Display Advertising: What It Is and How It Actually Works

What display advertising actually is

Display advertising is any ad that shows up as a visual unit on a page, app or video — as opposed to a text link next to a search result. A banner on a news site, a video that plays before a YouTube clip, a rich-media unit that expands when you hover over it, a product ad that follows you after you leave a retailer's site — all of that is display.

It comes in a handful of standard formats: banner ads (static or animated rectangles, sized to fit common ad slots), rich media ads (interactive, expandable, sometimes with embedded video), video ads including in-stream and in-display units that run inside a video player or feed, and native or sponsored display units that are styled to match the page they sit on. Google groups most of these under Google Display Ads, running across the Google Display Network — Gmail, YouTube, millions of partner sites and apps. Amazon groups its equivalent under Amazon DSP, running across Amazon's own properties and a smaller set of partner exchanges.

The thing all of it shares: you are buying attention on a page the person did not search for. That is different from search advertising, where the person typed the intent themselves. It changes what display is good at — reach, repetition, staying in front of a shopper who has not decided yet — and what it is bad at, which is proving on its own that a sale happened because of the ad and not despite it.

How a display campaign actually runs

Almost none of it is manual. When a page loads an ad slot, an auction runs in the background — usually in under a second — where advertisers bid on the chance to show that person an ad, based on who they are, what they've done recently, and how much each advertiser is willing to pay. Google's AI (or Amazon's, or Criteo's) picks the format and size automatically to fit the slot. You don't design forty banner sizes by hand; you upload a responsive asset set and the platform assembles the rest.

What you control is: the audience (interest, behavior, past-purchase, remarketing lists), the budget, the bid strategy, and the creative inputs. What you don't control, precinct by precinct, is which exact page your ad lands on inside a network of millions of sites — you're buying an audience, not a placement, unless you specifically restrict inventory.

Picture a kitchenware brand like HexClad running remarketing display against people who viewed a product page and didn't buy. The ad follows them for a set window, at a frequency cap, until they convert or the window closes. That's the mechanic behind most retail display — simple in concept, and easy to get wrong in the details covered below.

What display advertising costs — the math, with real numbers

Display is priced almost entirely on CPM — cost per thousand impressions — occasionally on CPC for specific formats. Neither Google nor Criteo publishes a flat rate on their own pages, and that's the honest answer: it isn't a fixed price, it's an auction, and it moves with competition, audience, and format. If a source quotes you one number for "display advertising cost," treat it as a snapshot, not a rate card — check current pricing directly and ask what's included.

Here's what that structure looks like on a real book. Across 30 advertisers Full Circle and reMKTR managed on Amazon DSP in July 2026, the spend bought 78.4 million impressions at a $4.00 CPM. Run the CPM math forward — impressions divided by 1,000, times the CPM — and that's roughly $313,600 in media behind those impressions. Blended cost per click across formats came out to $1.42. Cost per acquisition landed at $5.49 across 57,137 attributed purchases, 20.1% of them new to the brand, for a 6.04x return on ad spend.

The table below lays out that same math as a template — the metrics that matter, in the order they compound, so you can run the same arithmetic against your own numbers.

Google Display, the open web, and Amazon DSP — how the inventory differs

Three pools of inventory get called "display," and they're not interchangeable. Google Display Ads run across Google-owned properties (YouTube, Gmail) plus partner sites and apps in the Google Display Network — strong for broad reach and self-serve setup, with Google's AI handling creative sizing and bidding. Criteo and similar commerce-media players buy across the open web, tied to shopper and product data, and lean hardest into retargeting and retail-specific audiences. Amazon DSP buys across Amazon's own pages plus a smaller set of partner exchanges, tied to Amazon's purchase and browse signals — the advantage there is that the same shopper data driving your Sponsored Products campaign is available to the display buy.

One clarification worth making plainly: DSP here means Demand-Side Platform — the programmatic buying system — not Delivery Service Partner, the Amazon courier franchise. Same three letters, completely different business. If you landed on this page looking for the trucking opportunity, this isn't it.

None of the three is universally "better." Google's reach is unmatched if your audience isn't primarily Amazon shoppers. Criteo's strength is open-web retargeting tied to commerce data across many retailers, not just one. Amazon DSP's strength is narrow and specific: it's the only one of the three where the ad, the product, and the purchase all live inside the same ecosystem, which matters if what you actually want to know is whether display moved a shopper who was already looking at your listing.

The mistake that quietly ruins display measurement

The most common error, and one we've made ourselves early on a book: crediting a display impression with a sale because the shopper clicked the ad and bought within the attribution window — without checking whether that shopper would have bought anyway. Last-click attribution counts the click. It cannot tell you what would have happened without it. A shopper who already had the product in their cart, saw a retargeting ad, clicked it out of habit, and bought — that's not incremental lift, that's a click stealing credit from a sale that was already happening.

The only honest fix is testing, not smarter tagging: holdout groups who see no ads, matched control groups who look like your buyers but aren't exposed, and reconciling the result in a clean room like Amazon Marketing Cloud so display and sponsored ads stop double-counting the same purchase. It's more work than reading a dashboard. It's also the only way to know if display added anything or just repainted demand that was already there.

When the number looks wrong — what to check first

If your display ROAS looks implausibly high, check the attribution window before you check the creative. A 14-day or 30-day click window will catch purchases that had nothing to do with the ad. If frequency is climbing past 8–10 views per person per week with no lift in response, you're paying to repeat, not to reach — cap it. If a holdout test comes back flat, don't kill the channel on one read; check whether the control group was actually matched on past purchase behavior, and whether the test ran long enough to clear a normal purchase cycle for your category.

And if the platform's own dashboard and your last-click view disagree, that's not a bug — that's two different counting methods giving two different honest answers. Reconciling them, not picking whichever number looks better, is the actual job.