Four interlocking developments landed on Sunday and Monday across digital advertising's two dominant structural questions: who controls the performance channel, and what is happening to the underlying measurement infrastructure that makes performance possible at all.

OpenAI activates performance

The date June 5 now carries specific technical weight. Beta advertisers notified by OpenAI on May 30 learned that conversion-optimized ChatGPT campaigns will begin rolling out that day, with a hard eligibility condition attached: any account lacking at least one active conversion event flowing through either the JavaScript Pixel or the Conversions API before June 1 will not receive early access. That cutoff is not administrative. A conversion-optimized campaign must have historical signal to train against, and without that data pipeline in place before launch, there is nothing for the system to learn from.

The measurement architecture is specific. The Pixel captures a privacy-preserving identifier called oppref from landing page URLs, stores it in a first-party cookie, and batches events prior to transmission. The server-side Conversions API sends purchase data directly from an advertiser's backend to OpenAI, bypassing browsers and the ad blockers increasingly sitting in front of them. When both run concurrently, deduplication requires a shared event_id. The design closely mirrors Meta's Conversions API in principle, though the field names and matching logic are OpenAI's own implementation.

Digiday first reported on May 28 that OpenAI had quietly activated cost-per-action ads inside the ChatGPT Ads Manager for select advertisers, allowing brands to pay only when a user clicks through, signs up, or completes a purchase. Three weeks earlier, OpenAI's head of monetization Asad Awan had described outcome-based bidding as "in the plan" and "should be done soon." Screenshots captured by Juozas Kaziukenas had already shown a "Conversions" objective appearing in the Ads Manager interface ahead of any official announcement, confirming a controlled release was already underway.

The product timeline is striking in its pace. The platform opened to all US businesses without a minimum spend requirement on May 5. Custom audience targeting arrived May 14. Daily budgets and ZIP-code-level geo-targeting were introduced May 22. Conversion optimization follows June 5. Six significant product additions in 26 days. The revenue context documented by PPC Land helps explain the acceleration: the ChatGPT ad pilot crossed $100 million in annualized revenue within its first six weeks. Criteo, one of a small group of technology partners, reported conversion rates on ChatGPT approaching twice those of traditional search in retail categories as of May 5. Those figures come from Criteo's own client reporting, not an independent audit.

A tension runs through the platform's public positioning. A May 29 Digiday briefing noted that OpenAI has been telling marketers to treat ChatGPT as a test channel rather than a performance one, while the pixel infrastructure, Conversions API integration, bidding tools, and partner additions - now including commerce media firm Skai alongside Criteo - collectively describe a platform building toward performance at scale. The gap between the messaging and the product roadmap is narrowing.

Google's self-assessment, and what replaces it

Two separate interviews published May 31 made clear how the dominant player in the search advertising market is negotiating its own structural pressures.

On the Decoder podcast, recorded May 26 and covered in depth by PPC Land, Sundar Pichai was shown a phone query for "best Chromebook" - a result topped by an AI Overview with a confident recommendation, followed by sponsored boxes, then an organic Reddit result, then a publication offering a different answer still. Pichai looked at it and said the result was "probably more opinionated than it should be" for that particular query. He acknowledged room for improvement while floating the possibility that the result was personalized to the interviewer's browsing history, a framing the interviewer challenged.

Third-party research has tracked the consequences of that editorial posture at scale. A February 2026 study found AI Overviews absorbing 58% of clicks that previously reached publisher websites, up from 34.5% in an earlier Ahrefs measurement. Conde Nast CEO Roger Lynch has since directed the company's brands to plan as if search referral traffic will reach zero - not as a scenario to hedge against, but as an operating assumption.

Google SVP Nick Fox, interviewed by tipsheet.ai on May 20, framed the commercial response to those tensions around checkout infrastructure rather than content. The Universal Cart - a single cross-merchant shopping interface spanning Search, the Gemini app, Gmail, and YouTube - was announced at Google I/O on May 19 and is positioned as the friction-reduction layer Google controls at the point where users abandon purchases. Fox noted that abandoned carts represent lost value for both users and merchants, and that the Universal Cart resolves the authentication burden that drives abandonment by standardizing the checkout layer across participating merchants. As of late May 2026, only 26 sites had implemented the underlying Universal Commerce Protocol, despite four months of activity around the standard.

Programmatic CPMs: display gains, CTV retreats

Against the AI platform competition, the underlying programmatic market is showing meaningful price strength in most formats. DataBeat's May 2026 US Programmatic Trends Report, published May 31 and drawing on data from more than 35 billion monthly impressions and $55 million in tracked monthly revenue, registered a 33.9% year-over-year overall CPM gain in April 2026. Month-over-month, overall CPMs rose 25.9%.

The channel divergence inside those numbers is sharp. Display CPMs rose 24.2% year-over-year; video recovered to a 2.1% year-over-year gain after a prolonged contraction. Connected television moved in the opposite direction, with CPMs falling 6.8% month-over-month and 25.8% year-over-year to an average of $4.68. The explanation is structural: new streaming entrants are expanding CTV inventory faster than advertiser demand can absorb it, compressing prices even as the total volume of money flowing into the channel continues to grow. Mobile CPMs surged 51.6% year-over-year to $1.44.

At the integration level, Ad Exchange expanded from 27% to 35% programmatic share, posting a 71.8% year-over-year CPM gain, while Prebid contracted from 56% to 48%. Within Prebid, Rubicon registered a 68.9% year-over-year CPM rise; OpenX fell 26.4% month-over-month. The ads.txt data from the same period points in the same direction: premium publishers trimmed their authorized seller lists while low-traffic creators are still adding to theirs - a supply chain consolidation at the top tier, and a supply chain build-out at the long tail.

Ad blockers crack a first-party tracking assumption

A development that connects directly to the measurement architecture OpenAI and others are building surfaced May 31. Filter lists maintained by AdGuard and the EasyPrivacy project now contain explicit blocking rules targeting subdomains used by server-side Google Tag Manager deployments, including instances using the standard sgtm. prefix.

The AdGuard tracking_servers_firstparty.txt file holds over 3,400 blocking rules. Among them: 83 entries matching the gtm. subdomain pattern and 42 entries matching sgtm. prefixes. These are not generic pattern rules. Each entry is a specific named domain - sgtm.kaspersky.de, sgtm.myprotein.jp, gtm.wise.com, gtm.temu.com, and dozens more spanning industries and geographies. The EasyPrivacy specific file, bundled by default in uBlock Origin and AdBlock Plus, already includes a rule for gtm.wise.com.

Server-side GTM had been positioned as ad-blocker-resistant precisely because its events originate from the same registrable domain as the website itself, avoiding the calls to googletagmanager.com that generic blockers target. The blocklist additions suggest that filter list maintainers have adopted a different approach: cataloguing specific subdomains by naming convention and community reporting rather than waiting for third-party origin detection to apply. The assumption that first-party routing provides durable protection from blocking is now demonstrably narrower than it was a year ago - which raises the stakes attached to architectures like OpenAI's oppref cookie and server-side CAPI, both of which face the same evolutionary pressure.

Also noted