Ad Format

In the early years of online advertising, making an ad meant making one picture. A team would produce a wide strip and a square, then crop them to whatever dimensions each publisher asked for. The same picture ran beside search results, across the top of a news site, and along the bottom of a game app. Nobody was choosing an ad. They were counting how many places they had bought.

The results came back wildly uneven. In one spot two people out of a hundred clicked. In another it took a thousand views to get a single click. That is a twenty-fold gap with no vocabulary to explain it, so the meeting circled the same drain every month. Somebody would say the publisher seemed weak, and the team would switch publishers. When the same thing happened again in four weeks, they would change the picture instead. Six months of swapping publishers and pictures in turn left nobody any closer to knowing why only certain spots produced a reaction.

What made the difference was not the quality of the publisher but the state of the person sitting in front of the screen. Someone who typed "meeting notes tool" into a search box already feels a problem and is hunting for an answer. Someone flicking through a feed over lunch is hunting for nothing. In the first case a single line of text is the answer and no picture is needed. In the second, something has to stop the thumb before any copy gets read at all. The same human being reacts differently depending on which screen they happen to be on.

The idea of a format is a name put on that difference, and once the name existed the order of the work flipped. Teams used to build a piece of creative and then look for somewhere to hang it. Now they decide which state of mind they are going to meet and then build creative to the dimensions and conventions of that spot. An ad for a search results page has to land inside a headline of about thirty characters and a description of about ninety. An ad for a feed is read as a square image first and a line of copy second. The same message needs a different maker and a different number of days depending on the container.

That is why performance marketing puts format at the front of the list. Get this choice wrong and no amount of careful bidding or careful audience work later in the chain can undo it. Push search-style copy at somebody who was not looking for anything and doubling the budget only doubles the number of people who scroll past it.

A format settles two things at once: the place, meaning where the ad appears, and the form, meaning what shape it takes there. They are not treated separately because the place all but dictates the form. You cannot hang a video at the top of a search results page, and you cannot hang two lines of text in the slot that runs before a video plays. So teams talk about place and form as one unit, and they choose among four groups based on what the person in that spot is doing right now.

The first group stands in front of people who are already looking. Search ads are the only member. They appear only in the moment somebody types words into a search box, so the ad is not going out to find a person — the person emits a signal and the ad answers it. Because only people who already feel the need get caught, the share who go on to sign up runs several times higher than in any other format. At Pagedeck, visitors arriving from search converted to signup at around 6 percent while visitors from feeds came in under 1 percent. The catch is that the size of the pool is already fixed by how many people type those words. Double the money and you do not get double the people. Past roughly $2,200 a week, Pagedeck's budget started leaking into broad words like "meeting notes," and the people who arrived that way did not sign up.

The second group stops people who are not looking for anything. Four formats live here, and what separates them is how much of a stranger the ad looks like on that screen.

1. Display ads — banners laid into fixed slots on a website or an app. They read as advertising at a glance and never mix with the content around them. The unit price is the lowest of any format, but people have trained themselves not to look at those slots at all, so it takes a thousand appearances to earn about one click.
2. Social ads — posts slotted into a social media feed in the same shape as everything else there. They sit inside the scroll, so they do get seen, but the thumb is already moving, which means the first image and the first line decide everything.
3. Native ads — ads that borrow the shape of the surrounding content, appearing as an article in a list or a card in a recommendation rail. They don't interrupt the reading flow, which lifts the response, but a label marking them as advertising is mandatory.
4. Video ads — clips that cut in before or during another video. Sound and motion are available, which suits a product that takes explaining, but what you sell has to be visible within the five seconds before the skip button appears.

These four sit together because the problem they pose is the same. In all four the viewer did not come to see you, so if the opening frame fails to hold attention, the sentence after it never gets read. Work in this group starts with picking the one image that stops somebody, not with deciding what to say.

The third group has one action nailed to it. App install ads send a tap straight to the install screen in an app store rather than to a website, and the platform traces whether the install happened. They look like social or video ads, but because the goal is pinned to installs, the platform prioritizes people who look likely to install. Counting through to the install is the biggest practical difference. With other formats your own team has to stitch together what happened after the click; here the platform hands back the tap, the install, and the first launch as one connected line.

The fourth group calls back people who already came. Retargeting shows ads only to someone who visited the site once and left without doing anything. The form might be a banner or a feed post, so it overlaps with the groups above, but the audience is defined by visit history rather than by the slot. The same banner therefore pulls a response several times larger. An ad shown to someone who got as far as the pricing page and an ad shown to someone meeting the product name for the first time are the same picture landing on very different states of mind. The limit is that the audience is tied to the number of visitors, so it is no use as a way of growing scale.

The four are not parallel options sitting side by side; they feed each other in order. The fourth only works once the first three have brought people in, and the first only works once people know the words that describe your product. Nobody can buy search ads against a product name nobody has heard. Pagedeck keeps spending on the second group less for the signups it produces today than for the search terms it seeds. People who meet the product in a feed do come back weeks later by typing its name, and that traffic lands in the search ad column.

"This shouldn't run as display, it should run on search. These people are already looking."

If the audience feels the problem and is typing it into a search box, spend on being findable rather than on being noticeable.

"The product has to be on screen within three seconds. Don't open on the logo."

The slot is worth only the seconds before the skip button, so brand film conventions can't be carried over intact.

"If this goes out as native, don't drop the ad label. We get fined without it."

Confirming that a format shaped like the content around it carries a mandatory disclosure.

In the quarter Pagedeck first scaled paid advertising, Ravi, the growth marketer, put the same copy and the same image into all four formats. Two weeks later the numbers had piled up on one side. Search returned four clicks per hundred appearances, the feed returned one, and banners returned about one per thousand. Ravi's first conclusion was to switch banners off. Rebuilding the creative per slot changed his mind. A banner version stripped down to the two words "meeting notes" and a single screenshot tripled the response in the very same slot. The slot had not been bad. Search copy had been hanging in it.

When Noa, the product designer, pointed out that per-slot creative meant more work, Ravi cut the list to three formats and committed to making proper creative for those three. His reasoning was that three formats with real creative produce more signups on the same budget than four formats with something thrown together. When Grace asked in the quarterly review why video had been dropped, Ravi's answer was not an efficiency figure either. He had kept the formats the team could actually make good work for.

The most common mistake is believing one format is simply the most efficient and the others can be ignored. Pull cost per signup by format and search sits at the top every time, so the reading is an easy one to arrive at. But search is cheap not because the advertising is good — it is cheap because only people who are already leaning toward you get caught in it, and the number of those people is fixed by how many type the words. Pour everything into search and it works well for a while, then arrives at a stretch where the unit price climbs and the traffic does not. That is the point where leftover money starts flowing into unrelated search terms. The well is a fixed size and all you are doing is buying a bigger bucket. Switch off the other formats at that point and six months later fewer people are searching the product name at all, which dries up the search ads too.

Reading native advertising as a trick for hiding ads is the other frequent mistake. Something sitting in a list of articles looking like an article does invite the assumption of deception. But the point of borrowing the shape is not to conceal what it is — it is to avoid breaking the reading flow, and the label marking it as advertising is required by advertising regulation. Shrinking and fading that label to the edge of legibility loses two things at once. One is the fine when a regulator notices. The other is that people who feel tricked leave the moment they arrive. A campaign where clicks are climbing while time on page stays under ten seconds is usually in this state. Run a clearly labeled version against a faint one and the faint one often wins on clicks while the clear one wins on signups, because somebody who clicked knowing it was an ad is somebody with an actual intention to look.