Display CPMs fell 45%. They also rose 34%. Which number sets your Q4 budget?

On 22 September 2026 AdRoll reported that display prospecting CPMs had fallen 45% year over year. Earlier in the year, DataBeat's programmatic data showed US CPMs up around 34% year over year. Both figures circulated as "the display market in 2026", and a lot of Q4 budgets are about to be set by whichever one somebody read last. They are not in conflict. They are measuring different markets, and knowing which one you buy in decides whether cheap display is an opportunity or a trap.

The two numbers, side by side

Here is what each dataset actually covers. The differences in the second and third rows are the whole explanation.

Read the bottom row twice. A report on what one platform's advertisers paid for prospecting during a summer lull answers a different question from a report on average prices across all programmatic supply in a different quarter. Neither is wrong. Only one of them is about your media plan.

Four questions to ask any CPM report

This contradiction will repeat, because CPM benchmarks are published constantly and almost none of them state their denominator in the headline. Before a number changes your budget, answer these.

1. Whose spend is in the sample? A platform reporting on its own advertisers is reporting on a self-selected group with a shared buying style. That is useful if you look like them and misleading if you do not.
2. Advertiser side or publisher side? Advertiser-side data is a price paid; publisher-side data is a price received, across a different mix. The two can move in opposite directions for a full quarter without either being false.
3. Which supply path? Open exchange, private marketplace and curated deals are effectively separate markets with separate prices. An average across all three moves when the mix moves, even if no individual price changes.
4. What window, against what comparison? July to early September is a seasonal trough in most consumer categories. A year-over-year drop measured in the trough is not the price you will pay in November.

That last point is the one that matters most for a Q4 decision, and it is the reason we would not extrapolate a 45% decline into Black Friday. AdRoll framed its own finding carefully: as a lower-cost window to build awareness before media prices typically rise around Black Friday and Cyber Monday. The cheap window is the pre-peak window, not the peak.

The only CPM benchmark that decides anything: yours

A market CPM tells you what attention costs. It cannot tell you whether that attention pays, because that depends on your click-through rate, your landing page and your allowable cost per acquisition. So calculate the price at which display breaks even for you, then compare the market to that.

Run it across the four situations most advertisers are actually choosing between, and the answer stops being ambiguous. Allowable CPA is held at 120 dollars in the middle column and 60 dollars in the right-hand one, so you can see how a leaner unit economic changes the verdict.

Now put the market price next to it. DataBeat's April 2026 figures put mobile programmatic CPMs around 1.44 dollars and desktop around 1.45. Against that:

• Row one is not a business. A 0.36 dollar break-even against a 1.40 dollar market price means cold display with weak creative and a homepage landing page loses money by roughly four to one. No CPM decline of 45% rescues it, because 45% off 1.40 is still 0.77.
• Row two is the interesting one. A 1.20 dollar break-even is marginal at last year's prices and viable at this year's discounted ones. This is precisely the row that a 45% prospecting decline moves from "no" to "maybe", and it is the row where creative quality and landing page conversion rate decide the outcome rather than the media buy.
• Rows three and four were never about price. With eight to twenty-five times headroom over market CPM, retargeting economics are decided by audience size and frequency, not by CPM. Which is also why a 29.1% retargeting discount is not the headline it sounds like: it makes a profitable thing slightly more profitable.

The practical consequence is that a falling prospecting CPM is an argument for improving your landing pages and creative, not for raising your display budget. The cheap impressions only pay at all in the row where those two things are strong.

When cheap display is worth buying, and when it is not

Cheap media creates a specific temptation: buying volume because it is available rather than because it does a job. Here is the rule we apply.

The measurement row is not optional in a quarter like this. Display's contribution is mostly view-through, so last-click reporting systematically understates it while the platform's own attribution systematically overstates it. If you cannot run some form of incrementality test, the honest position is to buy a small, fixed amount and treat it as an awareness cost, not to scale it on a reported ROAS you do not believe. This is the same problem that makes blended versus platform ROAS worth getting straight before Q4 rather than during it.

Why prospecting CPMs fell further than retargeting CPMs

The gap between the two AdRoll figures, 45% against 29.1%, is more informative than either number alone, and it points at something structural rather than seasonal.

Retargeting demand is relatively price-insensitive: an advertiser bidding on somebody who abandoned a cart has, as the table above shows, an enormous amount of headroom, so they keep bidding when prices move. Prospecting demand is the opposite. It is the first budget line cut when performance targets tighten, and it is the line most exposed to advertisers shifting spend into walled-garden inventory and AI-driven campaign types instead of the open web. Softer demand against available supply is the textbook condition for prices to slump, and that is consistent with what AdRoll observed.

There is a less comfortable reading too, and it should temper any enthusiasm: a CPM can fall because the average impression got worse. Less viewable placements, lower positions, weaker inventory. A price decline driven by quality decline is not a discount, and the only way to tell the difference from your seat is to look at the placement report and the post-click behaviour rather than the CPM line.

One structural thing to get right before you buy

Whatever you decide about price, do not build a new Q4 display programme on a campaign type with a published end date. Google is folding standalone Display campaigns into Demand Gen, with new standalone Display campaigns ending in January 2027. Any naming convention, reporting view or automated rule you create now inside a standalone Display campaign is work you will redo in a few months. We covered the timing decision in detail in Google Display is becoming Demand Gen; the short version for this quarter is to run the buy in a structure that survives the migration, and to keep the audience assets, customer lists, site visitor pools, portable rather than campaign-bound.

The same caution applies to spend that sits outside Google entirely. If the cheap inventory you are buying is on a third-party demand platform, check what happens to your audience pools if you stop spending there in January. Pools you cannot export are a rental, not an asset, and the whole justification for buying cheap prospecting impressions in October is the asset it builds for November.

What we would actually do with a Q4 display line in 2026

1. Calculate break-even CPM from your own CTR, landing page conversion rate and allowable CPA, per the formula above. Do this before looking at any price.
2. Compare it against a live price from your own platform over the last 30 days, not a published benchmark. Benchmarks are for context; your auction is the fact.
3. If prospecting clears break-even, cap it at an amount you would be willing to write off entirely, and run it in the five to six weeks before peak rather than during it.
4. Define the success metric as pool growth plus an incrementality read, agreed in writing before launch. Retargetable audience size is a legitimate and measurable deliverable.
5. Protect the retargeting budget first. It has the headroom, and a cheaper retargeting CPM means the same budget reaches the same people more often, which is a frequency decision that needs a cap, not a spend increase.
6. Re-price in the first week of November. A summer discount is not a November discount. Decide then whether the line continues, using the same break-even figure you calculated in step one.

If your Q4 plan does not yet have those numbers in it, that is the more urgent gap, our Q4 planning guide covers the sequence, and ROAS and CAC benchmarks by industry gives you a sanity check on the allowable CPA you plug into the formula. If the whole exercise is new, it is the same work that sits at the front of any performance marketing engagement: agree the allowable cost of a customer first, then decide what media clears it.

Frequently asked questions

Are display ads cheaper in 2026?

It depends which market you are buying in. AdRoll's Q3 2026 report found display prospecting CPMs down 45% year over year between 1 July and 8 September 2026, with retargeting CPMs down 29.1%. DataBeat's publisher-side programmatic data for April 2026 showed overall CPMs up about 34% year over year. Both can be true because they measure different sides of the auction and different supply paths: advertiser-side buys through one platform's demand versus CPMs observed across publisher inventory.

Why do display CPM reports contradict each other?

Four things differ between reports and any one of them can reverse the sign: whose spend is in the sample, whether it is measured advertiser-side or publisher-side, which supply path is included, and the time window. A platform reporting on its own advertisers' prospecting buys during a summer lull is not measuring the same thing as an SSP-side view of all inventory in a different quarter. Always read the denominator before the headline.

Should I buy display ads before Black Friday 2026?

Only if you can name the job it is doing and you have a measurement plan that is not last-click. Cheap prospecting impressions are genuinely useful for filling retargeting and Customer Match pools four to six weeks before peak, which is a real and measurable job. They are a poor choice for direct last-click revenue in a quarter when every other channel's CPM rises. If you cannot run a holdout or a geo test, buy less than you are tempted to.

What CPM makes display worth buying?

Work it backwards from your own numbers rather than from a benchmark. Multiply your click-through rate by your landing page conversion rate to get the share of impressions that convert, then multiply by your allowable cost per acquisition and by 1,000. That gives the CPM at which display breaks even. For a campaign with a 0.10% click-through rate, a 1% landing page conversion rate and a 120 dollar allowable cost per acquisition, break-even is 1.20 dollars per thousand impressions.

Does a falling CPM mean cheaper customers?

No, and this is the most common error in reading these reports. CPM is a price for attention, not for outcomes. A CPM can fall because demand softened, because supply expanded, or because the average impression got worse, less viewable, lower in the page, on weaker inventory. Only the first two make your customers cheaper. Judge the change on cost per acquisition and incremental revenue, not on the media price.

Is display advertising still worth it in 2026 if Google is retiring Display campaigns?

The inventory is not going away, the campaign type is. Google is moving standalone Display campaigns into Demand Gen, with new standalone Display campaigns ending in January 2027. That is a reason to make any Q4 display buy inside a structure that survives the migration, and to avoid building new reporting or naming conventions on a campaign type with a known end date.

The takeaway

Two credible datasets disagreed about display prices in 2026 because they were never measuring the same market. The reconciliation is not a compromise between 45% down and 34% up; it is that advertiser-side prospecting demand softened while publisher-side average prices rose, and only the first of those is about your media plan. Calculate your break-even CPM before you look at any benchmark. If cold display clears it, the cheap pre-peak window is a real opportunity to build audience pools you will use in November. If it does not clear it, and for most advertisers landing cold traffic on a generic page, it does not by a wide margin, then the 45% discount is a discount on something you should not be buying.