Six working professionals with substantial web experience ran 24 short searches for Jan Panero Benway and David Lane of Rice University in the late 1990s. Some of the searches required clicking a red banner that contained a shortcut to the goal. The participants found those banners about 58 percent of the time, against 94 percent for ordinary control items. Benway and Lane wrote that they had "dubbed this phenomenon 'banner blindness'" and set out to reproduce it under controlled conditions.1
Their second experiment used 72 undergraduates and varied the banners. Some held information useful to the search, some were plain blue link text, and some were animated. Only 17 of 71 participants reported seeing the helpful banners, and neither grouping nor animation changed that.1 Benway's conference abstract put it briefly: "Users saw banners hardly at all when clicking a banner was not required to accomplish a task."2
The effect has been measured since the 1990s
Nielsen Norman Group defines banner blindness as "people's tendency to ignore page elements that they perceive (correctly or incorrectly) to be ads." The group first documented it in usability tests in 1997 and replicated it with eye tracking in 2007. A further eye-tracking study, reported in April 2018, found the behavior still common. Kara Pernice, its author, summarized the desktop finding: "Users read the text on the page, but they looked very little (if at all) at the ads."3
Two further observations matter to a publisher. Content placed where ads usually sit "is often considered to be an ad and is also ignored." Pernice adds that "anything that stands out from immediate surrounding context is likely to be considered an ad." Position and appearance, on that evidence, are the cues readers act on.3
Attention goes where the reader's task goes
Benway and Lane offered two explanations for what they saw. The first was that their participants had "learned to ignore advertisements when searching for information on the Web." The second was that "since searchers tend to look for linked text, they focus on small, blue text objects." Both point to the task, which decides what gets looked at, and a searcher's task is finding the next link.1
Magnus Pagendarm and Heike Schaumburg tested that reading. They gave 32 students one website; half searched for something specific, and half browsed with no goal. The browsing group later recalled and recognized significantly more of the advertising. Navigation style, the authors proposed, explained the conflicting earlier results.4
Text ads are missed as well
The finding is not confined to graphics. Owens, Chaparro, and Palmer asked in the Journal of Usability Studies in May 2011 whether text advertising blindness was the new banner blindness. It exists, they answered. It lowers search performance on a page. Users "tend to miss information in text ads on the right side of the page more often than in text ads at the top."5
Guillaume Hervet and three colleagues studied the question with an eye tracker. Thirty-two participants read eight pages about choosing a digital camera, each with a Google-style text ad of 180 by 150 pixels on the right. The abstract reports that most participants fixated the ads at least once. The British Psychological Society's digest of the study gives the counts. Eighty-two percent of the participants looked at one or more of the ads, and of the 128 ad exposures, 37 percent were looked at once or more. Ads whose subject matched the page were remembered better than ads whose subject did not.67
A 2023 eye-movement study in Cognitive Processing reached a related conclusion. Text-type banners had "a lower blindness rate" than picture-type banners, and the authors recommended "setting a consistent theme between the banner ad and the web content."8 Plain text and a matching subject each helped. In every study above that reports placement, however, the ad sat in a region of its own, beside or above the text.
A word in the sentence sits inside the reader's attention
AdBubbles sells the words that already appear in a publisher's articles. Its role ends there. We match a word on a page to the advertiser who bought that word for the month. We do not choose which people see a bubble, do not follow a reader between pages or sites, and do not promise a result. AdBubbles sets no cookie, stores no identifier, and collects no personal data from readers.
A publisher adds one script, adbubbles.js, and the embed scans only the scope the publisher configured, skipping headings, links, code, and navigation. The first occurrence of a bought word in the article body gets a light dotted underline. Nothing more appears until the reader hovers over the word, taps it on a phone, or presses Enter on it. Then a small bubble opens with an "Ad" label, a short headline and body, the advertiser's domain, and one link marked as sponsored.
The difference from the studies above is where the reader is looking when the ad appears. A banner requires the reader to look away from the text. Twenty-five years of measurement show readers declining to do so. A bubble opens from a word in the sentence being read, on the reader's own action, and closes on mouse-out or a tap outside it. The literature's lesson is that attention stays with the task, and here the task is that sentence.3
A field experiment by Avi Goldfarb and Catherine Tucker in Marketing Science bears on the same choice. Matching an ad to the page's content raised purchase intent, and so did making the ad more obtrusive. Doing both was worse: such ads "do worse at increasing purchase intent than ads that do only one or the other." The authors saw in that a "growing bifurcation in Internet advertising between highly targeted plain text ads and more visually striking but less targeted ads." A bubble is matched to the page by construction, since the word is the page's own, and stays closed until opened.9
The research does not measure a bubble
The first limitation is that no study above tested a bubble that opens from an underlined word on the reader's own action. Benway's banners sat apart from the menus, Hervet's text ads sat in a rectangle on the right, and Pernice's promotion looked unlike the text.173 Nothing here reports what share of readers open a bubble. We have published no such figure.
The second limitation concerns what "attention" measures. A fixation is a glance; Hervet's participants who glanced at an ad were not, on that evidence, persuaded by it.6 Goldfarb and Tucker measured stated purchase intent rather than purchases.9 Whether readers on a given site act on a bubble is a question for that site's dashboard, which counts views, opens, and clicks.
Setting up a publisher account and the embed costs nothing, and a site can be verified and crawled before any bubble appears. The crawl shows which words on the site are currently sellable, a fact no study above supplies. Adding a site is the first step, the installation notes cover the choice of scope, and how it works sets out the privacy position. What the crawl cannot show, and what no published study shows either, is how many readers will open what is sold.