The Coalition for Better Ads and a paper in the Journal of Advertising give different answers to one question: whether a page can carry a measurable amount of advertising beyond which readers leave. The Coalition says yes and gives a figure: on a phone, ads covering more than 30 percent of a page's main content height are one of the experiences its standards name.1 Cho and Cheon, writing in 2004, examined three explanations for why people avoid ads online: first, perceived goal impediment; second, perceived ad clutter; and third, prior negative experience. Goal impediment, the sense that an ad is stopping the reader from doing what they came to do, was the strongest.2 On that reading, what an ad interrupts sets the limit, and a share of the screen is only a proxy for the interruption.
The published thresholds measure height, not the number of ads
The Coalition's definition is exact. On a phone, ad density "is determined by summing the heights of all ads within the main content portion of a mobile page," then dividing by that portion's total height.1 Sticky and inline units count; headers, footers, navigation, and anything below the main content do not. The desktop threshold is 50 percent of the same height, and the desktop count includes ads in side rails.3 Neither definition counts ads; two large units or ten small ones can give the same percentage.
The Coalition's lists run to sixteen experiences, six for desktop and ten for the phone, built on research in which more than 150,000 consumers took part.4 Three of the sixteen are density thresholds. The other thirteen name a format or a behavior, such as auto-playing video with sound or a large sticky unit.4 Quantity is one thing the standard measures, and it measures quantity as area.
Google's policy uses a ratio, and Chrome enforces the Coalition's list
Google's publisher policies measure something else. Under the heading "Inventory value" the policy states: "We do not allow Google-served ads on screens: with more ads or other paid promotional material than publisher-content."5 That is a ratio of ads to content rather than a share of height, and it is judged screen by screen. Neither test says anything about the reader.
The Coalition's figure matters because a browser enforces it. Chrome first filtered ads on sites in North America and Europe that repeatedly violated the standards. From July 9, 2019, it would "stop showing all ads on sites in any country that repeatedly display these disruptive ads."6 Google's help page on ad experience reviews says a site rated Failing is notified at least 30 calendar days before filtering starts.7 A publisher who passes that review has cleared the only density test a browser enforces, and has learned nothing about the readers.
The research does not agree that the optimum is a number
A field experiment reaches a conclusion close to Cho and Cheon's. Goldfarb and Tucker, in Marketing Science in 2011, found that matching an ad to the page's content and making an ad more obtrusive each raised purchase intent on its own. Together they did worse: ads that were both matched and obtrusive raised purchase intent less than ads that were only one or the other.8 For a publisher whose ads already match the page, adding prominence to them is the combination the study found to fail.
Readers' own reports agree. In the Reuters Institute's 2015 Digital News Report, 29 percent of respondents in the United States said they found traditional banner advertising distracting. Those respondents also said they "will actively avoid sites where they interfere with the content too much," and 47 percent of the American sample had installed ad-blocking software.9 None of the three studies puts "too much" in pixels.
The per-page maximum is the density control the publisher holds
A publisher adds one script, adbubbles.js, and words that already appear in the site's articles get a light dotted underline. Until a reader hovers over or taps one of those words, the underline is all that is on the page. What opens then is the bubble: a small panel with an "Ad" label, a short headline and body, the advertiser's display domain, and one link. The Coalition's method sums the heights of ads within the main content, and here the reader's own action decides when a bubble has any height. The embed reports three things and nothing else: one view per page load, one open when a bubble has been visible for one second, and one click when a link is followed. AdBubbles sets no cookie, stores no identifier, and collects no personal data from readers.
The embed also underlines only the first occurrence of each bought word and places at most one bubble per word. It never places two bubbles in the same paragraph, and it never exceeds the maximum the publisher set for the page. That maximum is the publisher's number, set in the site's configuration beside the scope the embed may scan. AdBubbles matches a word on a page to the advertiser who bought that word for the month. It does not choose which people see a bubble, does not follow a reader between pages or sites, and does not promise a result.
Suppose a 1,400-word article carries nine bought words spread over seven of its fourteen paragraphs, and the publisher has set the page maximum at three. The embed underlines at most three of the nine, no two in the same paragraph, each at its first occurrence. The other six stay as plain text, and a reader who never hovers sees three dotted underlines and no ad. On Cho and Cheon's reading, the number the publisher chose is a decision about how often the reading is interrupted. Here the reading is interrupted at most three times, each time at the reader's own choice.
One paper priced annoying ads at more than they paid
Goldstein, Suri, McAfee, Ekstrand-Abueg, and Diaz, in the Journal of Marketing Research in 2014, set out to price the interruption, and their unit was the impression, the industry's word for one ad shown once. They measured how many fewer impressions participants produced when the ads had been rated annoying, and what a publisher would have to pay readers to make up the difference. In those terms the cost of the bad ads in their experiment came to US$1.53 per thousand impressions. Publishers, they note, are often paid less than US$0.50 per thousand impressions to run such ads. Their conclusion is that "in plausible scenarios, the practice of running annoying ads can cost more money than it earns."10
The optimum, on that arithmetic, sits where the last ad added still earns more than the readers it costs, and that point can lie well inside what a browser permits. A page can stay under the Coalition's thresholds, keep more content than ads on every screen, and never receive a filtering notice, yet still be the page that paper priced. On a page carrying bubbles the maximum per page is the one density setting nobody else sets. A publisher account and the embed cost nothing to set up, and adding a site shows which of its words are for sale before any maximum is chosen. If a cap set too high produces the effect that paper measured, the cost takes the form of readers who do not return, and no dashboard column shows that cost beside the earnings.10