Editor’s Pick Revenue Editors Picks

Google’s ad-count reset could lift CPM without lifting revenue

Ivan Massow 21 Oct 2026 · 14 min read

A red-shirted man holds a huge AD placard beside a blue-clad reader, forcing her to lean around it with her newspaper. Caption: A better reading experience. Signed Noah.

Cartoon · AI-generated

Noah gives a 77% probability that, on 17 February 2027, Google’s operative documentation confirms begin-to-render counting for display banners across web, mobile web and connected TV in both AdSense and Google Ad Manager. That is a forecast of the specified platform state on that date. It is not a forecast that every publisher’s income falls or rises. The commercial task is to prepare for a change in what gets counted.

A publisher can report fewer advertising impressions without losing a single reader. It can report a higher price per thousand without receiving another pound. That combination is about to become a practical management problem rather than a classroom exercise in denominators.

Google says Ad Manager and AdSense will begin changing display-banner counting to begin-to-render on 17 February 2027, covering desktop, mobile web and CTV banners. The count moves from the start of downloading to the creative loading and beginning to render. Google warns that reported impressions may fall.

The commercial question is what each business will make of that change. A finance team needs a comparable baseline. A salesperson needs to know which units a campaign promises. An ad-operations team needs to distinguish a creative that did not render from a reader who never arrived. An editor should not be told that journalism has suddenly become less attractive because a supplier changed the point at which an advertisement enters a report.

Advertisement · Noah

The preparation is already visible in Britain. In a 6 October article, Content Ignite’s James Hanslip says its team is reviewing configurations, lazy loading and refresh across its publisher network. That is a supplier’s account of preparation, rather than a published measurement of the eventual effect. Its value is that it connects a platform announcement to work a publisher can assign now.

Five forecasts to watch

These are Noah’s modelled probabilities as of 21 October 2026. The first concerns the operative platform state on a particular day. The other four include qualifying public evidence by their deadlines. They are not the chances that any given publisher will suffer a loss: an outcome can occur privately while its publication-gated forecast settles NO. Exact conditions, ranges and calculation references are retained in the claim audit.

77% — On 17 February 2027, Google’s operative AdSense and Ad Manager documentation confirms begin-to-render counting for display banners across web, mobile web and connected TV. A postponement, narrower rollout or continued count-on-download on a specified platform prevents YES. The published plan and existing measurement infrastructure support the call; day-specific scope and rollout execution remain the countercase.

30% — By 30 June 2027, a qualifying public case reports that a UK news or specialist publisher’s counted display impressions per 1,000 UK sessions fell at least 10% under begin-to-render versus count-on-download, while session totals differed by no more than 5%, using simultaneous counting or matched periods. A later count makes a gap plausible; its size and the required public comparison remain unobserved here.

30% — By 31 December 2027, a qualifying public UK news or specialist publisher case reports display CPM rising at least 10% after begin-to-render while net display-ad revenue per 1,000 UK sessions changed by less than 5% in either direction against a comparable pre-transition period. This is the potential CPM mirage: a stronger-looking rate without much change in session-normalised income. Gross revenue, unmatched periods or a CPM rise alone cannot settle the target.

20% — By 31 December 2027, a qualifying public UK display-campaign case shows at least a 25% relative reduction in the absolute billed-impression discrepancy rate after both sides align on begin-to-render, using both parties’ records for the same campaign or a matched predecessor. That means a relative reduction, not 25 percentage points. Compatible counting could help, but it cannot eliminate every source of disagreement.

17% — By 30 September 2027, a qualifying public technical study of at least 10 UK news or specialist publisher domains finds that at least 5% of downloaded display ads fail to begin rendering before the slot leaves the viewport or the session ends, with creative or multi-partner delivery latency identified as the primary cause. This is not an observed 5% failure rate. The technical mechanism is credible; a study meeting every sampling, timing and attribution condition is a much tougher event.

The common implication is practical. Prepare for the platform change, then measure the business consequence. None of the four public-case probabilities supplies an industry-wide revenue haircut or a ready-made loss assumption for a particular title.

A later count is a different promise

Advertising reports contain several events that are easy to collapse into a single word. A page can request an advertisement. An auction can select one. A creative can arrive. It can begin to render. It can have an opportunity to be seen. A person can notice it, remember it or act on it. Each event answers a different commercial question.

Moving a count further along that sequence can remove units which previously appeared in a report. It does not establish that the same number of people stopped reading, that an advertiser stopped bidding or that the publication lost the same proportion of cash. Those may be real problems, but each needs its own evidence.

There is a constructive interpretation. If buyers and sellers recognise delivery at more comparable stages, they may spend less time disputing whose number is correct. A campaign sold against a rendering-based requirement can be easier to reconcile with a rendering-based publisher report. The gain would be a clearer transaction and less operational friction; it would not require the publisher to discover a new audience.

The countercase is equally practical. A publisher can have commitments whose delivery assumptions were built around the old measure. A later counting point can expose a gap between the volume its planning system expects and the volume its contract recognises. That is a reason to examine actual agreements, not assume that every existing deal suddenly becomes unprofitable or invalid.

The CPM can improve while the business stands still

Effective CPM relates revenue to a quantity of impressions. Hold the revenue constant and reduce that quantity, and the calculated rate rises. That is arithmetic, not evidence that an advertiser increased a bid or a commercial team improved its pricing.

The same trap works in reverse. A dashboard can make a fall in counted impressions look like a loss of publishing scale when the underlying reader behaviour has not changed. If the team responds by commissioning more articles or adding more advertising positions, it could solve the wrong problem and make the reading experience worse.

I would want a bridge between the two measures before accepting either interpretation. It should show which inventory changed, which counting convention applied and what happened to net revenue, reader sessions and campaign delivery over the same period. A single company-wide percentage can conceal differences between formats and sales channels.

Revenue per reader session can provide a useful second view, provided sessions are defined consistently. It is not immune to changes in audience mix, consent, traffic quality or commercial terms. It simply helps prevent an impression denominator from carrying the whole explanation of performance.

The important question for a publisher is whether the business earns more from a useful reader relationship after costs and commercial adjustments. A more flattering rate on a smaller count may accompany that result, but cannot demonstrate it by itself.

Measure the gap before it becomes a surprise

Google’s current preparation advice includes comparing Ad server impressions with Ad server begin to render impressions, using a historical start after 12 August 2026. Additional Ad Exchange, yield-group and total BTR metrics are marked as coming soon. The timing and available channel coverage matter to any comparison.

A publisher should preserve the date, inventory definitions and scope of its extract. It should also retain the configuration changes that might explain a movement: a new template, an altered loading rule or a different creative supplier. The exercise should make future comparison possible, rather than produce a comforting one-off number.

That means comparing like with like. A direct-sales slice and an entire programmatic network can differ in more than counting method. Desktop and mobile readers can encounter different layouts and loading conditions. Seasonal campaigns can change the creative mix. A useful baseline acknowledges those differences instead of presenting them as a universal industry haircut.

No UK publisher-specific revenue-loss percentage is established by the sources checked for this report. The documentation explains the method and offers a way to examine it. It does not authorise taking one supplier’s expectation and applying the same number to every British news business.

A request-count reset has already shown the problem

Noah’s retained evidence pointed to an earlier change with a different unit. Google’s 23 June 2026 Multiplex notice says each ad in a grid would be reported as its own request, replacing the single-request treatment of the grid. It warned that request-related rates would shift while significant changes in total impressions and revenue were unlikely. This was an announcement about requests, not February’s new impression-counting point.

The connection is useful because a publisher can face several reporting breaks in the same financial history. Combining them into one undifferentiated advertising trend would obscure what changed. A metric dictionary needs an effective date and a countable event, as well as a familiar label. More requests need not mean more advertising sold; fewer impressions need not mean fewer readers.

Rendered does not mean noticed

Begin-to-render is not a synonym for viewability or attention. Google’s Active View explanation describes the usual display threshold as at least half the ad’s pixels visible continuously for one second, with a different pixel threshold for very large creatives. Even meeting an opportunity-to-see standard does not prove that a person paid attention or changed their mind.

This distinction matters in sales conversations. A cleaner delivery measure can be valuable without becoming a promise of human attention. Publishers should be able to explain the journalism, context, audience and creative opportunity they offer while remaining precise about what the measurement can establish.

It also matters when evaluating technology. A vendor could improve the number of ads that begin to render without improving the number that readers notice. A creative could render correctly but provide a poor experience. Those are different outcomes, and a publisher needs more than one measure to choose between them.

The standards discussion predates this particular announcement. The IAB document linked by Google treats rendered impressions as an input into viewability measurement, rather than an equivalent result. The current platform transition is therefore an implementation issue with commercial consequences, not the invention of a new definition of attention.

Do not make the page worse to rescue a count

The tempting response to a later count is to make more advertising load earlier. That can help a particular placement reach its required event before a reader moves on. It can also demand more from a device that is trying to display the article.

Google’s Publisher Tag guidance retains a distinction between important placements near the top of the page and positions further down. Its lazy-loading documentation separates fetching from rendering; under Single Request Architecture, reaching the fetch margin for one slot can trigger the request for the other slots too. One setting therefore should not be mistaken for evidence that every requested ad has rendered.

The sensible unit of work is a controlled change to a particular problem, with a record of both commercial and reader outcomes. A team might investigate whether a slow creative, a delivery path or a template is responsible. It should be able to say what it changed and what else moved, rather than present earlier loading as an automatic improvement.

This is a useful meeting point for product, engineering and commercial teams. Ad operations can identify delivery failures; product can explain the effect on the reading experience; finance can test whether the apparent improvement reaches revenue. None should be asked to optimise a single dashboard number in isolation.

The countercase to intervention deserves to remain open. A well-performing site may have little to repair. Broad changes made in anticipation of a problem can consume time and introduce new ones. Establishing the baseline is valuable partly because it tells a publisher where restraint is justified.

Billing has its own calendar

There is a buyer-side distinction too. Google’s Campaign Manager 360 capability notice labels February 2027 as the estimated move to begin-to-render for billed display impressions in Campaign Manager 360 and Display & Video 360. It warns that billed volume may decrease slightly. That is a separate billing statement; its month-level estimate should not be silently replaced with the publisher-side date of 17 February.

This matters when assessing the promise of easier reconciliation. Two systems can describe themselves as moving towards the same standard while their operative dates, covered inventory and contractual use of the number still differ. A publisher should ask which report the buyer is using and which count becomes an invoice. A measurement label alone cannot answer either question.

The practical gain comes when the parties establish compatible records and understand residual differences. The test is a reduction in the discrepancy for a defined campaign, with both sides’ records, rather than an assertion that shared terminology has eliminated every disagreement.

Fewer ads can be a real improvement

There is a useful counterpoint in Ezoic’s 17 September account of a change implemented on 4 August. The company reports that rendering display ads near the viewport removed about 25% of impressions, with viewability reaching 87% and Prebid win CPM rising about 54%, while publisher earnings held. These are supplier-reported figures, not an independently controlled UK publisher result.

The distinction between Prebid win CPM and the publisher’s overall return matters. A price measured on one auction subset cannot automatically explain all income on a page. Ezoic’s account also does not provide the UK-session-normalised net-revenue comparison required by Noah’s 30% CPM forecast. It is a relevant mechanism and a question to investigate, rather than a completed instance of that target.

Nor should a stable revenue total make the change worthless. Less advertising with a better measured opportunity to be seen could improve the product for readers and buyers. The lesson is to test both the real quality gain and the financial result. Ezoic’s near-viewport loading decision is also different from Google’s minimum rendering-count rule: an ad can begin rendering before it enters the visible viewport.

Britain’s shared infrastructure changes who does the work

Content Ignite’s August partnership announcement with Ozone describes bringing premium demand through Fusion to independent and specialist publishers. It names The Observer, Global, Haymarket and The Weather Channel among Content Ignite’s partners. These are stated commercial relationships, not evidence of each site’s exposure to the February counting change.

Ozone’s founding joint venture brought together News UK, Guardian Media Group, Telegraph Media Group and Reach. The relevance is the infrastructure connecting large media businesses, technology suppliers and smaller publishers. A technical migration need not be solved independently inside every newsroom, but responsibilities still need to be clear.

A publisher using a managed service should know who supplies its comparison, what the service can see and which parts of its business lie outside that view. A platform may observe the auctions and placements it manages while the publisher’s finance team holds the contract, invoice and makegood information needed to assess the result.

Shared expertise can spread practical preparation. Shared dashboards can also spread the same blind spot if everyone treats a definition change as market growth or decline. The governance task is to agree who owns the explanation, not merely who can export the data.

There is an opportunity here for suppliers to earn trust through specificity. A dated comparison, a clear channel limitation and a measured account of what did not change are more useful than a general promise to protect yield. Publishers should ask for that precision while recognising that a supplier cannot manufacture a matched history which was never recorded.

The sales conversation starts before the switch

Guaranteed campaigns that span the transition deserve attention before they become a delivery argument. Commercial teams should establish which measurement their agreements use and how both parties will reconcile a reporting break. That is a contract-specific operational review, not a claim that Google’s change rewrites the terms.

Forecasting also needs a note about the break. A budget based on an old impression count, a new CPM and an unchanged fill-rate assumption may combine quantities which no longer fit together. The forecast can look precise while depending on inconsistent units. The practical repair is to state the counting basis alongside the assumptions and update the affected comparisons.

External communication should be equally exact. Advertisers need an account of the inventory they are buying. Internal teams need a way to distinguish changed measurement from changed demand. Boards need to see whether net cash, commercial obligations and audience experience have moved together or separately.

Google’s documentation says no action is required for the transition itself. That leaves an important distinction: the platform can change its counting without a publisher changing its code, while the publisher still has work to do to interpret the result. The absence of a mandatory technical task does not make commercial preparation unnecessary.

Give the new number an honest history

The strongest publishers will be able to explain why a number moved before being asked to defend it. They will preserve the old basis, establish the new one and separate a reporting change from the real performance of their journalism and advertising business.

The gain is not a prettier February dashboard. It is the ability to make a commercial decision using quantities that mean what the people in the room think they mean. When advertising is already under pressure, that is a small discipline with substantial consequences.

Research note: Noah completed the planned research checks; the exact method, counts, calculations and source trail are retained in the Claim audit.

Sources

Source: Tomorrow’s Publisher