Few beliefs about a dashboard are held as firmly as this one: a click rate that was three percent one week and five percent the next has recorded something that happened on the site. On a site whose bubbles open a few hundred times a week, the belief is usually mistaken. A rate computed from a few hundred events moves by chance more than by cause, and the statistics of a proportion put a figure on how much. The counts underneath the rate, views, opens, and clicks, raise a separate question, because the standards that define an impression were written for a format that a bubble is not.
An impression is counted late, filtered of robots, and still not necessarily seen
The IAB's 2004 measurement guidelines define an ad impression as a delivery system's response to a browser's request, "filtered from robotic activity" and recorded "as late as possible" in delivery of the creative.1 Google Ads defines click-through rate as clicks divided by impressions, so that 5 clicks on 100 impressions is a CTR of 5 percent.2 Google's page says the ratio shows how often people who see the ad click it, and "see" claims more than the count establishes.
The Media Rating Council gave that word a threshold in 2015. A display impression is viewable when at least 50 percent of its pixels have been on an in-focus browser tab in the viewable space of the page for one continuous second.3 In the guideline's words, an "opportunity to see" exists with a viewable impression, "which may or may not be the case with a served ad impression."3
Invalid traffic is removed before a rate is computed
The Council's invalid traffic standard of June 2020 sorts the counts to be removed into two kinds. General Invalid Traffic, or GIVT, is identified "through application of lists or with other standardized parameter checks," and its examples include data-center traffic, crawlers, and non-browser user-agent headers.4 Sophisticated Invalid Traffic, which includes hijacked devices and falsified measurement events, needs "advanced analytics, multi-point corroboration/coordination, significant human intervention."4 The standard requires that "Known GIVT must be removed from monetized counts and metrics." A rate whose denominator still contains crawlers is low for a reason that has nothing to do with readers.
A view is a page view, and an open is a viewable impression's counterpart
The embed, adbubbles.js, sends three kinds of beacon and nothing else, each a small asynchronous HTTP POST of the kind browsers provide for analytics.5 One view is sent per page load, one open when a bubble has been visible for one second, and one click when a reader follows the link inside a bubble. A view is a page view on which the embed ran, and at that moment no ad has been shown. The dotted underline is on the page, but the bubble is not.
An open is the counterpart of the viewable impression, since both require one continuous second of visibility. What differs is how that second begins. A bubble becomes visible only after a reader's hover, tap, or Enter key, and a display ad whenever the viewable space of the page includes it, with no action by the reader. Two rates follow from the sequence: opens divided by views, and clicks divided by opens, which is the nearest thing to the CTR in the title. The advertiser dashboard shows an open rate and a click rate, and a publisher computes rates from the publisher dashboard's counts.
The match is between a word and the advertiser who bought it for the month. No reader is chosen to see the ad or followed between pages or sites, and no result is promised to anyone. AdBubbles sets no cookie, stores no identifier, and collects no personal data from readers. A publisher account and the embed cost nothing to set up, and a verified site can request a crawl that lists its currently sellable words. The publisher pages and the documentation describe the steps and the scope setting.
Earnings accrued is a sum, and a yield figure is arithmetic the publisher does
Yield, in the trade's usage, is a rate per thousand. AdSense's help page defines page RPM as estimated earnings divided by page views, multiplied by 1,000, and adds that RPM "doesn't represent how much you've actually earned."6 The AdBubbles dashboard shows earnings accrued per site per day, a sum rather than a rate, and AdBubbles publishes no RPM. The pricing page gives the share instead: publishers keep 70 percent of what advertisers pay for the campaigns that run on their pages.
A publisher who divides accrued earnings by views has a yield figure of their own, and it moves for a reason the AdSense formula does not contain. Each campaign's payment is shared among the sites it ran on in proportion to their bubble opens, accrued daily. A site's earnings from a word therefore depend on its opens relative to every other site the same campaign ran on that month.
A few hundred events give a rate a margin wider than its weekly movement
Lawrence Brown, Tony Cai, and Anirban DasGupta, writing in Statistical Science in 2001, give the standard interval for a proportion.7 It is the observed proportion, plus or minus a multiplier times the square root of the proportion times one minus the proportion, divided by the square root of the count. For a 95 percent interval the multiplier is 1.96, the value the NIST handbook tabulates at 0.975.8 The same authors found that interval's coverage erratic and recommend the Wilson, Jeffreys, or Agresti-Coull interval; the standard one is used below because it can be checked by hand.
Suppose a site's bubbles opened 400 times in a week and 12 of those opens ended in a click. The click rate is 3 percent, and the formula gives an interval from 1.3 to 4.7 percent. Suppose the next week brings 400 opens and 20 clicks, a rate of 5 percent, with an interval from 2.9 to 7.1 percent. A single underlying rate anywhere between 2.9 and 4.7 percent is consistent with both weeks. Suppose instead the twelve weeks of a quarter, 4,800 opens and 192 clicks: the rate is 4 percent and the interval is 3.4 to 4.6 percent. At a fixed proportion, twelve times the count narrows the interval by a factor of about three and a half, the square root of twelve.
Weeks of counts are the unit in which a change can be read
The controls a publisher holds are the scope the embed reads, the maximum bubbles per page, and the advertiser categories excluded. Each changes the counts. The rates report the change only after enough opens have accumulated on both sides of it. A change is visible once the intervals from before and after it no longer share any values. On a site with a few hundred opens a week, that takes weeks on each side, and for a small difference it takes months.
The industry's definitions are changing in the meantime: in November 2025 the IAB and the Media Rating Council published attention measurement guidelines. They separate viewability, "the opportunity to see the ad," from attention, "the measure of whether the ad was seen or heard by a person and to what depth."9 An open records a reader's own action and one second of visibility, and how the guidelines would classify that is for their readers to judge. A rate built on a finer event is still a rate, and on a small site it needs the same weeks before a movement in it means anything.