Programmatic advertising in 2026: display, CTV, DOOH, ad tech, targeting and the Google ad tech ruling

Programmatic advertising is the automated buying and selling of individual ad impressions through software, in the fraction of a second between a page request and the ad appearing. It carries more than $200 billion of US ad spend in 2026, most of it now through private deals rather than the open auction, and in September a US court rewrote the rules for Google’s exchange.

How programmatic actually works

Four pieces of software do the work. The publisher’s ad server (usually Google Ad Manager) decides which demand source gets to fill a slot. The supply-side platform or exchange packages that slot and offers it for sale. The demand-side platform receives the offer, prices it against the buyer’s targeting and bidding rules, and returns a bid. Then the auction resolves and a creative loads.

All of that happens inside a bid timeout the exchange sets, normally somewhere between 100 and 300 milliseconds. Miss it and your bid is not counted.

The most useful thing a marketer can learn here is what a bid request actually contains, because every targeting decision downstream is built from it. A typical open real-time bidding request carries the page or app it comes from, the ad slot size and position, the device and operating system, coarse geography derived from IP address, the user’s consent string, and whatever identifiers are available: a third-party cookie, a mobile advertising ID, a hashed-email-based ID, or nothing at all. Your DSP is bidding on that packet of signals. It cannot target anything the packet does not describe.

Two mechanics are worth knowing because they still trip people up. First, open web auctions moved to first-price pricing across the industry from 2019, so the winner pays what it bid, not a penny above the runner-up. Second, most large publishers run header bidding, usually with the open-source Prebid stack, which asks several exchanges for a price in parallel before the ad server makes its decision. That parallel auction is why a single impression can generate a dozen bid requests, and why duplicate bidding is one of the transparency problems the industry is still arguing about.

The buy side is under more pressure than it looks from the outside. The Trade Desk, the largest independent DSP, reported second-quarter 2026 revenue of $715 million, up only 3% year on year, well below its historical growth rate (Q2 2026 earnings call transcript). Amazon DSP has taken share by bundling retail data with cheap supply, and the walled gardens keep absorbing budget that used to reach the open web. The independents have answered with agentic AI, and buyers should treat vendor performance claims about it with the scepticism they would apply to any self-reported test.

Open auction, PMP, preferred and programmatic guaranteed

Programmatic is not one market. It is four ways of transacting, and the choice of deal type does more for outcomes than most targeting decisions.

The balance has tipped decisively. Programmatic direct, which covers guaranteed and non-auction automated deals, accounts for 76.3% of US programmatic spend in 2026 according to EMARKETER, and 50.4% once social platforms are taken out of the base. Private marketplace deals alone represent more than 60% of programmatic spend on ANA figures cited in The Trade Desk’s Q2 2026 product release. The open exchange is no longer where most of the money is.

Where a programmatic dollar goes

This is the least comfortable section in any programmatic guide, and the one with the highest return on attention.

Adalytics analysed observed supply fees rather than contracted rate cards and found the same SSP charging one publisher 18% and another 8%, and a made-for-advertising site 10.75% while charging a biopharmaceutical news publisher 20%. In extreme cases intermediaries kept up to 98% of the bid. The ISBA and PwC programmatic supply chain study, run in the UK, found 51% of advertiser spend reaching publishers with 15% unaccounted for at all. That unattributable slice fell sharply in the follow-up study, but the pattern of wide, undisclosed variance did not.

Supply path optimisation is the response: deliberately choosing fewer, shorter routes to the same inventory. In practice that means auditing your seller paths in log-level data, blocking resellers you can reach directly, checking ads.txt and sellers.json declarations, and asking each SSP to disclose its take rate in writing. Most buyers can reach the same publishers through three or four paths rather than fifteen.

The industry is finally organising around this. On 21 April 2026 IAB Tech Lab launched a Programmatic Governance Council with Dentsu, Omnicom Media Group, WPP, Disney, Magnite, PubMatic, Hearst, News Corp, Yahoo, Amazon Ads, The Trade Desk, Raptive and Mediavine among the founding participants, aimed at transaction IDs, bid duplication and auction mechanics.

“Programmatic grew incredibly fast, but governance hasn’t kept pace with scale and complexity.” Anthony Katsur, CEO, IAB Tech Lab, 21 April 2026

Targeting in 2026, after the cookie deadline died

If your programmatic plan still has a slide about preparing for the end of third-party cookies, throw the slide away. On 17 October 2025 Google announced it would retire the Attribution Reporting API, IP Protection, On-Device Personalization, Private Aggregation, Protected Audience, Protected App Signals, Related Website Sets, SelectURL, SDK Runtime and Topics, on both Chrome and Android, citing low adoption. Chrome keeps CHIPS, FedCM and Private State Tokens. On the same day the UK Competition and Markets Authority released Google from its Privacy Sandbox commitments, concluding the competition concerns no longer arise, as AdExchanger reported.

Third-party cookies remain in Chrome. There is no forced deprecation and no separate opt-out prompt.

What buyers are actually using in 2026 falls into five groups. First-party data onboarded into a DSP or matched in a clean room, which is the strongest signal you can bring and the one that ties back to your measurement and attribution setup. Contextual targeting, which improved a lot once natural language models replaced keyword lists, and which needs no identifier at all. Interoperable IDs built on hashed email or phone, such as Unified ID 2.0, RampID and ID5, which work where a user has logged in somewhere. Seller-defined audiences, where the publisher declares audience segments in the bid request using the IAB taxonomy rather than passing an identifier. And household and device graphs, which carry most CTV targeting through IP address.

Data clean rooms sit underneath most of this now. A retailer, a broadcaster or a bank holds identifiers it will not export; the advertiser holds its own; the clean room matches them without either side seeing the other’s raw file, and returns an audience or a measurement result. It is slower and more expensive than dropping a pixel, and it is the direction the whole market is moving, because it survives regulatory scrutiny in the EU, the UK and India in a way that identifier syncing does not. Retail media networks are the biggest users, which is one reason retail media keeps taking share from open-web display.

Contextual deserves more credit than it gets. It survives every privacy change, it is cheap, and on the open web it often reaches the same person that a behavioural segment would have, because they are reading about the thing they are interested in right now. Pair it with a first-party seed list and you have covered most of what a cookie-based plan did, without the fragility. The same logic underpins how brands are being surfaced inside AI answer engines, where context beats identity too.

CTV and digital out-of-home

Connected TV is where programmatic growth now lives. US CTV ad spend reaches $37.95 billion in 2026, up 14.5% year on year, and more than 84% of CTV spend is transacted programmatically, on EMARKETER and Nielsen figures compiled by StackAdapt in March 2026. Buyers now expect 47% of their CTV inventory to be biddable, up 13 percentage points on 2024, and 119.8 million US households watch CTV.

The formats are getting more specific. The Trade Desk moved CTV pause ads, full-screen units that appear when a viewer pauses, into open beta on 12 May 2026, sold through private marketplace deals. The same release added deal quality scoring and Sincera inventory signals covering more than 400,000 publishers directly inside the buying workflow, and a live event pacing mode that delivered 49% more bids while spending 99% of budget in testing. If you are planning creative for these placements, the production side sits with your video marketing workflow, not your display team.

Programmatic digital out-of-home is smaller but moving fast. In research published on 19 March 2026, the DOOH supply platform VIOOH found that 34% of US campaigns had included programmatic DOOH in the previous 18 months, that the figure is expected to reach 52%, and that investment is projected to rise by an average of 49%, as reported by Billboard Insider. Ninety-one percent of US respondents said programmatic DOOH is now planned as part of wider digital activity rather than as a separate out-of-home line.

That last number is the interesting one. DOOH stopped being an outdoor buy and became a screen in the same plan as CTV and display, triggered by the same audience and dayparting rules. It is the clearest evidence that “programmatic” now describes a buying method rather than a channel.

The Google ad tech ruling and what it changes

2 September 2026 · Source: AdExchanger

In April 2025 the Eastern District of Virginia found that Google had unlawfully monopolised the publisher ad server market, where it held about 91% share, and the ad exchange market for open web display, in violation of Sections 1 and 2 of the Sherman Act. The Department of Justice failed on a third count covering advertiser ad networks. A remedies trial ran from 22 September to 6 October 2025, with closing arguments on 21 November 2025.

On 2 September 2026 Judge Brinkema rejected all three structural remedies the DOJ had asked for: divesting AdX, open-sourcing the final auction logic in DFP, and a contingent divestiture of the remainder of DFP. Her reasoning is that a forced break-up would hurt publishers more than the conduct does. She accepted most of the behavioural remedies the two sides had proposed, as modified.

Here is the honest caveat that most coverage skipped. The memorandum opinion was filed under seal, so the specific obligations are not public yet. What we know is the menu the parties put in front of the court, which included making real-time AdX bid amounts available to rival ad servers, deprecating unified pricing rules, letting publishers set different price floors for individual bidders, banning the use of first look and last look on open web display, equal latency and data signals across pathways, no differential revenue share based on which ad server a publisher chooses, a commitment never to rebuild first look or last look, and a monitoring trustee. Google had proposed a three-year supervision period. Redaction motions were due by 16 September 2026 and a jointly proposed final judgment by 2 October 2026, according to PPC Land. Until that judgment is filed, nobody can say precisely which of those survived.

Separately, the European Commission fined Google 2.95 billion euros over ad tech self-preferencing on 5 September 2025, so the pressure is not only American.

What should a buyer do about it this quarter? Very little. What should a publisher do? Read the final judgment on 2 October, and if unified pricing rules genuinely go, rebuild your floor strategy per demand partner rather than as one blanket rule. That single change, if it lands, is worth more to publisher yield than anything else in the list, and it is the line I would watch. We track the remedy filings and the wider ad tech beat in our September 2026 ad tech news report.

Quality, invalid traffic and made-for-advertising sites

Made-for-advertising sites exist to arbitrage cheap traffic against programmatic demand. They are stuffed with ad slots, thin on content, and they clear the same brand safety checks as real publishers because nothing on them is unsafe. They are the reason a campaign can post excellent viewability and terrible results.

Two datasets tell the current story. The ANA’s benchmark, built with TAG TrustNet and Fiducia on log-level data, found MFA spend among participating brands falling from 15% to 4% of media dollars, and the average programmatic campaign running across roughly 23,000 websites instead of the 44,000 in the 2023 study, as AdExchanger reported. Twenty-three thousand sites is still an absurd number for one campaign. Participation has since grown from 54 marketers to 86 in Q1 2026, per the ANA’s May 2026 update.

The second is Integral Ad Science’s 21st Media Quality Report, published on 9 July 2026 and built on more than 300 billion daily interactions. Mobile web display accounts for 45% of measured open-web impressions but 72% of MFA failures and 55% of brand suitability failures. Video viewability ran 11.8 percentage points ahead of display. Invalid traffic held around 1.1% across 2025, and inside CTV the split is stark: campaigns bought without quality optimisation ran 9.1% invalid traffic against 0.1% for optimised ones, as summarised by Digital Applied.

The practical lesson is that an inclusion list beats a blocklist. Blocklists are always behind; new MFA domains appear faster than anyone can catalogue them. A list of two thousand publishers you have deliberately approved will outperform an open buy with fifty thousand exclusions, and it is easier to explain to a client. Publishers on the receiving end of this shift should read our note on how content and publishing economics are changing alongside it.

KPIs and how to run a programmatic account well

Programmatic gets measured badly more often than it gets bought badly. Click-through rate on display is close to meaningless as a success metric; a 0.08% CTR tells you almost nothing about whether the campaign moved anything. Use it as a creative diagnostic and no more.

The metrics that matter, roughly in order: incremental conversions or sales measured against a holdout, cost per qualified impression after stripping waste, reach and frequency at the household or person level, viewable completion rate for video and CTV, and working media percentage, meaning the share of your budget that reached a publisher. Attention metrics are worth testing but are not yet standardised enough to plan against.

Two things get ignored and shouldn’t. Creative carries more of the variance in display and CTV outcomes than bid strategy does, yet most accounts run three sizes and one message for a quarter. And frequency is the quiet budget killer: without a household-level cap across DSP, CTV and social, a small group of heavy internet users absorbs a large share of impressions while your actual target barely sees the campaign. Cap it, then check the reach curve rather than the impression total.

A workable operating rhythm looks like this. Once a week, pull the domain and app report and cut anything below your spend threshold that is not on the inclusion list. Once a month, review supply paths and remove duplicate routes to the same publisher. Once a quarter, run a geo or audience holdout to check that the incrementality story still holds, and re-request fee disclosures. Once a year, re-tender your DSP seat, because pricing moves.

Programmatic works best when it is not asked to do the job of search. It is a demand-creation and reach channel with excellent targeting and measurement, sitting alongside paid search for capture and retail media for conversion close to the purchase. If you are building a channel mix from scratch, start with our guide to every digital marketing channel and come back here once search and analytics are stable. And if you want the fortnight’s ad tech developments in one place, the weekly digital marketing roundup covers them.

Last researched and updated: 7 September 2026.