GOOGLE DEMAND GEN AD EXAMPLES: WHAT SPENDS WELL, BY INFERENCE

Most collections of google demand gen ad examples show you the finished frame and ask you to admire it clean product shot, tight headline, a colour palette someone clearly agonised over. That's the least useful way to learn from an ad. After enough launches I stopped asking "is this a good-looking ad" and started asking "what can I infer about how this will spend", because on live budgets those two questions have almost nothing to do with each other. The ad that wins the design review and the ad that wins the auction are rarely the same one, and the gap between them is the whole subject of this piece.

So I'm not going to paste a gallery of pretty screenshots. I'm going to walk how I read an example by inference what its asset coverage tells me about which placements it can serve, how much room the system has to optimise it, and whether it'll accumulate the signal it needs to leave the learning phase. If you want the platform background first, the primer on what Demand Gen is sets the stage; this article is the creative-diagnosis layer, the part nobody photographs because it doesn't show up in a thumbnail.

One framing before the patterns. A Demand Gen ad isn't a single image, it's an asset pack the system recombines on the fly across YouTube, Shorts, Discover and Gmail. So when I look at a demand gen google ads example, I'm not grading a picture I'm counting how many combinations it gives the algorithm to test. The examples that spend well are the ones that hand the system a deep, varied pack; the ones that stall are usually a single beautiful asset with nothing to test it against. Hold that distinction and almost everything below falls out of it.

It helps to remember what these campaigns are for. Google demand gen campaigns sit in the mid funnel: they're built for driving demand and brand awareness among people who aren't searching yet but are in a buying mindset, which is the territory the old Discovery ads used to cover before they were folded in. That mid-funnel job is exactly why the creative carries so much weight there's no keyword doing the qualifying, so the ad itself has to stop the scroll and move someone from passive interest toward action. An example that's read purely as decoration misses that it's the engine of demand generation, not the garnish on top of it.

How I read a Demand Gen ad example

When someone sends me an ad and asks why it isn't working, I don't look at the creative first I look at the asset coverage behind it. The four things I check, in order: does it supply all three aspect ratios (vertical, square, landscape), does it mix at least one video with the images, how many distinct headline and description variants it carries, and whether the message is native to the surface it's meant to serve. None of those is a judgement about taste. They're structural facts about how much inventory the ad can physically fill and how many combinations the system can run.

This is why a screenshot is such a poor teacher. The thumbnail shows you one frame in one ratio with one headline exactly the slice that tells you nothing about delivery. The same ad might be a deep pack with three ratios, a video and five headlines behind it, or it might be that single frame and nothing else, and the picture looks identical either way. So I read every example as a coverage map: what placements does this unlock, and what does the system have to work with. That's the same lens I bring to the broader Demand Gen asset specs the specs aren't a checklist you tick, they're the dimensions of the box the algorithm gets to play in.

There's a second layer I infer alongside coverage: fit. Demand Gen offers a handful of ad formats single image, video, carousel and product-feed ads and the strong examples pick the format that matches the product or service and the audience segments they're chasing, rather than defaulting to whatever was easiest to produce. A considered service with a story to tell leans on video; a catalogue of physical goods leans on a feed-driven carousel so the real products carry the ad. When the format obviously suits what's being sold and who it's being sold to, that's usually a sign the rest of the build was thought through too, because the same discipline that picks the right format tends to fill the right coverage.

What the creatives that spend well have in common

Across the accounts I run, the examples that scale share three traits, and being attractive isn't one of them. First, full ratio coverage: vertical, square and landscape for both images and video, so the ad is eligible everywhere instead of competing for a fraction of the inventory. Second, a mix of formats at least one video paired with images rather than going image-only because Demand Gen leans hard on YouTube and Shorts, and image-only simply can't serve there. Third, depth of text: several genuinely different headlines and descriptions the system can recombine, not one hero line with filler underneath.

The reason this works is mechanical, not aesthetic. Demand Gen finds winners by testing combinations and pushing budget toward the ones that convert. An asset pack with three ratios, a couple of videos, a handful of images and five distinct headlines gives it hundreds of combinations to explore; a single ratio with one image and one headline gives it almost none. Google's own data runs in the same direction advertisers mixing video and images tend to see more conversions per dollar than image-only and the inference I draw from every winning example is the same: the spend follows coverage and variety, because that's what the system has room to optimise. The polish is real, but it's downstream of whether the ad can serve at all.

The quality of the individual assets still matters it just matters second. High quality, high-resolution visuals with a clear subject build the brand trust that turns a click into a conversion, and tight, distinct calls to action give the system something concrete to test against. But quality only pays off once coverage lets the ad serve; a beautifully shot clip that can only fill a third of the inventory caps your return on investment no matter how good it looks. The examples that protect ROI do both high-quality assets and full coverage and when I have to choose which to fix first on a stalled ad, it's always coverage, because that's the constraint the budget is actually hitting.

Why the "beautiful" ad often loses

Here's the pattern that surprises people most. A team pours its effort into one stunning hero creative a single perfectly art-directed landscape video and it underperforms a scrappier ad that simply covers more ground. The beautiful ad isn't bad; it's narrow. One ratio means it never competes for Shorts. One asset means the system has nothing to rotate when a placement wants a different shape or a fresh frame. The effort went into making one combination excellent instead of giving the algorithm a deep enough pack to find its own excellent combination, which is the job Demand Gen is actually built to do.

I read these examples as a misallocation of effort, not of taste. The fix is almost never "make it prettier" it's "make it wider": cut the same concept into vertical and square, add a couple of image variants, write three more headlines. A merely good creative with full coverage beats a gorgeous one that can only serve a third of the inventory, every time I've compared them on a live account. That's the same instinct that runs through Demand Gen best practices generally: the system rewards the advertiser who hands it options, not the one who hands it a single masterpiece and hopes.

This trips up advertisers coming from the old Discovery ads especially. There, a single strong image often carried a campaign, so the instinct is to repeat that habit and ship one hero asset. But Demand Gen is an AI-powered campaign in a fuller sense its job is to assemble and test combinations, not to display the one frame you chose for it. Handing an AI-powered campaign a single creative is like hiring an analyst and giving them one data point: there's nothing to compare, so there's nothing to learn. The beautiful single asset doesn't fail because it's beautiful; it fails because it starves the very mechanism the campaign type was built around.

Why some examples never exit the learning phase

The thin-pack problem has a specific symptom: the ad never leaves the learning phase. The learning phase needs two things combinations to test and conversions to learn from and a single beautiful creative supplies neither. With one ratio, one or two assets and one headline, there's nothing for the system to compare, so it can't run the experiments that exit learning, and it stalls at low volume while the dashboard cheerfully reports the ad as "serving". Nothing throws an error. The ad is just quietly starved of the variety it needs to learn.

So when an example is stuck in learning, I don't diagnose the creative's quality I diagnose its depth. The questions are always the same: how many ratios, how many assets, image-only or mixed, how many text variants. Widen the pack and the same campaign that was crawling starts to move, because now the system has combinations to test and can begin accumulating signal. This is also why I'm wary of relaunching a stalled ad with a single "better" creative; that just rebuilds the same trap with nicer pixels. The escape from learning is breadth, and it's the same logic that decides whether a whole Demand Gen campaign ramps or sits flat the algorithm can only optimise what you've given it room to optimise.

Creative depth isn't the only thing that gates learning, but it's the one I see misdiagnosed most. Bidding strategies and the budget you set the pace, the audience you point at supplies the signal: custom audiences and tight segments of high-intent users or existing customers give the system cleaner conversions to learn from than a vague broad reach does. But none of that rescues a thin creative point a perfect bid and a perfect audience at one beautiful asset and you've still given the optimiser nothing to compare. When an example exits learning fast, it's almost always because the asset breadth and the audience signal were both built deliberately, not one at the expense of the other.

The examples that feel native to the surface

There's one more thing the strong examples share that's easy to miss because it looks like polish but isn't: they feel native to the surface they serve on. A creative built for YouTube Shorts is vertical, caption-first and a little raw it reads like content, not an ad. The same concept for Discover and Gmail is cleaner, more product-led, calmer. The advertisers who win aren't shipping one master creative and letting it crop; they're shipping assets that look like they belong on each surface, which is exactly why a centre-cropped landscape video on Shorts reads as a repurposed afterthought because it is one.

The inference here is about intent. When I see an example with a genuinely vertical, sound-off-legible, caption-led Shorts cut alongside a tidy square for the feed, I can tell the advertiser thought about where the ad lives, and that almost always correlates with the coverage and depth that make it spend. Native-feeling assets and full coverage tend to travel together, because both come from the same habit: building for the system and the surface, not for the slide that gets shown to a client. Google's own playbook lands in the same place its Creative Excellence Guide for Demand Gen pushes the rule of three assets per ratio and a video-plus-image mix, which is coverage and nativeness restated as official guidance.

What I notice actually moves spend

Strip away the gallery and the lesson is blunt: meeting the spec gets an ad accepted, but covering it fully and feeding it variety is what gets it spending. Demand Gen can only deliver against assets it has, so a thin pack however beautiful under-delivers, while a deep, varied, native one ramps. No bid or budget buys back the placements your creative can't physically fill, which is why I treat creative breadth as the real lever and the individual frame as almost incidental.

The signal I watch for this is ad strength, read as a coverage gauge rather than a grade. Google ties a strong rating to full coverage across all three aspect ratios for both images and video, which is precisely the coverage that lets an ad serve broadly and leave learning. An ad group with both orientations, video plus images, and several assets to rotate consistently shows stronger ad strength, and stronger ad strength means the system has room to push budget toward the combinations that convert. Varied ratios, a real format mix, depth of text and native-feeling assets that's what I've seen move spend on every account, far more reliably than any single "best" creative I could point to in a showcase.

The practical takeaway

If you take one thing from these google demand gen ad examples: stop judging the picture and start counting the coverage. The creatives that spend well aren't the prettiest, they're the deepest all three ratios, video paired with images, several distinct headlines and descriptions, and assets that feel native to each surface. The ones that stall in learning are nearly always a single beautiful asset with nothing to test against, and the cure is breadth, not a nicer redesign.

So I no longer read a demand gen google ads example as a thing to admire. I read it as a coverage map and a question: how many combinations does this hand the system, and which placements does it unlock. Get that right and the algorithm has room to find its own winner and spend into it; get it thin and even a gorgeous ad sits at the edge of the learning phase, technically serving and going nowhere. Build the pack wide, make it native, and let the system do the part it's actually good at.