Google will stop counting display ad impressions that never render in February 2027

On February 17, 2027, a display ad served through Google's AdSense or Ad Manager will only count as an impression once it has loaded and started rendering on screen. Today, the impression is recorded as soon as the creative begins downloading. Under the new rule, a slow-loading ad that has not rendered by the time the user navigates away will not be counted.

Eytan Sayag laid out the change in a MarTech360 guest column published September 28. He warned that reported impressions are likely to fall, and that sites not technically ready could see ad revenue drop with them. Google also expects reported display impressions to decline. That suggests revenue operations teams that forecast from impression volume could face a planning problem, even though the update looks like a small change to how impressions are counted.

Count-on-Download versus Begin-to-Render

For years Google has counted display ads using Count-on-Download, or COD. The impression is recorded when the creative begins downloading to the device, whether or not it ever appears. Under Begin-to-Render, or BTR, the ad has to finish loading and start drawing on screen first.

Sayag's example is a reader who taps the wrong link and hits back within two seconds. The page fired an ad request and the creative started downloading. Nobody saw anything, and COD logged an impression anyway. That is how download-based counting has hidden latency problems for years.

Publishers do not have to carry out any manual technical integration for the change itself. It covers display banner inventory on desktop web, mobile web and CTV.

The seconds that go missing

Under BTR, impressions that never render will drop out of reports. The number that matters is how big that gap is on a given site, and what's causing it.

A page can post a healthy impression count while still losing inventory between request and render. He writes that latency can come from several places: heavy ad scripts, long chains of bidders, wrappers that aren't needed, and lazy loading that is badly configured. The basic complexity of the page adds to it as well. After February, each of these could cost counted impressions.

Reader behavior compounds it. A creative that takes several seconds to appear on a page people leave quickly won't earn what the same ad earns when it shows up while they're still reading.

Count-on-Download let latency hide inside healthy impression counts, and on February 17, 2027 that cover goes away.

There's a fair argument that the drop matters less than it looks. Sayag himself notes that an impression that downloaded but never rendered was always worth little to the buyer, so BTR mostly removes inventory advertisers weren't getting value from. The risk falls on publishers whose revenue planning leans on raw impression totals, and on sites where the gap turns out to be wide.

Comparison metrics in Ad Manager

Sayag's suggested first step is to compare standard impressions against the new ad server BTR impressions metric. The gap between the two shows how much inventory is currently being counted before it renders, which gives a baseline. He also recommends tracing the path from ad request to creative render and finding the unnecessary wrappers, heavy buyer scripts and bidder dependencies that slow it down.

A similar split between larger and smaller publishers could emerge this time. For publishers with ad tech staff, the BTR comparison could become the starting point for a latency audit. For leaner operations, running the comparison before February 17 could at least establish how large the gap is before the new count takes effect.