Agentic ad infrastructure races ahead as AI market share reshuffles
Mediaocean, Magnite, and Walmart Connect rewire agentic ad setup, CTV buying, and retail measurement as ChatGPT drops 24 points of AI traffic share in 2026.
Three separate product announcements published on June 11, 2026 arrived at what feels like the same structural moment: the advertising industry is no longer debating whether to build AI agents into campaign workflows, it is now competing over which layer of the stack those agents will inhabit. Mediaocean launched NIVO AI on top of Innovid’s ad serving infrastructure, Magnite introduced an orchestration layer designed to let buyer-side agents transact directly with premium supply, and Walmart Connect tied its first-party purchase data into Google’s Display and Video 360 platform. Taken together, the three announcements describe a supply chain for automated campaign execution that, until recently, existed only in vendor roadmaps.
The backdrop to all of this is a market-share chart that has been moving faster than most forecasts projected. Similarweb data published on June 11 showed ChatGPT’s share of global generative AI website traffic falling from 76.4% to 52.7% in twelve months, while Gemini climbed from 8.9% to 27.3% over the same period. That redistribution is not merely a product ranking story. It carries direct implications for where brands need to be present to be found, and therefore for how advertising infrastructure must adapt.
Mediaocean’s NIVO AI: twelve agents, one intelligence layer
Mediaocean’s NIVO AI, announced on June 11, is the first product to formally carry Innovid’s name as an artificial intelligence layer running across the combined Mediaocean stack. The company framed NIVO in architectural terms: “NIVO thinks, Orchestrator connects, and Agents act.” That is not marketing shorthand. It describes three distinct functional layers. NIVO handles reasoning and decisioning. Orchestrator, a workflow coordination product Innovid introduced the previous year, connects systems and passes instructions between them. Specialized agents then carry out discrete tasks at execution.
The acquisition context matters. Mediaocean closed a $500 million deal to acquire Innovid in early 2025, merging Innovid with Flashtalking, its earlier acquisition, to create a combined entity covering ad delivery, creative personalization, measurement, and optimization across CTV, digital, and social. NIVO is the product emerging from that consolidation. Mediaocean processes more than $200 billion in annualized ad spend through its software, and its Prisma media management platform is used by more than 100,000 people globally.
The agent architecture is the most technically specific element of the launch. Twelve agents are grouped into four categories. The two creative agents address opposite ends of the content cycle: the Creative Generator Agent produces and adapts creative at scale, while the Predictive Scoring Agent estimates performance before launch. Four delivery agents handle operational mechanics. The Campaign Trafficking Agent builds and traffics campaigns automatically; the Campaign QA Agent catches errors before assets go live; the Decisioning Agent translates strategy into dynamic execution; the Taxonomy Agent standardizes data labeling for cleaner reporting downstream. Two measurement agents surface performance. The Reporting Agent answers cross-channel performance questions through natural language; the Creative Insights Agent surfaces real-time asset-level data. Four optimization agents work in-flight, including the Reach and Frequency Agent, which identifies overexposed and underexposed households in connected television in real time.
That last agent addresses a problem the CTV industry has carried for years. Advertisers running campaigns across multiple streaming services have had limited visibility in real time into how often the same household sees the same ad. Innovid’s structural position as an independent omnichannel ad server, not owned by any publisher, places it in a rare position to surface that cross-publisher signal. Whether it delivers in practice depends on individual publisher data-access agreements, but the architectural positioning is distinct from anything a walled-garden tool could offer.
Pilots with five organizations produced claims of up to 90% improvement in speed to campaign launch compared to manual setup. The five participants were Canvas Worldwide, FanDuel, Optimum, Paddy Power, and Tailwind. Stephen Rubino, Media Operations Manager at FanDuel Group, described the shift as moving “from managing campaigns to orchestrating outcomes.” Raul Tafur, VP of Paid Social and Strategy at Canvas Worldwide, pointed specifically to the gap NIVO addresses: identifying performance opportunities and failing to operationalize them at speed.
FanDuel’s context is worth noting. Sports betting advertisers operate some of the highest-complexity programmatic environments in existence, with large creative volumes, rapid geographic regulatory variation, and tight timing constraints tied to live event schedules. If NIVO holds up under those conditions, the pilot results carry credibility that a simpler advertiser test would not.
NIVO at launch connects to Meta and Snapchat platforms. The company has not published an integration timeline for other platforms.
Magnite Orchestration: agent-to-agent transactions in programmatic
On the same day, June 11, Magnite launched Magnite Orchestration from New York, describing it as a coordination layer designed to let buyer-side AI agents connect directly to the company’s sell-side infrastructure. The first partners testing expanded capabilities in beta are dentsu and DIRECTV Advertising.
The product’s architecture sits above the transaction layer. Magnite built Orchestration as a shared environment where AI-driven buying systems can discover, evaluate, and activate premium omnichannel inventory without requiring custom point-to-point integrations between individual buy-side and sell-side systems. Buyers connect their own agents using open integrations designed for interoperability across planning, activation, and optimization tools.
The sell-side component is represented by a Seller Agent that gives publishers two capabilities: packaging inventory and audiences with flexible pricing and targeting controls, and making that inventory discoverable and purchasable by buyer agents. The result is what Magnite describes as agent-to-agent transactions. Automated systems on both sides of a deal negotiate and execute buys without a human manually matching supply and demand at each step.
Buyers who prefer their own agentic infrastructure can connect proprietary systems into Orchestration rather than adopting Magnite’s native buyer tooling. That design choice is significant. Magnite is positioning Orchestration as infrastructure that third-party systems plug into, not a closed platform that requires exclusive adoption.
How audience data moves through the system is one of the more specific technical elements. According to Magnite, buyers, publishers, and data providers can make their own audiences available to agents operating inside Orchestration, and those audiences are packaged alongside premium supply within the agent workflow. Dentsu integrated its dentsu.Audiences segments into the system during beta testing. Nick Halas, Head of Product Strategy at dentsu, described the integration as “grounded in a unified identity and data layer” and said Magnite’s infrastructure enables dentsu to bring “real-time data, optimization, and decisioning capabilities closer to activation.”
The historical friction that matters here is latency. The gap between a data signal existing inside a data management platform and that signal actually influencing a live auction has long measured in hours or, in many cases, days. Embedding audience segments inside the coordination layer is an architectural attempt to reduce that gap for agentic workflows.
DIRECTV Advertising’s participation follows a pattern of programmatic infrastructure expansion. The company opened its DOOH network to programmatic buying in January 2026 and has since extended its programmatic reach into additional inventory types. Drew Groner, SVP and Head of Sales and Marketing at DIRECTV Advertising, framed the beta involvement as ensuring “seamless access to premium inventory” as new buying models take hold.
Walmart Connect and DV360: closing the retail loop on YouTube
Where Mediaocean and Magnite are building new agent infrastructure, Walmart Connect’s June 11 announcementaddresses a different problem: connecting the ad exposure to the confirmed sale. The partnership with Google ties Walmart’s first-party omnichannel shopper data directly into Display and Video 360, letting advertisers activate purchase-based audiences on YouTube and measure whether those video impressions resulted in actual Walmart transactions.
The integration operates on three components. Audience activation is the first: advertisers working inside DV360 will find Walmart Connect audience segments available as targeting options without switching platforms or configuring a separate data integration. Those segments are built from verified purchase behavior spanning both Walmart.com and physical store locations. Walmart operates more than 4,500 stores and reaches approximately 280 million customers and members weekly, giving the data set a transactional breadth that modeled or interest-based alternatives cannot replicate.
Closed-loop measurement is the second component. Once a campaign runs, advertisers can measure how YouTube impressions influenced confirmed purchases at Walmart, across both digital and store channels. The attribution rests on Walmart’s own transaction records, not a probabilistic model. For a media environment where many offsite campaigns still rely on correlation-based attribution, the distinction is material.
Workflow continuity is the third. Both audience data and measurement reporting are accessible directly inside DV360, meaning the entire process stays within the platform that agencies and large brands already use for planning.
Courtney Rose, VP of Retail at Google, stated that viewers watch an average of 110 million hours of shopping content on YouTube every day, framing the platform as a commerce context rather than purely an awareness channel. The Walmart Connect deal fits into a pattern of retail data entering DV360 that PPC Land has tracked through 2026. Kroger Precision Marketing joined DV360 through Google’s Commerce Media Suite in March 2026, introducing SKU-level sales attribution. Costco and Dollar General joined the same suite at YouTube Brandcast in May 2026. Walmart, as the largest US retailer by revenue and physical footprint, extends that architecture further.
The partnership begins with YouTube inventory, with additional inventory types described as planned to follow.
It is also worth situating the announcement in the broader retail media context. Walmart Connect has been pursuing a distribution strategy rather than a consolidation strategy, pushing its first-party data onto platforms where media buyers already operate rather than requiring those buyers to shift budget into Walmart’s proprietary DSP. PPC Land covered a similar move in March 2026 when Walmart made its shopper audiences available through The Trade Desk. The DV360 partnership extends that same philosophy to a different buyer base. Enterprise brands and agencies that use DV360 as their primary programmatic surface can now access Walmart’s purchase-based audiences without learning a new interface or renegotiating a separate data deal. The commercial logic is straightforward: Walmart’s first-party data becomes more valuable as its reach increases, and placing it where buyers already work accelerates that reach.
ChatGPT at 52.7%: the AI traffic map changes
The infrastructure announcements land against a market-share backdrop that changed substantially in twelve months. Similarweb data published on June 11 showed ChatGPT holding 52.7% of worldwide generative AI website traffic as of May 26, 2026. Twelve months earlier it held 76.4%. The 23.7 percentage point decline is large in absolute terms, but the trajectory is what commands attention.
The rate of loss was not constant. Six months ago ChatGPT still held 65.2%. Three months ago it stood at 56.7%. One month ago it sat at the same 52.7% mark as the latest reading. The pace of decline appears to be slowing, though the overall direction remains downward.
Gemini is the primary beneficiary. Google’s platform held 8.9% twelve months ago, reached 20.3% six months ago, climbed to 25.5% three months ago, and stands at 27.3% in the latest data. In proportional terms, Gemini’s traffic share tripled over twelve months while ChatGPT lost nearly a quarter of its global standing.
The measurement question that circulated in comments on Similarweb’s LinkedIn post is worth noting. The dataset covers generative AI website traffic worldwide, which in Similarweb’s methodology generally means direct visits to AI platform domains rather than AI-generated answer surfaces embedded in search results. If that reading is correct, the 27.3% figure represents Gemini as a standalone chat product, and the growth is not inflated by AI Mode or AI Overviews traffic. That interpretation makes the competitive shift more, not less, remarkable.
The fastest single-month gain in the latest period came from Anthropic’s Claude. One month ago Claude held 6.0% of global generative AI traffic. By May 26 it had reached 8.9%, a 2.9 percentage point increase in a single month. Twelve months ago Claude held just 1.6%. The fivefold multiplication of its traffic share over the year is the kind of trajectory that changes assumptions about competitive positioning. PPC Land has tracked this since Comscore’s March 2026 consumer AI rankings showed Claude’s usage surging 130% year over year.
For marketing and media buying teams, the implications are structural. Optimization for AI-generated answers, sometimes described as GEO or AEO, has concentrated heavily on ChatGPT as the dominant surface. A Gemini platform commanding more than a quarter of global AI traffic demands equal attention. So does a Claude platform growing at 2.9 points per month. The channel mix has changed, whether media plans reflect that or not.
The traffic redistribution across AI platforms connects to a parallel trend documented in search. Search Engine Roundtable reported on June 10 on updated SparkToro and Similarweb data showing that 68.01% of Google searches now end without a click to the open web. The more granular finding is starker: when AI-generated responses are involved, only 27.6% of interactions result in a click to an external site. Taken together with the AI traffic share data, the picture is of an information environment where users increasingly receive answers without leaving the AI surface, whether that surface is a generative AI chat product or Google’s own AI Overviews. Publishers and advertisers who have built traffic models assuming consistent referral volumes from search are confronting a structural change that neither the Similarweb data nor the zero-click research alone fully captures, but that both confirm from different angles.
Beijing unwinding a $2 billion deal: Meta and Manus
The most consequential story from the past 24 hours does not fit neatly into the agentic advertising narrative, but it intersects with it in significant ways. PPC Land reported on June 11 that Meta has blocked Manus from its internal data systems and ordered staff to migrate existing projects off the platform, following China’s National Development and Reform Commission order to reverse the $2 billion acquisition of the Chinese-founded agentic AI platform.
Meta closed the Manus acquisition on December 29, 2025. By February 21, 2026, the Manus AI shortcut had appeared in the navigation menu of Meta Ads Manager, positioned under the “Manage” section between Instant Forms and Media Library. Manus had been designed to handle report building, audience research, and campaign analysis tasks autonomously inside the ad management workflow. By some accounts, it had also been connecting to Similarweb market intelligence data as part of the Manus Pro feature set, a detail PPC Land covered when tracing the data partnership implications.
The NDRC opened a probe almost immediately after the December announcement. The regulator concluded the transaction violated China’s foreign investment and technology export rules regardless of where the company was formally incorporated. Manus had relocated its headquarters from Beijing to Singapore in 2025, a step that plainly did not insulate it from Chinese regulatory authority. As of this week, Meta’s internal memo instructs staff to migrate projects and describes Manus as being “sunsetted.”
The advertising integration consequences are unresolved. Manus still offers advertisers the option to connect to Meta’s Ads Manager and Instagram in its interface, but the data firewall that Meta erected at the start of June blocks deeper system access. The broader implication for the AI industry is that incorporating a Chinese-origin entity through a Singapore holding structure does not automatically place it beyond Beijing’s reach. That precedent will be watched carefully by any company evaluating similar acquisitions.
Measuring what is actually new: Google Analytics Source Group
On the quieter end of the day’s announcements, PPC Land reported on June 12 that Google Analytics has introduced Source Group, a new dimension that consolidates fragmented source values for Facebook, Instagram, and TikTok into single standardized entries. The update also adds native groupings for AI traffic sources including ChatGPT and Perplexity.
Source Group is distinct from the existing Source Platform field. Where Source Platform operates one level up, labelling Instagram traffic as “Meta Ads,” Source Group identifies the specific origin, collapsing variant strings such as “facebook,” “fb,” and “Meta-facebook” into a single clean “Facebook” value. The practical effect is that cross-channel budget analysis no longer fragments a single platform’s traffic across a dozen rows in a report.
The AI traffic groupings connect directly to the market-share picture described above. In May 2026, Google Analytics added a native AI Assistant channel to the Default Channel Group, automatically tagging traffic from ChatGPT, Gemini, and Claude without requiring configuration. Today’s Source Group update adds source-level standardization for that same traffic. As AI referral traffic grows from platforms that are each adding web browsing capabilities, the measurement infrastructure is beginning to reflect a traffic mix that practitioners are already experiencing but struggling to quantify cleanly.
The combination of accurate AI traffic measurement and the market-share data from Similarweb creates an environment where practitioners can now begin to observe, with some precision, how much of their actual site traffic originates from which AI platform, rather than relying on estimated totals that aggregate every variant string into a single opaque bucket.
Also noted
June 11, 2026: Teads launched EngageOS, a publisher feed operating system designed to unify editorial and advertising auctions in a single environment, with Magnite as the launch partner for programmatic demand, explicitly positioning the product as a response to declining AI-driven publisher traffic.
June 11, 2026: Viant launched its free SupplyIQ portal, giving publishers direct visibility into how Viant’s DSP scores, bids on, and values their CTV and programmatic inventory, with the portal available to all publishers from that date.
June 11, 2026: MiQ expanded its Sigma AI platform one year after the original launch, adding a Planning Agent, Total Measurement product, 2.5 petabytes of daily data processing, and activation across 16 environments.
June 11, 2026: Spotify redefined podcast plays as 30-second listens, aligning with the AMP industry standard, and launched five new creator analytics tools including Audience Segments and Episode Trends, alongside access to historical performance data.
June 10, 2026: Dexerto launched Omnidex, its first supply-side platform, on June 10, processing 90 billion monthly ad requests across more than 600 gaming publishers, with McDonald’s and PlayStation among the initial demand partners.