On 17 February 2027, Google Ad Manager and AdSense change how they count a display impression.
The current method is count-on-download. An impression is recorded when the ad server responds to a request and the creative starts downloading to the device. The new method is begin-to-render: the impression is recorded only once the creative has actually started rendering on the page.
Publishers will see their display impression totals go down. The number of advertisements human beings see will not change at all.
What the gap between those two numbers was
Every impression that disappears is an ad that began downloading and never rendered — the user navigated away, the connection dropped, the tab closed, the creative failed, the page was destroyed before the pixel painted.
Under count-on-download, all of those were impressions. They were reported to publishers as delivered inventory and to advertisers as delivered advertising. Nobody saw them.
That is not a small technical nicety. It is the definition of the unit that the entire open-web display market prices in, and for as long as the method has been in use, some share of what changed hands was inventory that did not exist in the only sense that matters — a person and a screen.
So the change is right
Worth saying without hedging, because the framing in the trade coverage — publishers may see impressions fall — invites the wrong reaction.
Begin-to-render is the more honest measure. It moves the count closer to the event anybody actually cares about, it aligns with where the measurement standards bodies have been pointing for years, and an advertiser paying for an ad that rendered is getting what they think they are buying. A publisher whose count drops has not lost audience or revenue-generating capacity. They have stopped being credited for a failure.
What it does to publisher economics
Less than the headline implies, if pricing adjusts, and considerably more if it does not.
If advertiser demand is unchanged and the supply of counted impressions falls, the price per counted impression should rise by roughly the same proportion. Same money, fewer units, higher CPM, flat revenue. That is what should happen in a market that is clearing.
The problems are in the interim and in the contracts. Any publisher who sells guaranteed impression volume has commitments denominated in a unit that is about to shrink, and they will have to deliver more real impressions to satisfy the same number. Forecasts, rate cards, year-on-year comparisons and every internal target built on the old count break simultaneously in February.
Google's advice — run a historical report from after 12 August and compare the two figures — is genuinely the right preparation, and the fact that it must be done publisher by publisher, with no universal percentage available, tells you the variance between sites is large. A publisher with heavy pages, slow connections and aggressive lazy loading will lose far more than a fast, simple one. The change quietly rewards page performance, which is a defensible outcome nobody is describing as a goal.
The part that belongs with this week's other Google story
On Wednesday a federal judge declined to make Google divest its ad exchange and imposed behavioural remedies instead — bid transparency, no unified pricing rules, an end to first and last look. This paper wrote at the time that a behavioural remedy is a rule somebody has to keep watching, and that the watching has to happen inside a real-time auction understood in full by exactly one party.
Three days later, that party changed the definition of the unit the auction transacts in.
The two things are not connected and there is no reason to think they are. That is the point. This is an ordinary product decision, made for good reasons, that materially alters what every publisher in the market is paid for — and it demonstrates the enforceability problem more clearly than any adversarial act would. A remedy can prohibit self-preferencing in an auction. It cannot easily reach a change to what is being counted, because the change is technically correct and would be defensible in any forum.
The buy side has already drawn the obvious conclusion by replacing one number with several, which is what you do when you no longer trust a single reported figure from anybody's system.
What to watch
Not the impression decline, which will be reported loudly and means little on its own.
Watch open-web display CPMs across the February–April window. If they rise roughly in proportion to the fall in counted impressions, the market repriced and the change was a clean accounting correction. If impressions fall and CPMs do not move, then a share of publisher revenue has quietly gone away — and it will turn out to have been revenue for advertising that nobody ever saw, which is a fair outcome arriving abruptly at people who had built businesses on the old count.
The switch from count-on-download to begin-to-render impression counting in Google Ad Manager and AdSense effective 17 February 2027; the definitions of both methods; the application to display banner inventory across web, mobile web and connected TV; Google's expectation that publishers may see total display impressions fall while declining to forecast a universal percentage; the dependence of the size of the change on creative performance, page behaviour and delivery; and the recommendation that publishers compare existing and begin-to-render impressions using a historical report starting after 12 August 2026 are as reported by Storyboard18, PPC Land, SMBtech and Google's own Ad Manager documentation in September 2026. The ad tech remedies ruling referred to here is as reported on 2 September 2026 and covered in this publication the same day. The analysis is our own.
Topics mediaadvertisingpublishersmeasurementgoogle
