Three separate stories collided this week in ways that rarely happen simultaneously. OpenAI's ad business crossed from announcement into reality, confirmed brands now appearing inside ChatGPT responses. The Trade Desk's flagship supply-path initiative, OpenPath, lost two of the world's largest holding companies. And Google issued a string of product changes — some loud, some nearly invisible — that together tell a coherent story about a company tightening its AI-automation grip over advertiser controls. None of these stories exists in isolation. The same holding companies pulling back from The Trade Desk are simultaneously writing checks to OpenAI. The same Google that expanded AI Mode to 53 new languages on February 19 is the platform against which the Omnicom-IPG merger is explicitly positioned. What follows is the week's industry news read as a connected sequence of events, not a list.

ChatGPT's first real ads: who, how much, and how rarely they appear

OpenAI officially began serving ads inside ChatGPT on February 9. But the first confirmed public sightings only arrived this week. Adweek reported on February 20 that Adthena, a search intelligence firm, analyzed more than 500 prompts and found ads appearing in roughly 0.8% of responses. The confirmed advertisers: Expedia, Qualcomm, Best Buy, Enterprise Mobility, and The Knot Worldwide. Best Buy and Enterprise Mobility independently verified their participation. Expedia and Qualcomm did not respond to requests for comment.

The mechanics: ads are determined by the content of a user's current conversation, as well as past chat history and prior engagement with ads, per OpenAI's own blog post. Placements sit at the bottom of answers, are clearly labeled as sponsored, and do not appear near sensitive or regulated topics including health, mental health, and politics. Users under 18 are excluded entirely. The rollout covers both mobile and desktop — a point worth clarifying because, as AdExchanger reported on February 13, ChatGPT itself told a reporter the opposite when asked about ad availability. An OpenAI spokesperson confirmed the chatbot's answer was false.

The minimum spend to participate is $200,000, confirmed by OpenAI to Adweek. Some Adthena clients were asked for $250,000. The planned CPM rate is $60 — comparable to NFL broadcast inventory, and more than three times the prevailing CPMs on Meta. Adweek's earlier report identified Omnicom Media Group, WPP, and Dentsu as the first holding companies to commit, with Omnicom Media Group alone counting more than 30 client placements in the pilot. Named brands across the three holdcos include Adobe, Audemars Piguet, Audible, Ford, Mazda, and Mrs. Meyer's.

Search Engine Roundtable's February 20 coverage noted the Expedia sighting as the first time anyone in the broader community had spotted a live ChatGPT ad since the February 9 launch — eleven days after the official start. That gap between launch and public confirmation speaks directly to the controlled scale of the pilot. The 0.8% frequency Adthena measured is not a glitch. It reflects deliberate restraint, not a platform struggling to fill inventory.

PPC Land covered statements from Asad Awan, OpenAI's monetization lead, who outlined a longer-term vision: small businesses creating campaigns through conversational prompts rather than through agency-managed platform interfaces. That framing matters. It positions ChatGPT not just as another ad placement surface but as a potential alternative to the agency layer itself — a claim that, if taken seriously by the market, has structural implications for the entire holding company model.

AdExchanger's February 13 podcast laid out the core structural tension clearly: OpenAI is burning through billions in training and inference costs, revenue lags behind, and ChatGPT still lacks a pixel or any standard attribution system. The lack of attribution infrastructure is not a temporary omission. It reflects the fundamental design challenge of measuring conversions in a conversational interface where the user's next step is a typed response rather than a click on a landing page URL. Digiday's February 17 podcast added another layer: OpenAI required pilot partners to receive pre-approval before making any public statements about their involvement, and insisted those statements be labeled as a pilot rather than a launch product.

Omnicom posts first earnings as a merged company. The number is $17.5 billion

On Wednesday February 18, Omnicom reported its first earnings since closing the $13.5 billion IPG acquisition. AdExchanger's reporting put Q4 revenue at $5.5 billion and full-year revenue at $17.5 billion. The combined company is now the world's largest agency holding company by revenue. CEO John Wren and the executive team framed the entire deal as a "data-led AI transformation." The Omni platform, which is Omnicom's centralized identity and analytics infrastructure, is now integrated with Acxiom's Real ID — an IPG-owned identity solution acquired in the deal — alongside Flywheel's Commerce Cloud and Omnicom's own first-party data assets.

The more operationally concrete number is $1.5 billion. That is the projected cost savings over the next 30 months from the merger — double the original $750 million target announced when the deal was first proposed. Most of those savings come from restructuring: streamlining regional and brand structures, offshoring and outsourcing functions, and eliminating duplicative roles. IPG had already cut roughly 3,200 employees in the run-up to the merger. Omnicom has separately announced plans to eliminate around 4,000 additional roles.

CTO Paolo Yuvienco described a change in creative workflow that captures what "AI transformation" means in practice at the holdco level: where creative teams previously brought two or three concepts to a client, AI tools now allow teams to generate 20 or even 50 ideas before client review. Whether that translates into fewer creative staff or into faster iteration at the same headcount is the open question. Digiday's February 18 report on the same theme found that at S4 Capital, humans still anchor about 85% of the production process. AI flags anomalies and trends; humans act. That mismatch between the pace of AI capability and the pace of CMO adoption is, according to the piece, one of the defining patterns of early 2026.

The Trade Desk's OpenPath loses Dentsu and WPP — what that means

Adweek's February 20 exclusive reported that Dentsu and WPP have quietly exited The Trade Desk's OpenPath initiative. OpenPath, which launched as The Trade Desk's attempt to give ad buyers direct access to publisher inventory without routing through sell-side platform intermediaries, was positioned by CEO Jeff Green as entering a "steep acceleration phase" in 2025. The reality was different. Both holding companies cited lack of transparency around where ads were actually running, alongside what they described as hidden fees on the platform.

The withdrawal matters for several reasons. OpenPath is the mechanism through which The Trade Desk has argued it can serve as a viable structural alternative to Google's ad stack. If the world's largest holding companies find the platform opaque rather than transparent, that narrative becomes harder to sustain. The Trade Desk's stock is already down roughly 65% from its peak, weighed by competitive pressure from Amazon DSP and by tough year-over-year revenue comparisons. Digiday's reporting on The Trade Desk's fee structure noted that when all data, identity, and measurement layers are loaded, the effective platform fee can run from the mid-teens to roughly 20% — a level that becomes a direct performance variable at large budget scales.

The same holding companies walking back from OpenPath — Dentsu and Omnicom — are the ones committing hundreds of millions in combined client budgets to ChatGPT's ad pilot. That juxtaposition is worth sitting with. Agency leverage in programmatic exists partly through their volume with platforms. Shifting that volume toward new surfaces, even experimental ones, sends a signal about where agencies see future pricing power residing.

Freestar CEO Kurt Donnell, speaking on AdExchanger Talks on February 17, framed the broader publisher-side version of this problem: it is "freaking tough for publishers right now," but the industry is moving toward greater transparency that should ultimately benefit publishers who are not gaming auctions. The TID disputes of 2024 — where buy-side and sell-side players fought publicly over transaction IDs — at least forced both sides to articulate their grievances about information asymmetry. That same transparency pressure is now landing on The Trade Desk.

Google's search volatility reached a fever pitch this week

Search Engine Roundtable's February 20 video recap described search ranking volatility as "heated all week." That pattern has been documented across multiple tracking platforms and is consistent with Google's continued post-December core update environment. The December 2025 update, which PPC Land covered in detail when it launched December 11, completed after 18 days. But ranking instability continued into February 2026, suggesting the algorithmic systems governing search quality remain in active flux.

Alongside general volatility, Search Engine Roundtable reported on February 19 a renewed wave of complaints from businesses seeing Google local reviews disappear. The issue last appeared at scale in October 2025 and was fixed in December. Its reappearance in February suggests the review filtering mechanism may be miscalibrated again. On February 20, Search Engine Roundtable followed up to note Google had updated the Prohibited and Restricted Content section of its Google Business Profile review policies — a change that may explain some of the removals.

Also spotted this week: a new ad unit in testing called "Sponsored Places," which Search Engine Roundtable described as distinct from the standard "Sponsored results" label and from local packs with sponsored listings. Whether this is a formal product test or an isolated observation remains unclear, but its appearance is consistent with Google's ongoing effort to expand commercial ad surfaces within search results.

John Mueller of Google discouraged large sites from force-indexing their pages, saying on LinkedIn on February 20 that he "strongly recommends not relying on trying to force indexing." Mueller separately addressed concerns about bad title tags on Bluesky, stating that Google does not have a "we don't like this one guy's titles" filter — clarifying that title quality affects individual pages, not sites as a whole.

Google's own product changes: Demand Gen targeting, parked domains, Ad Manager, and AI Mode

Several discrete Google product changes landed this week that individually look routine but together represent a consistent directional shift.

The most significant for advertisers managing Demand Gen campaigns: Google announced on February 17 that Lookalike segments in those campaigns will shift from hard targeting constraints to audience suggestions, effective March 2026. The change is not a minor parameter update. Lookalike segments in Demand Gen have previously enforced strict similarity thresholds — narrow at 2.5%, balanced at 5%, broad at 10% of the target location. Starting March, the system treats those same thresholds as directional signals that can be expanded by Google's AI models to optimize for conversions or CPA. Advertisers who have used narrow Lookalike segments to maintain tight audience control, or to exclude certain users, will find that control removed by default. An opt-out exists but requires a form submission. UI controls to opt out directly from the Google Ads interface are expected later in 2026. PPC Land covered this in detail, including the technical interaction between this change and Optimized Targeting — a separate Google Ads feature that can expand reach beyond defined audiences when the system identifies likely converters. The two mechanisms now operate in parallel within the same ad group, which is new and consequential.

On February 17, PPC Land also reported that Google removed parked domains — also called AFD (ads on parked domains) — from its Search Partner Network effective February 10, 2026. Parked domains are inactive websites displaying ads to users who navigate to unregistered URLs. The format had persisted for years and was consistently associated with poor conversion rates. Its removal reduces a low-quality inventory surface but also eliminates a traffic source that some arbitrage-focused advertisers had relied on for volume.

Google Ad Manager's legacy Reports tool has a shutdown date: May 4, 2026. Publishers using scheduled reports that run automatically will find those reports stop executing on that date. The replacement infrastructure is the newer, unified reporting environment within Google Ad Manager. The transition window is relatively short given how deeply embedded some publisher reporting workflows are in legacy tooling.

Google AI Mode — the conversational search surface that now reaches over 75 million daily active users — expanded to 53 new languages, Search Engine Roundtable reported on February 19, citing an announcement by Nick Fox from Google on X. The expansion brings AI Mode to just under 100 languages total. The geographic and linguistic reach of AI Mode matters for advertisers because shopping ad placements inside AI Mode — announced by Google on February 11, as PPC Land covered — now have a much larger potential global surface. AI Mode's ad placements are explicitly commercial in format: sponsored shopping results appearing beneath product recommendations in response to conversational product queries.

Google Ads also updated its budget pacing for campaigns using ad scheduling. Search Engine Roundtable reported on February 20 that Google sent emails to affected advertisers about changes taking effect March 1, 2026. Ad scheduling allows advertisers to restrict when ads run during the day or week. The pacing update affects how budget is distributed across those scheduled windows — a change with downstream effects on dayparting strategies.

Walmart's $6.4 billion ad business and the retail media context

On February 19, AdExchanger reported that Walmart's full-year 2025 advertising revenue totaled $6.4 billion, representing 37% growth globally and 41% growth domestically for Walmart Connect. CFO John David Rainey told investors during the earnings call that fully a third of profit in the most recent quarter came from advertising and membership income. For reference, Snapchat — a social network with over a billion users and a mature ad business — generated $5.9 billion in total ad revenue for 2025, growing 11%. Walmart's ad business is now larger than Snapchat's and growing at nearly four times the pace.

Walmart's trajectory sits within a broader retail media context that PPC Land tracked earlier this year when Walmart Connect announced its AI strategy in January — including an advertising assistant in beta for Sponsored Search campaigns and integration with the Sparky AI shopping agent. The $6.4 billion figure is also the competitive benchmark against which Amazon's ad business is measured. Amazon does not break out retail media revenue separately from its broader advertising services segment, but Amazon remains structurally dominant in retail media at a scale that Walmart is still working toward. What the Walmart number confirms is that the retail media race is real, the growth rates are exceptional, and the platforms outside the traditional duopoly are generating ad revenue at a pace that merits attention.

AI and agentic standards: two competing frameworks

Digiday reported on February 17 on the competing protocol frameworks emerging for agentic AI in advertising: the Ad Context Protocol (AdCP), developed with Scope3 as a founding member, and IAB Tech Lab's Agentic Roadmap. The two approaches represent different visions of how AI agents should interact with advertising infrastructure. AdCP is a more open, publisher-side-oriented framework. IAB Tech Lab's roadmap is more comprehensive and standardization-focused. The dispute is not academic — which framework achieves industry adoption will determine how agentic buying systems interact with publisher inventory, how attribution flows in a world where AI agents initiate transactions, and who controls the identity layer when an agent completes a purchase on a user's behalf.

Prebid has also entered this space. As of early February, it is taking charge of a seller agent as part of AdCP architecture, and Amazon's integration with Prebid is in beta. PPC Land covered the IAB Tech Lab taxonomy side of this infrastructure question, reporting that Mixpeek donated an open-source AI-powered taxonomy mapper to IAB Tech Lab that accelerates Content Taxonomy 2.x to 3.1 migration from what was previously a weeks-long manual process to a matter of seconds. The mapper uses TF-IDF and LLM methods to match categories — a practical piece of infrastructure that matters for publishers classifying inventory within programmatic systems.

ICE, location data, and ad tech's threshold moment

AdExchanger reported on February 19 a developing story around Immigration and Customs Enforcement (ICE) exploring how ad tech and location data could assist federal investigations. The question — whether ad tech companies will cooperate with such requests or draw a line — is not new in the abstract, but its practical urgency is sharpened by the scale and enforcement posture of the current federal immigration crackdown. The same week, AdExchanger published a detailed profile of how the Minnesota Star Tribune is navigating the business tension between aggressively covering ICE operations in the state — including an active lawsuit against the city of Minneapolis over police misconduct records — and protecting its advertiser relationships. Roughly 80% of the Star Tribune's advertising revenue is now digital, with the remaining 20% from print. Of the digital slice, approximately three-quarters comes from agency services. Direct advertising represents about 60% of digital revenue with programmatic filling the remaining 40%.

TV convergence, still not converged

AdExchanger's February 17 report surveyed ten industry executives on the state of TV and video convergence. The summary verdict from buyers and sellers: "Transitional." "Developing." "Work-in-progress." With connected TV advertising in the US expected to grow 13.8% in 2026, buy-side respondents identified sports as their single biggest investment priority — ahead of interactive formats, generative AI, and agentic AI. Two of three sell-side respondents cited "underwhelming growth in investment" as their top concern, reflecting pressure as the upfront season approaches and streaming competition for ad dollars intensifies. The gap between where buyers say they will put money and where sellers say growth is actually landing is one of the persistent frictions in the CTV market.

The March Madness tournament is the next major activation window. AdExchanger reported February 17 on DTC agency Rain the Growth's use of show-level metadata from Spectrum Reach last year to identify the highest-performing programmatic CTV placements across a major food and beverage brand's March Madness campaign — an approach that delivered the agency's highest response rate for the quarter. Matching budgets to specific games and time slots rather than spreading evenly across programming windows is the emerging operational model for sports CTV buying.

Apple's HLS video podcast move and what it means for ad inventory

On February 16, PPC Land reported that Apple announced HLS (HTTP Live Streaming) video podcast capabilities for Apple Podcasts, with Acast as one of four launch hosting providers. The technical significance is dynamic video ad insertion: automated placement within video episodes while preserving host-read audio integrity. The announcement coincided with The Best One Yet podcast — 80 million lifetime downloads — signing a multi-year exclusive with Acast for global ad sales and distribution. The combination of Apple's infrastructure announcement and a major podcast deal on the same week signals the industry's direction: audio-first podcasts gaining video ad inventory as a parallel revenue stream. Podcast advertising grew 32% year-over-year in Q4 2025, according to Magellan AI data cited in PPC Land's coverage. The ad insertion infrastructure Apple is now enabling makes video podcasting a programmatic surface, not just a brand-deal surface.

The state of the agency sector: employment, consolidation, and AI pressure

Digiday's February 20 graphic analysis of the agency sector painted a consistent picture: fewer employees, more reliance on tech credentials and data capabilities, and ongoing consolidation as agencies either merge or, as in Dentsu's case, restructure to shed divisions. The employment figures published by Britain's Institute of Practitioners in Advertising earlier this month showed a long-term decline that has been accelerating. Omnicom shed thousands of staff last year; its leadership plans hundreds of millions more in "synergies." The agency model is under pressure from in-housing (which has grown steadily for over a decade), from AI-driven workflow compression (which reduces the hours-based billing model's economic logic), and from the platform giants whose programmatic tools are designed to reduce the need for agency intermediation at lower spend levels.

Timeline

Share this article
The link has been copied!