy on this channel, but the movements you would need to see in blended efficiency, new customer count, branded search and survey responses to keep funding it. Deciding what would change your mind while you are calm is the difference between a test and an expensive opinion.
The brands that make YouTube work are rarely the ones with the best video. They are the ones who bought it for the right job, judged it on the right numbers, and left it alone long enough to find out. If the wider question is how much of your budget should sit in demand creation at all, that is covered in our guide to paid media budget allocation.
Inside the system
How we build this for brands
Channel decisions like this one are downstream of two things we build first. A voice of customer engine mines reviews, support messages and social comments into the language customers actually use, which is where the opening five seconds of an in-stream ad comes from rather than from a copywriter guessing. Alongside it sits a profit and cash-flow dashboard built from live Shopify and ad platform data, with a reporting agent that surfaces blended efficiency, new customer count and leakage every week, so a demand creation test can be judged on the numbers that move rather than the ones the platform prefers to show you.
Underneath both, the lifecycle flows have to hold the customers the ads introduce, which is why welcome, replenishment, post-purchase and win-back sequences get built and deployed in Klaviyo before we recommend spending at the top of the funnel. Part of this runs live for portfolio brands today; the full system is what we deploy when we take a brand on.
Paid Media Audit
Find out which channel is creating demand and which is just claiming it
We pull your platform reporting apart against blended performance, branded search and survey data, and show you where the credit is landing versus where the customers are actually coming from. You get the channel mix, the creative gaps and the order to fix them in.
Book Your Paid Media AuditFrequently asked questions
How much should a DTC brand budget to test YouTube ads?
Treat it as an eight to twelve week commitment with a real monthly floor rather than a two week experiment. YouTube is a demand creation channel, so the return shows up in blended performance and branded search before it shows up in platform ROAS. If the budget you can commit would not survive six weeks of flat last-click numbers without being pulled, the test will be killed before it can tell you anything. Brands spending under roughly five thousand a month on paid media in total are almost always better off fixing Meta creative and retention first.
Are YouTube ads cheaper than Meta ads for ecommerce?
On a CPM basis, usually yes. YouTube CPMs typically run $5 to $10 for ecommerce advertisers and Shorts inventory sits lower still at roughly $4.85, against a DTC median of $13.52 on Meta. But CPM is the cost of being seen, not the cost of a customer. Cheap impressions against the wrong audience with the wrong creative produce cheap nothing. Judge the channel on incremental customers acquired, not on the price of reach.
Should a DTC brand start with Demand Gen or skippable in-stream?
Start with Demand Gen. It uses social style audience building across YouTube, Discover and Gmail, it accepts creative closest to what you already run on Meta, and it gets to a readable result fastest. The trade off is that it blends surfaces, so it tells you less about YouTube specifically. Once Demand Gen is working, add a separate skippable in-stream line with a genuine long form asset to learn what the channel can do on its own.
Can I reuse my Meta or TikTok creative on YouTube?
TikTok creative transfers reasonably well to YouTube Shorts because the surface, the aspect ratio and the two second hook window are broadly the same. Meta creative rarely transfers to skippable in-stream. On Meta you are interrupting a feed the viewer controls. On in-stream you are standing between the viewer and content they chose, with five seconds before the skip button appears. That is a different job for the opening, and most failed YouTube tests are failed openings rather than failed channels.
How do you measure whether YouTube ads are actually driving sales?
Do not trust platform reported ROAS and do not trust last click. View-through windows are generous, so in-platform numbers overstate the channel, while last-click models hand the credit to branded search and email, which understates it. Use a holdout or conversion lift study, a geo split test, or a post-purchase survey, and watch blended marketing efficiency, branded search volume and new customer count while the test runs. In 2026 independent incrementality testing was the measurement method senior decision makers trusted most, roughly twenty points ahead of media mix modelling.
About the author
Caner Veli is a DTC operator who has helped 350+ brands fix broken growth engines. He built Liquiproof from zero to 3,000+ global retailers in under 6 years. He now runs the same playbook, supported by AI systems he built himself, for DTC and CPG brands.