Does an ad do better when the page it sits on is about the ad's own subject, and if it does, on which measure? A matched ad raised purchase intent in one study and attitude in another, and one experiment found the unmatched ad was noticed more. The usual claim for such placement is that it converts better, and what the studies found is narrower.
Intent was first read from the query, and a page's words carry it too
Andrei Broder's 2002 paper in SIGIR Forum sorted web searches into three intents. A navigational query means "the immediate intent is to reach a particular site." An informational query means "the intent is to acquire some information assumed to be present on one or more web pages." A transactional query means "the intent is to perform some web-mediated activity."1 In the query log, the words typed were the only evidence of which intent applied.
Five years later Broder and three colleagues separated the two kinds of ad. Sponsored search places ads on result pages "with ads driven by the originating query." Contextual advertising is "the placement of commercial textual advertisements within the content of a generic web page."2 For a content ad, the words that drive it belong to the page. Kaifu Zhang and Zsolt Katona, in Marketing Science in 2012, call this "the potential that consumers' content preferences are indicative of their product preferences."3
A matched ad raised purchase intent, and prominence took the gain back
A 2011 field experiment by Avi Goldfarb and Catherine Tucker, in Marketing Science, found that "matching an ad to website content and increasing an ad's obtrusiveness independently increase purchase intent." Ads that did both "do worse at increasing purchase intent than ads that do only one or the other."4 For an advertiser buying a word, the finding has a half to keep and a half to avoid. The matched half is the word itself. Buying a word is the most literal form of matching, and it claims only that the phrase appears in the page's text, not that the whole page is about it. The obtrusive half is what a bubble does without: the bought word carries a light dotted underline, and the bubble opens only when the reader hovers over it or taps it.
Matching changed what readers thought of an ad more than whether they saw it
Robert Moore, Claire Stammerjohan, and Robin Coulter ran two experiments for the Journal of Advertising in 2005, with 195 and 90 participants. In the first, "incongruity has a more favorable effect on recall and recognition, whereas congruity has more favorable effects on attitudes." In the second, once ads drew enough attention to be recalled, "moderate congruity offers the most favorable attitudes toward the ad."5 The unmatched ad was noticed more, and the matched ad was liked more.
Guillaume Hervet and three colleagues, in Applied Cognitive Psychology in 2011, tracked eyes and reported that most participants fixated the text ads at least once. Matching made no difference to how long an ad was looked at, but matched ads were remembered better than unmatched ones.6 The two studies disagree on memory, and neither found that a match got an ad looked at more. Only Moore's group measured attitude, and it found the matched ad was liked more. The advertiser's problem is therefore getting the ad seen, and an in-text bubble answers that with the word itself. AdBubbles sets no cookie, stores no identifier, and collects no personal data from readers.
A specific ad works once the reader's preferences have narrowed
The retargeting literature answers which word to buy. Anja Lambrecht and Catherine Tucker, in the Journal of Marketing Research in 2013, compared ads showing products a consumer had viewed against generic brand ads. The data came from an online travel firm's field experiment. The specific ads were "on average, less effective than their generic equivalents." For consumers whose browsing, such as visiting review sites, suggested their preferences had evolved, the specific ads "no longer underperform."7 The authors' explanation is that "only when they have narrowly construed preferences do they respond positively to ads that display detailed product information."7
An article that uses a specific phrase is detailed information of the kind that explanation describes, and a reader partway through it chose to read it. Suppose a business that installs tankless water heaters is choosing between the keywords "water heater" and "tankless water heater." The first would appear on pages about repairs, rentals, and tank models. The second would appear on pages that have already narrowed to the product the business sells. Which phrase produces more clicks on a set of sites is what the campaign's own counts show, per keyword, per site, and per day.
The looseness of a match is a setting, and a bought word has none
On one large search platform a keyword has three match types. Under broad match, "ads may show on searches that are related to your keyword, which can include searches that don't contain the direct meaning of your keywords." Phrase match covers searches that "include the meaning of your keyword," and exact match covers searches with "the same meaning or same intent as the keyword." Google Ads Help adds that "broad match is the default match type that all your keywords are assigned."8 An advertiser who buys a word there buys an interpretation of it, and the widest interpretation applies unless the advertiser picks another.
Zhang and Katona's model says why an intermediary might prefer a loose default. The intermediary in their model "has incentives to strategically design its targeting technology, sometimes at the cost of the advertisers." When competition among advertisers is high, it "lowers the targeting accuracy such that the consumers see less relevant ads," which "encourages advertisers to bid for multiple content topics."3 That is a result about a model, not a finding about any named company.
A word bought on AdBubbles has no match type. The embed underlines the first occurrence of the phrase in a page's text. One advertiser holds a keyword on a site for a calendar month, and other advertisers see the word as taken. We match words on pages to the advertiser who bought them. We do not choose which people see an ad, we do not follow a reader between pages or sites, and we do not promise a result.
An advertiser can build a campaign and see, per keyword, how many sites carry the word and its estimated monthly page views, before paying anything. The advertisers page describes the steps. Both the underline and the bubble are shown on the how it works page.
The direction of the field, in its own documents, is away from the typed word and toward inferred meaning. One platform's default match type reaches searches that need not contain the keyword's direct meaning.8 Broder's 2007 paper argued for matching pages by a classification of meaning as well as by terms.2 Lambrecht and Tucker advised varying an ad with the stage of the consumer's decision.7 Each of those adds an inference about intent to the signal, and the signal itself is still a word in a sentence that a reader chose to read. That word is the one part of the arrangement that can be bought without inferring anything.