Search Only Catches People Who Already Know You Exist

There is a pattern I see in almost every DTC Google account I open. Brand Search is delivering an eye-watering return. Performance Max is comfortably above target. Shopping is fine. The whole account looks like the healthiest line in the media plan, and the founder is quietly wondering why they are spending anything on Meta at all.

Then you pull the new customer split, and the picture inverts. Most of that Google revenue is people who already knew the brand, typed the name, and were going to buy anyway. Median Google Ads ROAS across DTC stores sits around 4.07 against roughly 2.70 on Meta, and the gap is not skill. It is job description. Google is largely catching demand. Meta and TikTok are largely making it. When Meta gets expensive and you shift budget into the channel with the better reported number, you are not improving efficiency. You are quietly defunding the thing that fills the funnel.

Demand Gen is Google's answer to that gap, and it is the campaign type most DTC operators have either never run properly or tried once and switched off. It deserves better than that, but it also has to be run as what it is rather than as another performance campaign with a different name.

What Demand Gen actually is

Demand Gen is Google's visual, feed-based campaign type. It runs across YouTube in-stream, YouTube Shorts, Discover and Gmail. The ad units are images, carousels and short video, not text. It replaced Discovery campaigns and inherited a lot of their reputation, most of which is out of date.

The important thing to internalise is that it behaves like a social channel that happens to be billed by Google. The user is not searching. They are scrolling. Your ad has to earn the stop, the same way it does on Meta. Everything that follows from that, the creative you need, the timeline you allow, the metrics you judge it on, is closer to paid social practice than to anything else in your Google account.

The incrementality argument, in one number

Roughly 68 percent of Demand Gen conversions come from people who had not seen that brand's Search ads in the previous 30 days. That single figure is the whole case for the channel. These are not customers you were about to win through another line in the same account. They are additions.

It also explains why Demand Gen looks bad in a last-click report and looks fine in a holdout test. If you attribute on last click, a customer who first met you in a Shorts ad in week one and searched your brand name in week three shows up as a brand Search conversion. Demand Gen paid for the introduction and Search took the credit. Most accounts are running on exactly that reporting model and drawing exactly the wrong conclusion from it.

The practical fix is not a better attribution model. It is a new customer metric and a geo holdout. Track new customer acquisition cost rather than blended ROAS, and if the spend is meaningful, hold out a region for four weeks and read total revenue rather than platform revenue. That is a fortnight of setup that settles an argument brands otherwise have every quarter.

How to structure it for a DTC brand

Attach the product feed first. A Demand Gen campaign with a Merchant Center feed behind it is a shoppable format and sits in the 0.5 to 2 percent conversion band. The same campaign without a feed is an awareness unit and converts under 0.05 percent. If your Shopify feed has missing GTINs, thin titles or stale pricing, fix that before you spend a penny here, because feed quality is the ceiling on everything downstream.

Then separate prospecting from retargeting into different campaigns, with different budgets and different targets. Google will happily let you blend them, and the blended campaign will always report well because the retargeting half carries it. You will learn nothing about whether you can acquire. Keep them apart so the prospecting number has to stand on its own.

On audiences, seed lookalikes from your best customers rather than all customers. Push a list of your repeat buyers or top LTV decile from Klaviyo or Shopify into Google, and build the lookalike from that. Layer in the new customer acquisition goal so bidding is optimising for people who have not bought before. This is where most accounts leave the largest amount of money on the table, because the default is to upload everyone who ever placed an order and let the model average your best customers with your worst.

Start on a target CPA rather than a target ROAS while the campaign is learning, and set that target at the CPA you can genuinely afford on a first order rather than the one you wish you were paying. Move to target ROAS once you have consistent conversion volume. Sitewide tagging and enhanced conversions are not optional here, they are the difference between the model seeing your buyers and guessing at them.

The creative bar is higher than people expect

The rule of three is the baseline: at least three images or videos in each aspect ratio, vertical, square and landscape, in every ad group. Vertical video matters most because Shorts is where the cheap reach lives, and a vertical cut of an existing Meta asset will usually outperform a resized landscape hero image by a wide margin.

Keep text overlay under about 20 percent of the image. Put the call to action inside the visual rather than relying on the headline field. Cut for the surface rather than resizing one asset three ways, because a Discover card and a Shorts ad are being consumed in completely different postures. Google's own guidance is that advertisers hitting the bar on at least three of the four pillars, audiences, bidding, creative and data signals, see over 40 percent more conversions, and creative is the pillar most brands fail.

The good news for anyone already running paid social properly is that you have most of this. Your best-performing TikTok and Reels assets are the right shape, the right length and the right tone. Demand Gen is often the cheapest place to get a second life out of creative you have already paid for and already validated.

The four mistakes that make it look broken

The first is killing it early. Demand Gen needs four to six weeks before the numbers mean anything, partly for the learning phase and partly because the people it reaches were not in market on the day they saw you. If you cannot fund six weeks at a spend level that generates real conversion volume, do not start. A starved campaign that never exits learning is worse than no campaign, because it produces a confident and wrong conclusion.

The second is comparing it to Search. Benchmark it against your Meta and TikTok prospecting numbers. Google says the same thing, and it is not a defensive position, it is just an accurate description of where the ads appear and what the user is doing when they see them.

The third is running one campaign that mixes prospecting and retargeting, then celebrating a number the retargeting produced. The fourth is treating it as a set and forget budget line. Demand Gen fatigues on roughly the same curve as paid social, which means a refresh cadence, not a quarterly review.

Underneath all four is one honest limitation. Demand Gen gives you less placement transparency and less granular control than you would like, and the reporting is thinner than Search. You are handing more of the decision-making to the model than you might be comfortable with. That is a real trade, and the way to make it survivable is measurement discipline at the account level rather than control at the campaign level.

What this looks like in practice

For a brand doing between 200,000 and 500,000 a year, the sequence I would run is this. Week one, clean the Merchant Center feed and confirm sitewide tagging and enhanced conversions are live. Week two, export the top LTV decile from Shopify and Klaviyo, push it into Google, and build lookalikes from that list rather than the full customer file. Week three, launch one prospecting campaign with the feed attached, target CPA set to a genuine first-order affordability number, and nine creative assets across the three aspect ratios with vertical video weighted heaviest.

Then leave it alone. Read it at week six on new customer acquisition cost, not blended ROAS, and cross-check what happened to brand Search volume and direct traffic over the same window. If Demand Gen is working, you will usually see it in those two lines before you see it in the Demand Gen conversion column.

If it clears the bar, add the retargeting campaign as a separate line and start a creative refresh cadence. If it does not, the diagnosis is almost always feed quality, creative that was resized rather than cut, or a lookalike seeded from the wrong list. It is rarely the channel.

Inside the system

How we build this for brands

The seed list is the part that most brands get wrong and it is the part we automate first. A reporting agent reads live Shopify and Klaviyo data, builds the top LTV decile rather than the full customer file, and refreshes the audience push on a schedule so the lookalike does not drift as the customer base changes. The same agent watches new customer acquisition cost as a separate line from blended return, so a campaign that is buying repeat buyers cannot hide inside a healthy-looking ROAS.

On the creative side, our VOC engine mines reviews and support messages for the objections and the language customers actually use, and that feeds the briefs for vertical video and carousel assets across Shorts, Discover, Meta and TikTok at the same time. Assets get cut for each surface rather than resized, and performance comes back into a profit dashboard built from live Shopify and ad data so the decision to scale is a margin decision rather than a platform one. Part of this runs live for portfolio brands today; the full system is what we deploy when we take a brand on.

Frequently asked questions