Four years and change after Performance Max reached general availability, the campaign type has started handing back the levers it took away. Google is running an alpha in which selected buyers can deselect two inventory sources that were previously compulsory: the Search Partner Network, which serves Google results inside search boxes embedded on third-party sites, and the Google Display Network. The mechanism is unremarkable in form, appearing in the campaign console as two checkboxes that are enabled by default and can be unticked. Its significance sits in what it reverses.

Sam Clarke, managing director and head of search at Crossmedia, called the change significant, pointing to the absence of strategic levers as the original complaint when the format arrived. David Dweck, president at Go Fish Digital, described both networks as remnant supply that Performance Max advertisers had been compelled to accept. Brian Pappas, director of integrated search at Moroch, said his agency received access in the final week of June 2026. Kyle Rovinski, associate director of search at Duncan Channon, put the historical objection in four words: the product could not be trusted.

A Google spokesperson confirmed to Digiday that the feature remains in alpha, framed it as a pilot with a limited group of advertisers, and declined to say when broader availability might follow. Access is being rationed rather than announced. Buyers at three agencies said their requests were refused by Google representatives; James Viney, senior paid media account manager at Roast, said obtaining it had not been easy. Others found the checkboxes in their dashboards with no notification at all.

The commercial logic behind a concession of this kind is not hidden. Alphabet reported on July 23, 2026 that search advertising revenue rose 17 percent to $63.27 billion in the second quarter, with network revenue falling 1 percent and capital expenditure guidance climbing to $205 billion. Growth of 17 percent follows growth of 19 percent in the same quarter a year earlier. A deceleration of two points is not a crisis, but it arrives alongside the first credible competitor for paid search demand in two decades, and the cheapest available response to a restless buying community is to concede controls those buyers have requested since 2022.

Kaitlin McGrew, head of SEM at PMG, offered the clearest evidence that the strategy works. Reporting and control additions have already lifted Performance Max spending among the agency's clients, she said; after tests comparing Shopping campaigns against Performance Max, several PMG clients shifted budget into the latter, producing an average 10 percent increase in Performance Max investment. Transparency, in this reading, is not a cost centre. It is a retention product.

AdExchanger picked up the thread on July 27, 2026, adding the historical accounting: branded search controls did not arrive until 2023, after retailers spent millions bidding on their own trademarks, and the Search Partner Network has spent years drawing criticism as a source of brand-unsuitable placements whose delivery data Google withheld. PPC Land documented the intermediate steps along that path, including account-level placement exclusions arriving in January 2026 as a single unified block list spanning Performance Max, Demand Gen, YouTube and Display, and the November 2025 extension of channel performance reporting to cover search partners alongside Waze inventory for store goals campaigns. Industry research cited in PPC Land's earlier coverage of the partner network alpha put Search Partner Network return on ad spend 37 percent below Google Search proper.

A second exclusion appears the same week

On July 27, 2026, Search Engine Roundtable reported that household income exclusions had surfaced inside a Performance Max campaign, documented by paid search practitioner Thomas Eccel from a European account. The control operates at campaign level and offers seven brackets: top 10 percent, 11 to 20 percent, 21 to 30 percent, 31 to 40 percent, 41 to 50 percent, lower 50 percent, and unknown. Ticking a box removes that segment from delivery.

Household income exclusion is old machinery in Google Ads. It has been available in Search, Display, Demand Gen and Video campaigns for years, restricted to roughly twenty countries where Google models purchasing power, and it has never been the most accurate demographic signal because it is always inferred. What is new is its presence inside a campaign type built on the premise that the advertiser supplies goals and assets while the system decides everything else.

Two exclusions in one week, both landing at campaign level, both reversing the original design principle. The direction is consistent even if the pace is not. Neither has a confirmed general release date, and neither was announced through Google's usual product channels.

The same July 27 news cycle carried three further Google Ads items from Search Engine Roundtable: Google Ads Editor version 2.13 reaching general availability, an update to Google's conversion measurement documentation covering timing and data sources, and a reported issue with request indexing on robots.txt files in Search Console. Barry Schwartz also logged ranking volatility running from July 24 through the weekend, unattached to any confirmed update.

Demand Gen drifts into the same auction

While Performance Max was giving ground, the campaign type positioned beside it kept taking it. Smarter Ecommerce, the Austrian software provider and performance agency known as smec, published data on LinkedIn on July 27, 2026 showing that product feed adoption inside Demand Gen campaigns reached 35 percent in 2026, up from 26 percent in 2025 and 16 percent in 2024. Adoption climbed roughly ten points between the first two readings and nine between the second and third.

Feed support inside Demand Gen is not new. Behaviour is what changed. Attaching a Merchant Center feed alters which inventory and which formats a Demand Gen campaign can serve into, pulling catalogue data directly into units running across YouTube, Gmail, Discover and the Display Network, and unlocking shoppable connected television placements that now exist in both Demand Gen and Performance Max. For merchants without the budget to produce landscape, square and vertical creative in the ratios Demand Gen expects, a feed substitutes catalogue imagery for bespoke assets.

Smec's warning was blunt about the consequence. Demand Gen has drifted down the funnel, the company wrote, and feed-based campaigns can carry an uncomfortably high degree of overlap with what Performance Max is built to do. That collision has an audit trail. Google set out a formal separation between the two formats on February 13, 2025, casting one as end-to-end automated conversion optimisation and the other as controlled visual placement for consumers who are not searching. Eighteen months of releases narrowed the gap from both sides. At Google Marketing Live on May 20, 2026, Demand Gen gained Maps inventory, automotive feeds, checkout links in nine additional markets, and an AI-assisted creation flow that prefills a new campaign from an existing Performance Max setup. Six days later, Google confirmed that standalone Display campaigns are being retired and the Display Network folded into Demand Gen, with full transition expected during 2027.

Matt Rubinstein, paid search manager at 829 Studios, argued the inverse case in the comments: search, Shopping and Demand Gen together leave little advantage for Performance Max. The disagreement is real and unresolved by published data. A controlled programme run by measurement firm Fospha with Google between October 29 and December 16, 2025, covering 25 retail ecommerce brands across 28 market deployments, found brands allocating 10 to 20 percent of Google budget to Demand Gen recorded double the return on ad spend of brands allocating under 5 percent. No equivalent dataset yet measures what happens when a feed-enabled Demand Gen campaign runs beside a Performance Max campaign on the same catalogue in the same account.

Timing sharpens the question. Mike Ryan, head of ecommerce insights at smec, presented the company's fourth quarter session on July 28, 2026 at 3:00 PM Vienna time. Google alters bidding behaviour for budget-limited Target CPA and Target ROAS campaigns on August 17, 2026, a change that reaches Demand Gen line items in Display and Video 360 as well as Search, Shopping, Performance Max, Travel and Display in the main interface. Research published on July 21, 2026 found that made-for-advertising traffic rose 5 percent over Christmas 2025 while impressions climbed 219 percentbetween November 2 and December 5, a divergence that widens precisely when budgets peak.

OpenAI rebuilds the plumbing under ChatGPT Ads

The competitive pressure behind Google's concessions became measurable on the same day. Search Engine Roundtable published a consolidated list on July 27, 2026 of changes OpenAI shipped to ChatGPT Ads Manager over the preceding week, and the list reads like a systematic import of the machinery search buyers have used inside Google Ads for a decade.

Daily budgets changed on July 27 from fixed daily ceilings to average daily budgets measured across a seven-day period. Spend on any individual day may run above or below the stated figure, with the system compensating in later pacing so the campaign tracks toward its overall budget. Total budget does not change and no advertiser action is required. Alongside that, daily budgets now pace automatically through the day rather than exhausting early.

A Conversions objective arrived in campaign setup, creating conversion-optimised cost-per-click campaigns that steer delivery toward clicks more likely to produce the specified conversion event while billing continues on valid clicks. Craig Graham, who documented the option publicly, noted that a conversion-optimised bid cap can also be set at ad group level. Geo exclusion landed, letting advertisers remove specific locations from targeting. Automatic advanced matching arrived, improving website conversion measurement by matching conversions using hashed customer information, enabled through the conversions data source settings. Mobile measurement partner integrations with AppsFlyer and Adjust now attribute app installs and in-app events originating from ChatGPT Ads. A bulk API allows asynchronous creation and updating of campaigns, ad groups and ads. And a refreshed product card format for feed campaigns began rolling out, carrying price and star ratings.

Read as a list, these are small features. Read as a sequence, they are the components a performance channel needs before agencies will commit structural budget: an objective that maps to revenue, a pacing model that tolerates volatility, geographic negative controls, deterministic measurement, mobile attribution, and an API that scales beyond manual entry. The launch of ChatGPT advertising in February 2026 was the first serious challenge to Google's share of paid search demand; the six months since have added a conversion pixel, video formats and market expansion.

Measurement of that expansion has already begun. Adthena counted 7,378 advertisers running in ChatGPT with placements growing 97 times in a single quarter, with the United States holding 60.1 percent of observed advertisers and Booking.com leading all three tracked markets. The same firm reported UK cost-per-click swings of 278 percent inside ChatGPT auctions without published explanation, with three dentsu clients among the top nine UK advertisers on the surface. Similarweb data put generative AI web visits at 9.5 billion monthly, up 70 percent year on year, with AI Overviews appearing in 43 percent of US searches.

Passkeys become the price of API access

Google spent the same week tightening a different valve. Starting August 5, 2026, users following the Google Ads API authentication workflow will need a passkey to generate new OAuth 2.0 refresh tokens, with the rollout expanding to all users over subsequent weeks. Passkeys replace password-only authentication and displace SMS codes and time-based one-time passwords for that specific workflow. Anyone without a passkey will be prompted to create one during authentication, and existing refresh tokens continue working without reauthorisation.

The operational trap sits in the enrolment mechanics. A newly generated passkey enters a seven-day trust period on the device before Google treats it as fully trusted, which means access granted on August 5 to someone who enrols on August 5 may not behave as expected. AdExchanger flagged the likely casualties on July 28, 2026: agencies and software vendors rotating staff onto accounts, where a new team member needing immediate API access would suddenly not have it.

The security rationale is not abstract. AdExchanger described the scam pattern the change targets, in which operators searching for a Google Ads login click a sponsored result leading to a convincing fake, approve the two-factor prompt believing they are authenticating themselves, and hand control to an attacker who then drains the campaign budget to fund the next round of the same ads.

This is the second passkey deadline of the year. PPC Land reported in May 2026 that Google had emailed advertisers about a July 15, 2026 requirement covering sensitive in-account actions including account linking updates and user access changes, and that Google Ads API v24.1, released May 13, 2026, introduced a boolean field named passkey_enabled so developers could check enrolment status programmatically before attempting those operations. Version 25 of the same API, covered by PPC Land on July 23, 2026, removed the CustomerLifecycleGoal and CampaignLifecycleGoal resources and forced code rewrites for automated goal configuration.

Taken together, the week's Google news moves in two directions at once. Campaign controls loosen for a selected group. Account access hardens for everyone.

The AI bill lands inside principal media

The most consequential commercial story of the cycle involves no product at all. Digiday reported on July 27, 2026 that holding companies are offering to absorb clients' entire AI infrastructure bill in exchange for a committed share of spend routed through principal inventory, inventory the holdco buys in bulk at wholesale and resells bundled with its own targeting and data at a markup.

One CMO, mid-renewal, received an offer pitched at 70 percent of the media budget running through principal. Service costs, discounts and pricing terms all sat in the same negotiation, but principal media was among the largest items on the table. The mechanism is elegant in a way that should worry buyers: no separate token invoice ever appears. The markup on the principal allocation funds the AI commitment, and the larger the allocation, the more predictable the pool and the more compute the holdco can cover.

Two further executives confirmed to Digiday that similar conversations are underway. One recounted a CMO being offered a zero fee arrangement in exchange for routing all spend through principal media, with terms eventually negotiated back toward the middle. Another framed AI as the newest entry in a list that already includes offshoring and contract length extensions, the familiar levers by which efficiency promises get financed.

Robert Webster, a former WPP executive who founded the AI marketing consultancy TAU, was direct about the incentive: much of the claimed investment, he said, is manufactured to justify skimming money out of media. The counter-argument is less about bad faith than about timing. Nobody, including the holding companies, knows what a fair token price looks like. Compute costs are already appearing on profit and loss statements, capital decisions are already being made, and clients want a number now rather than in three years.

Daniel Knapp, chief economist at IAB Europe, described agencies as effective operators of futures markets through principal-based media, with the risk assessment skills and credit lines to support it, and said applying that model to a token world would fit the agency's natural DNA if the outcome can be priced. That conditional is the whole argument. Outcome-based pricing has been the industry's stated destination for years and remains concentrated in isolated pockets among large advertisers with the budget and internal alignment to anchor spend to revenue.

Ana Milicevic, co-founder of the consultancy Sparrow Advisers, placed the pattern in a longer arc: the industry accumulates evolutionary billing models, few of which still make sense five years on, and the transparency backlash follows. Transparency in these deals is negotiated rather than guaranteed, which means clients lacking the leverage to demand audit rights are adding a new layer of opacity to a practice many already distrusted. AdExchanger amplified the story on July 28, 2026, noting that building an AI billing model on top of existing media-buying infrastructure is a logical move given the ambiguity around pricing.

What Michaels found when shoppers stopped typing keywords

A retail data point published the same day cuts across all of the above. Michaels, the arts and crafts chain, announced its first customer-facing AI shopping assistant on July 21, 2026 and disclosed early performance on July 27. The tool, called Ask Mike, runs on the retailer's website across desktop and mobile as well as its iOS and Android apps, and was built on Google Cloud's Gemini Enterprise for Customer Experience, the same package Ulta Beauty, Macy's and The Home Depot have used.

The numbers are specific. Ask Mike has handled 75,000 conversations since a quiet launch in May 2026, and 27 percent of interactions end with the shopper clicking a product link or adding to cart. Heather Bennett, president and chief customer officer at Michaels, said shoppers who engage with the assistant convert at more than double the rate of those using traditional site search. Paul Tepfenhart of Google Cloud said the retailer moved from concept to production in six weeks; Kapil Dabi, Americas lead for retail and CPG solutions, attributed that pace to existing cloud infrastructure and a product catalogue already organised for machine consumption.

Bennett's explanation of the gap is the part worth holding onto. Traditional search forces a project to be decomposed into isolated keywords, leaving the shopper to sift results. Prompts such as planning a themed party or sourcing fabric for curtains arrive intact, and the assistant returns a bundle rather than an item. Shoppers using the assistant assemble core supplies and finishing touches within one session; those using site search more often arrive with a single product already in mind.

That distinction describes the same shift straining the paid search auction from the other end. If intent expressed as a project outperforms intent expressed as a keyword by a factor of two on a single retailer's own property, the pressure on keyword-shaped ad formats is not a forecast. Both Google and OpenAI spent this week building for a world in which the query is longer, the session does more work, and the advertiser controls less of what happens in between.

Also noted

  • July 28, 2026: InMobi Advertising launched Frame Attention, an in-app video attention model spanning mobile in-app video and connected television. MediaPost
  • July 28, 2026: Philippines-based artist Elmer Saflor filed a copyright infringement suit against Memes Apps, alleging his comic was offered to customers as an ad template through a paid AI service. MediaPost
  • July 27, 2026: Omnicom is folding Hearts & Science and Mediahub into a single agency under a new brand, with Mediasense chief strategy officer Ryan Kangisser predicting further consolidation across the holding company's portfolio. Digiday
  • July 27, 2026: A Lunio survey of 131 marketers found 5.3 percent run a dedicated invalid traffic platform while 75.6 percent report losing ad budget to bots, with half saying bidding now optimises toward automated traffic. PPC Land
  • July 28, 2026: YouTube Premium will carry a $10.99 Peacock tier from early 2027 under an NBCUniversal agreement, placing ad-supported inventory inside an ad-free subscription product. PPC Land