US$0.15 divided by 25 page views, multiplied by 1,000, is US$6.00. That is the worked example on Google's AdSense help page, which gives the page RPM formula as estimated earnings divided by page views, times 1,000.1 The same page says the figure does not represent what a publisher has earned. It is a rate, and a rate has two parts. A higher rate comes from a larger numerator or a smaller denominator.
The two parts are usually treated as independent. In the usual account, earnings go up when a page carries more units, larger units, or units that move. The published research finds instead that the denominator responds to the numerator. An ad that is harder to ignore earns more per view and lowers the number of views, and the report counts only the views that remained.
Relevance and obtrusiveness cancel each other in the same ad
Avi Goldfarb and Catherine Tucker tested relevance and obtrusiveness together in a field experiment of 2,892 web advertising campaigns, averaging 852 survey takers each.2 Matching an ad to the site's content raised purchase intent on its own. Making the ad more obtrusive did the same. Ads that did both did worse than ads that did only one; combining the two, the authors write, "nullifies the positive effect that each strategy has independently."
The authors relate the failure to privacy concerns. The negative effect was strongest among people who refused to state their income, and in product categories where privacy matters most.2 On that reading, an ad that is both hard to ignore and well matched draws attention to how the match was made. A bubble is matched to a word printed on the page, not to the person reading it. AdBubbles sets no cookie, stores no identifier, and collects no personal data from readers.
The readers an annoying ad loses never appear in the RPM report
Daniel Goldstein, Preston McAfee, and Siddharth Suri put a price on the readers who leave. They paid 1,223 Mechanical Turk workers to classify emails, at three pay rates, beside good, bad, or no ads.3 Their 2014 paper defines the compensating differential as what a publisher would pay users to generate as many impressions with annoying ads as without them.4 The differential came to US$1.53 CPM (cost per thousand impressions) for bad ads against no ads, and US$0.38 CPM for good ads against no ads. Every one of the 21 ads rated most annoying was animated, and every one of the 24 rated least annoying was static. Animation alone, they write, is a cause of annoyance.
The 2014 paper in the Journal of Marketing Research concludes that "in plausible scenarios, the practice of running annoying ads can cost more money than it earns."4 A reader who leaves because of an animated unit does not load the next page, and views that never happen are absent from the denominator. The RPM figure then reports earnings against the smaller audience that stayed. It can rise while total earnings fall.
Chrome enforces a clutter limit whether or not the publisher measures it
Surveys of over 40,000 internet users in North America and Europe, run by the Coalition for Better Ads, turned that reaction into the Better Ads Standards.5 The mobile web definition of excessive ad density is "Ads that take up more than 30% of the vertical height of a page."6 Ad heights within the main content portion are summed and divided by that portion's height, sticky and inline units included. Google's announcement of Chrome's filter names prestitial ads, which cover the page's content, and flashing animated ads among the most intrusive experiences.5 Each occupies the screen without the reader's asking.
Chrome applies the standard. The announcement says sites are evaluated on a sample of pages and rated Passing, Warning, or Failing.5 A site still non-compliant 30 days after notice will have its ads blocked. The filter then refuses any request matching a list of known ad-related URL patterns, so the ad never displays. The page views stay, and the earnings on them do not.
A bubble waits inside the text the reader chose
An in-text bubble is built around the reader's action, not the page load. A publisher adds one script, the embed, and words already in the article get a light dotted underline. On a desktop, a short hover opens the bubble, and moving the mouse away or pressing Escape closes it. A phone reader taps the word to open the bubble and taps elsewhere to close it. Until then the page shows only the underline, which adds no height and no motion. What opens is a small panel with an "Ad" label, a headline of at most forty characters, a body of at most ninety, the advertiser's domain, and one link.
Nielsen Norman Group's 2018 review of banner blindness defines the effect as people's tendency to ignore page elements they perceive, correctly or not, to be ads.7 The review also reports that people attend to what relates to their goals. In one eyetracking example, the right rail held 25 percent of the content area and received 0.8 percent of the user's attention. A bought word is underlined inside the text the reader chose, and is looked at for that reason. Whether the reader then opens the bubble is the reader's decision, and the embed sends an open beacon only when a bubble has stayed visible for one second.
AdBubbles matches words on a page to an advertiser who bought them for that site and that month. AdBubbles does not choose which people see an ad, does not follow a reader between pages or sites, and does not promise any result. Targeting, in Goldfarb and Tucker's sense, is done against the page, and obtrusiveness is left to the reader, who opens the bubble or does not. A page that carries none of the bought words carries no ad at all.
A bubble cannot fill a rectangle the way a banner does
Both the publisher account and the embed cost nothing, and a publisher who verifies a site can request a crawl showing which of its words are currently sellable. The installation notes cover the scope, the part of the page the embed may scan, and the per-page maximum on bubbles. A site with few sellable words has little to sell this way, and the limits do not end there.
A banner earns on the page load. The Media Rating Council counts a display impression as viewable once at least half its pixels have been in the visible part of an in-focus tab for one continuous second.8 The Council's term for that is an opportunity to see. A banner can also carry an image, and it can be sold for a page whose text is unrelated to the advertiser.
A bubble has none of those properties. It appears only where a bought word is present in the scope, it holds text and one link, and earnings are shared among sites in proportion to opens. A page whose readers never hover or tap therefore earns nothing from the format. That dependence on a reader who chose to act keeps the page uncluttered, and it is also the upper limit on what the format can earn.