Ezoic stopped loading display advertisements that readers were unlikely to reach on August 4, 2026, and says the change removed about a quarter of the ad units it served while the share meeting the industry's visibility standard climbed to 87 in every 100. The company set out the results in a blog post published on September 17, 2026, exactly five months before Google moves AdSense and Ad Manager to a stricter method of counting impressions.

In Short

Ezoic, a company that puts ads on websites, changed its code so that an ad only loads when a reader is about to scroll to it, and that removed about one in four ads from its pages. Advertisers paid more for each ad that remained, so publishers kept roughly the same income, according to figures the company published itself. Google changes how it counts ad impressions in February 2027, and Ezoic says it made its move early, although the Google rule and the Ezoic change measure different things.

What Ezoic changed on August 4

The post, written by Alyssa Mitzel and published on the Ezoic blog on September 17, 2026, describes a single technical adjustment. Before August 4, display units on Ezoic-served pages loaded when the page itself loaded, regardless of where on the page they sat. After August 4, according to Ezoic, "Ads now render only when they're about to enter the viewport instead of the moment the page loads."

The practical consequence is that a unit positioned far down an article, below the point where a typical reader stops scrolling, is never fetched or painted at all. The company is blunt about what that inventory had been. "Before August 4, a meaningful share of the display impressions we sold never entered a reader's viewport," the post states. "They loaded, sat below the fold, and were never seen by anyone."

Ezoic frames the problem as one of supply rather than buying technique. According to the post, "That's not a creative or bid-strategy problem. It's an inventory problem, and it sits upstream of anything a buyer controls."

The headline figures

Four numbers carry the announcement. Display impressions fell by about 25% "across the board," according to Ezoic. Viewability, measured since the August 4 change, reached 87%. Prebid win CPM - the average price paid on impressions won through the company's Prebid auction - rose by about 54%. A graphic in the post adds a fourth figure: revenue per thousand requests up 41%.

On revenue itself, the company offers no percentage. The post says only that "Revenue held rather than falling with the impression count," and later that "publisher earnings held." Ezoic does not state the length of the comparison window, the baseline period, the number of publishers or pages included, the absolute CPM levels before and after, or which measurement vendor produced the 87% figure.

For readers outside the display trade, viewability is a measure of whether an ad had a realistic chance of being seen. Under the industry standard, a display unit counts as viewable when at least 50% of its pixels sit inside the visible part of the browser for one continuous second, and larger units of 242,500 pixels or more may use a reduced 30% threshold if that is disclosed to buyers. An 87% rate means that, of the impressions Ezoic now serves, roughly 13 in every 100 still do not meet that bar.

Buyers, sellers and three exchanges

The post organises its argument around four groups: advertisers bidding on the inventory, publishers whose pages carry it, the demand partners routing bids into the auction, and Ezoic itself. "Most changes that help one of those groups cost another but this one didn't," the company writes.

For buyers, the claim is that price followed quality. "Advertisers aren't bidding on inventory they'd never get credit for anymore," according to Ezoic. "They're paying for something that's actually there." For publishers, the company argues that the price increase offset the loss of volume: "The CPM lift did the work the volume cut would otherwise have undone."

The post also contains a line aimed at the wider market rather than at Ezoic's own customers: "A viewability gain that doesn't move price is a vanity metric." It is a fair test, and one that the company's own data only partly answers, as the following sections set out.

PubMatic, OpenX and TripleLift

Ezoic names three supply-side platforms among its demand partners and gives viewability changes for each. According to the post, viewability on PubMatic rose by 33 percentage points, on OpenX by 34 points, and on TripleLift by 31 points, landing "into the low 80s" for all three. Revenue per request and win CPM for those partners "moved up directionally alongside it," the company says, without supplying figures.

The partner numbers carry an implication the post does not spell out. If viewability rose by 31 to 34 points and finished in the low 80s, then the starting point for those three exchanges was somewhere around the high 40s to low 50s. On the company's own account, roughly half of the display impressions passing through those pipes before August 4 failed the one-second, 50%-of-pixels test. "We changed what we sent them to bid on, not how they bid," according to Ezoic.

There is also a gap between the two viewability figures. The network-wide number is 87%; the three named partners sit in the low 80s. Ezoic does not explain the difference. Possible explanations include a different mix of placements routed to each exchange, or a different measurement source, but the post does not say which applies.

The TrustedStack comment

The only outside voice in the post comes from Scott Schoenbeck, founder and chief executive of TrustedStack. "We've seen a clear increase in demand for Ezoic inventory following their move to reduce lower-viewability impressions," Schoenbeck said, according to the post. "Spend and eCPMs have both increased, reinforcing what we consistently see from buyers: higher-quality, more viewable inventory attracts stronger demand." The statement confirms direction but gives no numbers for spend or eCPM.

What the numbers do not show

The figures do not reconcile into a single revenue outcome, and the arithmetic is worth doing. If volume fell by 25% and every remaining impression sold for 54% more, revenue would have risen by about 15% (0.75 multiplied by 1.54 gives 1.155). Ezoic does not claim a 15% gain. It claims revenue "held."

The likeliest reason is that the 54% applies to Prebid win CPM only, meaning the subset of impressions won through the company's header bidding auction, rather than to all demand on the page. The post does not say what share of revenue flows through Prebid, or how prices moved on other demand sources.

The 41% rise in revenue per thousand requests offers a second check. If revenue was flat, a 41% increase in revenue per request implies that request volume fell by about 29% (one divided by 1.41 is roughly 0.71). That would be slightly steeper than the 25% impression cut, which is plausible, since units never reached by a reader no longer generate a request at all. Ezoic does not publish request counts, so the calculation remains an inference.

All of the figures are company-reported and unaudited. Ezoic has a record of publishing performance data on its blog, and PPC Land has covered those releases with the same caveat. In July 2026 the company credited a 23.6% quarterly rise in network earnings per thousand visitors to its strategy of extracting more from existing audiences, and days later Mitzel's earlier post reported that AI Ad Optimization produced a weighted 7.7% uplift in EPMV across more than 900 publisher sites between January and July 2026. None of those figures has been independently verified.

Timing matters too. The post was published about six weeks after the change, and Ezoic does not state whether seasonal demand, which typically strengthens in the autumn, was controlled for in the price comparison.

Rendering is not the same as being seen

The post opens by tying Ezoic's change to Google's. "Starting February 17, 2027, Google will only count a display impression once an ad has actually started to render, not the moment it begins loading," according to Ezoic. It then summarises the principle behind Google's rule as: "if nobody could have seen it, it shouldn't count."

That summary goes further than Google does. The date is correct: AdSense emailed publishers on September 1, 2026 to say that display ads which load but never begin painting to the screen will stop counting as impressions from February 17, 2027, with a companion Help Center article extending the same date to Google Ad Manager. On the buying side, Google's Q3 2026 roadmap places the move of billed display impression counting in Campaign Manager 360 and Display & Video 360 onto the same basis in February 2027, with Google expecting only a slight decrease in billed volume.

But begin-to-render is a counting rule, not a viewability rule. It replaces count-on-download, which recorded an impression as soon as the creative started arriving on the device, with a trigger that waits until the creative starts to paint. An advertisement rendered into a slot below the fold, in a background tab, or in a unit the reader scrolls past in under a second still qualifies under begin-to-render. The standard behind it was finalised in October 2017 guidelines from the IAB, the Media Rating Council and the Mobile Marketing Association, and those guidelines say explicitly that rendering does not establish exposure.

Ezoic's August change goes further than Google's rule requires. By deferring rendering until a unit approaches the viewport, the company removes impressions that would have counted under both the old and the new Google methods. That is why its viewability moved so sharply. It also means the claim that "Ezoic got there first" compares two different things: the February 2027 change is a narrower accounting correction, while the August 2026 change is a supply decision.

The latency trade-off

Deferred rendering carries its own risk under the new counting method. Google's Help Center article, as described in PPC Land's September coverage, states that lazy rendering directly affects begin-to-render measurement because it governs when creative assets start painting, and names two Google Publisher Tag parameters, renderMarginPercent and fetchMarginPercent, that move rendering earlier. Raising them increases the load on the network and on the reader's device.

The tension is mechanical. The later an ad renders, the higher its viewability is likely to be, but the greater the chance that a fast-scrolling reader passes the slot before the creative paints, in which case no impression is recorded at all after February 17, 2027. Ezoic does not disclose what rendering margin it uses, or how many units now begin loading but fail to paint before the reader moves on.

Deferring page resources has also produced unintended results elsewhere. In November 2025 an SEO agency reported a 20% traffic decline after applying lazy loading to every image on a client site, because the technique had been applied to the hero image and pushed Largest Contentful Paint from 1.8 to 4.2 seconds. That case concerned images rather than ad slots, but it illustrates why the placement of the rendering threshold matters as much as the decision to defer.

Why this matters for the market

The sharpest relevance of Ezoic's data is not viewability but volume. The programmatic supply chain has spent two years trying to reduce the number of bid requests that never become paid impressions, and a quarter fewer impressions almost certainly means fewer requests sent to exchanges.

Those requests now carry a direct cost. PubMatic, one of the three partners Ezoic names, began charging publishers $0.001 CPM for inventory above allocated daily caps under a supply policy effective April 16, 2026, with enforcement notices arriving in June and the threat of blocked domains for persistent overages. Publishers including Chegg and the Daily Mail had already deployed bid throttling to cut duplicate requests created by header bidding, a problem that PubMatic's own processing of 56 trillion impressions in a single quarter of 2022 illustrated.

The other two partners have worked on the same problem from the exchange side. TripleLift joined Amazon Ads' Dynamic Traffic Engine beta in February 2025, replacing its own predictive models with direct demand signals, and OpenX integrated the same system on December 22, 2025 to route higher-value impressions to Amazon DSP. Amazon then donated the Dynamic Traffic Engine to IAB Tech Lab in April 2026 as a way to address wasted queries per second across the industry.

Those efforts filter requests after the publisher has sent them. Ezoic's change removes them before they exist. Whether that is more efficient depends on how accurately the rendering threshold predicts real scrolling behaviour, which the post does not measure.

The publisher baseline problem

For publishers outside Ezoic, the timing has a specific consequence. PPC Land reported that historical data suitable for estimating the impact of Google's begin-to-render switch begins on August 12, 2026, giving Ad Manager publishers roughly six months of baseline and AdSense publishers considerably less. Google has published no estimate of how far impression counts will fall.

Ezoic's change landed eight days before that baseline window opened. For its publishers, the months running up to February 2027 already reflect a network that renders far fewer units, so the drop they see in February, if any, will be measured against inventory that has already been thinned. That makes their pre- and post-Google comparisons cleaner, but it also means the 25% reduction will not appear in any Google-side comparison that begins on August 12.

Buyers and the price of what is seen

For advertisers, the useful signal is the price response. A 54% increase in Prebid win CPM after removing low-viewability units suggests that buyers were already discounting that inventory heavily, and that at least part of the lower price previously paid for Ezoic supply reflected units nobody saw. Buyers on the three named exchanges now receive fewer requests from Ezoic pages, but on the company's figures, each carries a much higher probability of meeting the standard viewability threshold.

What the post cannot show is whether the price gain lasts once the rest of the market adjusts. If Google's February 2027 change, and other intermediaries following Ezoic's approach, push more viewable supply into the auction, the premium that Ezoic inventory currently commands may narrow. The company argues the reverse, writing that "Advertisers stop paying for ads nobody saw, which builds more durable demand for what's left." That is a prediction, and the post offers six weeks of data to support it.

Timeline

Summary

Who: Ezoic, the publisher monetisation platform, in a September 17, 2026 blog post by Alyssa Mitzel, with a statement from Scott Schoenbeck, founder and chief executive of TrustedStack. The change affects Ezoic publishers, advertisers buying their inventory, and demand partners including PubMatic, OpenX and TripleLift.

What: From August 4, 2026, Ezoic renders display ads only when they are about to enter the viewport. According to the company, this cut display impressions by about 25%, raised viewability to 87%, lifted Prebid win CPM by about 54% and revenue per thousand requests by 41%, while publisher revenue held. Viewability on the three named exchanges rose 31 to 34 percentage points into the low 80s. All figures are company-reported and unaudited.

When: The change took effect on August 4, 2026; the results were published on September 17, 2026. Google's begin-to-render counting change for AdSense and Ad Manager takes effect on February 17, 2027.

Where: Across display inventory on sites monetised through Ezoic, and across the header bidding auctions and exchanges that carry it.

Why: Ezoic argues that removing ads nobody sees raises prices enough to protect publisher revenue, and positions the change as preparation for Google's February 2027 counting rule. That rule measures rendering rather than viewability, so Ezoic's change is broader than what Google requires.