Observations on Advertising
An analysis earlier this year by the Internet advertising-services firm Doubleclick revealed that only 4-5 people out of a thousand click on Internet ads -- the continuation of a long downward click-through trend. Figures like those once fed skepticism about the effectiveness of advertising online, particularly when the dot-com bubble burst.

Not anymore. These days, advertisers count click-throughs where they matter -- such as in sponsored links like those that Google provides -- and elsewhere they assess impact based on brand awareness and by using other measures. This is one of the reasons that Internet revenue is increasing.

The observations below from Eyetrack III are a different indicator of where and how ads are effective, based on what the researchers saw as 46 participants viewed mock news website homepages and article-level pages that we created specifically for research purposes. Five website designs contained a variety of ad formats and presentations, and we could compare their performance. (Unlike other parts of Eyetrack III, the advertising testing was not conducted using tightly controlled variables. Ergo, we offer these observations rather than scientifically sound findings.)

Observation: People avoid looking directly at some ads while looking at others; performance depends on placement.

The so-called "invisibility effect" of Web advertising is no myth. We found among our test subjects that there were often instances when they did not look directly at ads, even for a fraction of a second. That doesn't mean they didn't see them at all -- in some cases eyes fixated close enough to the ads to be able to view them in peripheral vision; in other cases they looked at ads directly; and in many cases, they didn't see them at all. And placement of ads mattered a lot.

That being said, let's put the viewing numbers within context of click-through rates, which are typically in the low-single-digit percentages. The ad viewing percentages in Eyetrack III range between 14 and 68 percent. This is good news for advertising that focuses on branding, because it says that even if the units are not getting clicked on, the percentages that are being viewed are higher than click tracking might suggest.

Take a look at the images below (click to enlarge). The left image is how one of the mock homepages (No. 8) looked. The colorful image on the right is a heatmap or aggregate representation of the eye fixations of all the participants who looked at this page. Colors indicate the percentage of participants who looked at specific parts of the page. Red-orange means most people set their gaze in that spot; blue-to-gray means fewer did. (There's a key along the top of the heatmap image.)

Homepage No. 8

Heatmap - homepage No. 8

Notice how the heatmap image shows that the majority of the group we tested focused on the editorial content (in this case, headlines and blurbs) but the direct gaze largely avoided the large ad on the left. You can see the significantly smaller number of participants who looked directly at the ad (at least briefly), and in this case it was a very large ("skyscraper") ad.

Between 50 and 60 percent of the group glanced directly at the left-column ad at least once, which isn't bad (though still much less than performance of the nearby editorial content). However, the results were not so good for the top banner ad on that page; it was mostly "invisible," with only 20-30 percent of the group ever looking at it.

That particular homepage was unusual because we put the navigation elements in the right column (a practice that's uncommon, but not unheard of, among news websites) and an ad in the left column. Our experiment suggested that when ads are put in the left column, they perform better than right-column ads. (This is supported by Eyetrack III observations on homepage viewing behavior, which show that normal initial eye movement around the page focuses on the upper left portion of the screen.)

In examining ad-viewing behavior across four distinct homepage designs, we found that on average, most ads were seen by less than half of the group.

Here's a table showing the percent looking directly at banner ads based on their placement on the page.

Overall, based on looking at all the homepage designs tested, advertising placed on top and in the left column of the page was viewed most. Ads placed at the bottom of the page were viewed least. When viewing homepage No. 8, more than 50 percent of participants moved their eyes all the way down the page to the last headline. However, fewer than 25 percent looked beyond that headline to the ad beneath it.

Observation: Visual breaks in design serve as barriers to seeing ads.

As seen in the heatmaps (aggregate images) produced for our collection of homepages, participants tended to avoid ads when a visual barrier of either white space or a border sat between the ad and the editorial content. A border or rule, or a visible area of white space, seemed to stop many people from viewing an ad.

An example of this was seen with homepages No. 4 and 9 (which are variations of the same design). Click the images below to see these pages.

Homepage No. 4

Homepage No. 9

Most participants who saw these homepages stopped at the end of the headline and blurb under the "Lifestyle" heading; they did not tend to see the VW ad beneath, in the lower right of the page. What's interesting is that on homepage No. 9, there was more white space between the blurb and ad -- and that page had less viewing of the VW ad than did No. 4. More white space between editorial and ad content seems associated with less eye action on the ad.

Below are closeups of this section of homepage No. 4:

Here are closeups of this section of homepage No. 9:

Click the images below to see full-page heatmaps of those two pages.

Heatmap - homepage No. 4

Heatmap - homepage No. 9

Observation: Ads that blend into the look and feel of the page -- especially text ads -- draw more eyes.

The Eyetrack III researchers found that ads that blended into surrounding editorial content on a news homepage attracted people's eyes more often than ads that featured contrasting colors or designs. If an ad was approximately the same background color as the rest of the page, it received more eye fixations.

An example of an ad element that was viewed more -- perhaps because it had the same background color as the site itself -- appeared on homepages No. 4 and 9. The large horizontal ad for VW ran the full width of the page and was positioned above the flag/masthead, with a white background.

Here's a detail of this ad, followed by its heatmap:

(Note: This particular mock website has a name that resembles a real news site in Michigan; they are not the same, as we tested mock websites designed for this research.)

A contrasting background color -- instead of the predominant white -- likely would have been seen less, based on what we observed on other pages tested.

We believe this ad also generated a greater percentage of people viewing it than some ads due, in part, to visual bleed (spillover) from the editorial content, specifically the nearby site flag/masthead.

In this study, the site flag/masthead on all the homepages received considerable visual traffic. Researchers speculated that this may have been a result of the fact that this trial featured fictitious prototype news sites that were modeled after real sites, but which were unfamiliar to participants. When looking at the viewing sequence across the group for each of these homepage designs, the participants overwhelmingly looked at the flag early in their visit to each site. Since the flag is the "editorial content" that brought visual attention within the vicinity of the banner ad, it is possible that regular users of a favorite site would not focus as much on the flag because they would already be familiar with the site.

For homepages No. 2 and 7, we included text advertising rather than web banner ads. These were job listings for the San Francisco area (where Eyetrack III testing was conducted).

Here's a detailed view of the text ads for homepage No. 2. (The corresponding text ads for page No. 7 were identically designed, but extended further down the page.)

These text ads had the highest viewing of all ads on all of our homepages, by far. Overall, 82 percent of people who visited homepages No. 2 and No. 7 looked at the text job ads. (The next highest performer was a "skyscraper" ad, which was viewed by 68 percent of participants.) On homepage No. 2 -- a compact page with fewer content choices than other homepages -- the text ads attracted viewing from 90 percent of study participants.

The text ads on these pages recorded the longest average duration of viewing. On average, participants spent 6.9 seconds perusing the text ads. The longest a banner ad was viewed, on average, was 1.6 seconds.

Observation: Size matters; "half-page" ads perform well.

Larger ads are seen before smaller ones, our observations suggest.

The larger ads on homepages do not, however, get viewed for any longer than smaller ads. With the exception of text ads, as noted above, all banners of all sizes on homepages were seen, on average, for between 0.6 and 1.6 seconds. That doesn't sound like a lot of time for a website advertiser to make an impression, but for branding purposes that might do.

We included some extra-large ads on some of our article pages -- "half-page" banners, sized at 368 x 850 pixels. (Here's an example.) On those pages, the half-page ads received more average fixations per person than any other ad size presented alongside articles, as the chart below shows.

As you can also see from this chart, the next best-performing ads were 300 x 250 pixels (example) -- medium size, placed within article text -- followed by large "skyscraper" ads (example) placed in the left column of some of our article pages (160 x 800 pixels). Smaller ads placed on article pages -- including the ubiquitous 468 x 60 banners -- did poorly in comparison. Indeed, the 468 x 60 banners just barely out-performed smaller right-column ads (184 x 90 pixels).

Observation: Ads inset within article text are seen more than most other ads.

Notice in the chart in the item above this that a 300 x 250 pixel ad inset into text of an article performed better even than "half-page" (368 x 850) ads on article pages, in terms of the percentage of people who actually looked at the ad while viewing the page.

On visits to article pages with the ad inset into the text, 56 percent of our test participants looked at the ad -- giving it an average of 4.6 eye fixations while they were viewing these pages. No other size ad on article pages got looked at by more than half of our test pool. The half-page ads did get more fixations per person on average, though, among those who looked at them -- that is, the people who looked at them were more engaged with the half-page ads. This means that both the half-page and the text-inset ads performed well, but each achieved different results.

Observation: Mouseover-expand ad were viewed more than other banner ads.

On homepages No. 5 and 10, we presented identical designs with one difference: One page had a static 468 x 60 pixel banner ad at the top of the page below the flag/masthead; the other page had the same size and position ad, but when the user moved his/her mouse over the ad, it expanded downward (temporarily covering editorial content) to reveal a much larger ad. (There was no clue on the mouseover ad that it would expand. Click the thumbnail below to see the expanded-view ad.)

Half of our participants saw homepage No. 5, the other half saw No. 10. Article pages associated with these homepages also used either the standard banner or the mouseover-expand version.

Homepage 10 - expanding ad view

The mouseover-expand ad performed very well in comparison to non-expanding ads. Only 40 percent of the people who looked at homepage No. 5 fixated their eyes on the 468 x 60 pixel ad, while 93 percent of people who visited homepage No. 10 looked at the expanding ad in the same position (but not that many triggered the mouseover expansion).

Part of the explanation may be that the ad on page No. 10 included a close-up of a handsome young man's eyes. Other observations from our research have established the draw of images of the human eye. The non-expanding ad on page No. 5 had a less-compelling image.

On homepage No. 10, 64 percent of people triggered the expanding part of the ad by moving their mouse over the banner -- which is significantly more than the number who saw the ad at all (40 percent) on homepage No. 5.

There was no clue that this ad expanded; the only way that a person would discover the expand behavior was to move the mouse over it. It might have been the model's eyes that led to this common mouse movement. More likely, it was the common user behavior of concentrating on the upper left portion of most homepages (as demonstrated by the homepage viewing patterns findings elsewhere in Eyetrack III).

The expanding ad was viewed for almost 2 seconds, on average, by those who saw it. That makes it the best-performing banner ad in our test -- though it's a distant second to the right-column text ads (job listings) that we published on one set of homepages.

The mouseover-expand ad also ran with article pages associated with homepage No. 10, and we saw similar behavior.

The mouseover and non-mouseover banner ads on homepages No. 5 and 10 were placed below the masthead/flag, which we believe improved the ads' performance. These ads did better in terms of eye fixations than the same size (468 x 60 pixel) ads positioned above the masthead/flag. We believe that this placement close to the editorial content of the page improved the chances of being viewed.

Observation: Static ads vs. animated ads revealed mix results.

For the homepages portion of Eyetrack III, we mostly used static ads. On one homepage where we did use an animated ad -- a "skyscraper" in the left column of homepages No. 3 and 8 -- we noticed a modest, but nevertheless significant, increase in spillover from the adjacent editorial content to the ad, which we believe was mostly a result of the animation pulling the eye over to investigate.

For our article pages, we included a mix of animated and static ads. Perhaps surprisingly, we couldn't identify a clear trend. As you can see from the chart below, best performance of static vs. animated ads varied by ad type. Static ads got a higher number of average fixations for most ad types, but not for "skyscraper" ads. Static ads were seen by more test subjects for most ad types, but animated ads inset in article text reversed that trend; animated in-text ads were seen by four times as many people as static ads in that position.

Observation: Ads closer to the top left part of the page are generally seen before ads elsewhere.

Ads located at the bottom or bottom right of a homepage get fewer viewings than ads positioned elsewhere. Ads closer to the upper left tend to be seen much more quickly -- within the first 5 seconds, on average. Ads in the right column were first seen between 12 and 45 seconds after a participant had entered a homepage. And two ads that we placed at the bottom of some of our homepages weren't seen for 29 and 96 seconds, respectively. (These numbers include only those people who did see an ad, with their time-to-view averaged. The bottom ad that took 96 seconds to be seen was seen by only 10 percent of participants.)

Observation: Ad creative quality, content affect viewing behavior.

Differences in the content of banner ads can affect how they are viewed. Some images in ads perform better than others. (We saw that above in the mouseover-expand ad where the closeup of a handsome man drew more and longer fixations than another, less visually interesting ad placed in the same position.)

On article pages in our Eyetrack III testing, we rotated several creatives in some placements. These were not tightly controlled variables, but we can still offer some observations.

"Skyscraper" ads. On one set of our test article pages, we included a 160 x 600-800 pixel ad in the left column. Ads rotated included:

• American Express/Marriott: cream-colored, with text and a small image of a couple two-thirds down the ad (not visible without scrolling); animated text scroll
• IBM: bright blue background with text only, no images; animated text scroll
• IBM: bright orange; type plus graphic images; animated text scroll
• Auto Europe: static ad with several photos of a castle and various automobiles, plus type; blue background

The Auto Europe ad (the only one not animated) got 3-6 more viewings than the other ads. Thirty-two percent of people who visited article pages containing this ad looked at the ad; those who looked at the ad averaged 3.1 fixations.

The ad that got the most fixations (among those who looked at the ad at all) was the blue IBM ad, with 6.8. That ad contained nothing but text on a color background.

We hesitate to draw conclusions from this, but it would appear that the photographs in the Auto Europe ad drew more eyes in the first place, but didn't hold them as long as did the animated graphical text ad.

Half-page ads. On one set of our test pages, we included "half-page" ads (336 x 850 pixel) at the article level (but not on the homepage). We rotated three ads through these pages:

• Exxon: white background; mostly text, with animation; small static image halfway down ad
• British Airways: dark blue background; text "above the fold"; series of illustrated graphics; with animation of text and illustrations
• HBO's Carnivale: colorful animated graphic and text

The HBO ad generated the most intense viewings; among those who looked at the ad, there were an average of 9.6 fixations. That compared to only 2.7 and 3.0 fixations for the British Air and Exxon ads, respectively.

The HBO ad seemed, to us, to be the most visually interesting. However, only 10 percent of people who visited pages including the HBO half-page ad ever looked directly at it to record a single fixation. British Air was looked at by 37 percent; Exxon was looked at by 30 percent.

We noticed this pattern elsewhere. It's fairly often the case that ads so visually compelling that you'd expect people to look at them are avoided entirely by the majority of people exposed to the page -- but those that do look directly at them spend more time exploring their content.

Less visually interesting ads seem to attract more eyes, but less intently. Could it be that the "louder" ads lead people to avoid them altogether, whereas subtler and less flashy ads draw more people in?

Ads inset into article text. On one set of our test pages, we included ads (336 x 850 pixel) inset into article text. We rotated four ads through these pages:

• Einstein: image and graphic; animated text; question and radio buttons
• Jennifer Aniston: Text and three photos of women; photos move up and down for several seconds to attract attention
• Mercedes: Text and images of two cars; animated
• Atkins: Text and photo of woman in bikini from behind; no animation

The best performing of these ads in terms of average number of eye fixations (among those who looked at the ads at all) was the one featuring Jennifer Aniston, with 6.6 (followed by the Atkins bikini ad at 5.7). This fits with results from our testing that showed the strong draw of people's faces. This ad featured head shots of three well-known actresses, and the images moved up and down.

When we look at these ads in terms of percentage of people visiting the page, we again see an inverse order. That is, while the Aniston ad was examined most intently by those who did look at it, it had the smallest percentage of page visitors looking at it at all (15 percent).

The Mercedes ad had the lowest average number of fixations, but the highest percentage looking at it (30 percent). Again we see the ads we consider least visually interesting looked at most often and less intently, and the most compelling ads seen less often, but -- when looked at -- they are looked at most intently.

Observation: Small pop-ups are quickly viewed, then closed or hidden.

On homepages No. 4 and 9, a small pop-up ad (200 x 200 pixels) appeared the first time a participant saw the homepage. (On subsequent visits, the ad was programmed not to reappear.) The interaction with the ad across the two groups of people who saw it was very similar.

Approximately 70 percent of participants did see (look directly at) the pop-up. The other 30 percent never looked at the ad.

The ways in which people dealt with the pop-up ad typically fell into two categories. The most common participant behavior was to close the pop-up window within 3 seconds of it becoming visible; typically, the person looked at it just long enough to close it. The other common behavior was to ignore it -- not to try to close it, but not to look at it either -- and then after about 25 seconds (on average), to click somewhere else on the page, which caused the pop-up to disappear under the viewed webpage.

Of the participants who did look at the pop-up, on average they had 2-3 fixations on it for a total viewing time of, on average, 1 second. This is within the same average viewing time range as other banner ads on the homepages. The pop-up didn't perform any better or worse.

No participant clicked on the ad -- only on the "close" box to get rid of it.

It's worth noting that this was a small pop-up ad. Participant viewing patterns might have been different on a larger pop-up.

One unexpected consequence of the pop-up ad was that it brought attention to the top of the page by appearing on top of the banner ad above the flag/masthead. Even though the pop-up ad itself did not receive significant viewing, it did draw attention to the VW banner ad placed immediately underneath it.

In at least one notable example, the person was so drawn into the banner that he/she read it completely. The image below is a single-session page detail of the VW ad as viewed by this individual. (The orange lines indicate deep reading of the content; thin lines indicate a quick gaze path between two fixation points.) As a sole bit of data, it's not particularly useful -- but it is fascinating to see how someone interested in an ad viewed it over a period of a few seconds. Perhaps this person needed a new car?

Tips

Here are some tips based on what we observed about advertising in this part of the research. You might consider these Eyetrack III advertising observations when placing and pricing ads on your homepage.

• If you are responsible for creating the ad content for your advertiser clients, think about making sure that your design can deliver its message in a single glance, because that might be all you'll get. If you want to insert more text on an ad unit than can be consumed in a single (less-than-a-second) glance, then assume that the unit will have one glance to hook the reader's attention. Once hooked, you have the opportunity to draw the reader in closer, but only if that initial hook is effective.
• Consider designing news homepages so that ads are not set apart from editorial content too much with horizontal or vertical rules and excessive white space, which can act as barriers to viewing ads.
• The researchers' observations suggests that you'll get better viewing for banner ads that do not contrast too severely with surrounding editorial content. An ad that broadcasts "I'm an ad!" by using bright, contrasting colors sometimes has the opposite of the intended effect. (Of course, the content of a contrasting ad can be compelling enough to counter this tendency; as is so often the case, the quality of the content can override other factors.) We are NOT recommending that ads be presented as camouflaged editorial content. While that may attract more visual traffic, this practice would diminish your credibility.
• Text ads work, in part, because they look similar to editorial content, and that helps bring visual traffic to them. It also helps if they are in close proximity to editorial content, which helps draw eyes.
• Size isn't always the dominant factor in Web ad performance. To get the most people to actually look at an ad (for them to fixate on it for at least a fraction of a second) on an article page, insetting it into the text flow seems to work better than any other placement. But in-text placement may not give you the most intense user engagement with an ad; sheer size appears to perform better in this regard.
• You might consider using expandable banner ads if you want better performance than static ads offer. If you do, you might want to let viewers know the banner will expand. It's not always a great idea to surprise users with this sort of behavior. If you're going to use a mouseover-expand ad, we suggest positioning it in a normal path of user mouse movement as a way to get the expanding part of an ad seen by a lot of people. (Note: We only tested a mouseover-action ad, but many sites now use ads that automatically appear on top of editorial content and must be closed by the user. This would make for an interesting future eyetracking test. We can make an educated guess based on these findings that such an ad would be seen by most if not all users. The trade-off is that such ads annoy some users.)
• Should you choose to use pop-up ads on your homepage, be aware of their poor performance relative to other ads.

Add/view feedback on this report