"It seems budget is the new quality score." Daniel Toledo, director of group media and performance marketing at evoke, the gambling group that owns William Hill and 888, wrote that line on LinkedIn on Saturday, October 10, in reply to a post published a day earlier by Joey Bidner, a freelance Google Ads manager and coach. Bidner's complaint was compact: Google now treats "limited by budget" as a performance measure, and in his words "budget should not be a lever for performance." PPC Land documented the exchange on October 10, and noted what was missing from the thread. Nobody from Google replied, and nobody posted data.
The remark lands because of what Quality Score was supposed to be. When Google introduced it in 2005, alongside Ad Rank and quality-based minimum bids, the promise was that a relevant ad could beat a richer rival. Relevance was the currency an advertiser could earn. If budget has taken that place, the currency is simply money, and the party that benefits most from more money entering the auction is the one that runs it.
That is the argument of this special edition, and it rests on a sentence that sounds obvious once it is said plainly. Google optimises for Google's revenue. Advertisers have to optimise for their own. Most of the time the two lines move together, which is why the auction works at all: a click that sells a product pays both sides. The trouble sits in the gaps where they part, and through 2026 those gaps have widened one default, one help-page edit and one removed control at a time. What follows draws on PPC Land's reporting and on coverage in Search Engine Roundtable, PPC News Feed, Digiday, AdExchanger, Adweek and MediaPost, going back to 2022 where the record requires it. It is not a claim that every change Google makes hurts advertisers. Several clearly do not. It is a record of who holds the dials.
Luis Rijo
A target that only moves one way
The change behind Toledo's line was disclosed on June 15, 2026, as the third item in a three-part announcement whose other two items were friendlier: a promotion mode beta and a wider rollout of Smart Bidding Exploration. From August 17, campaigns marked "Limited by budget" that use Target CPA or Target ROAS across Search, Shopping, Performance Max, Demand Gen and Travel would "more consistently perform toward" their stated targets. Hotel and Display already worked that way. App, Video reach and Video view campaigns were left out. The change reached Search Ads 360, Display and Video 360, Google Ads Editor and the Google Ads API as well.
Google's own example explains the stakes better than any critic could. A campaign with a $10 Target CPA that has been converting at $5 will be pulled towards $10. A campaign set to a 200% Target ROAS and delivering around 400% will drift back towards 200%. Under the old behaviour, a budget-capped campaign could beat its target and keep the difference as efficiency. Under the new one, the target stops being a ceiling and becomes, in the words of Jarah Burke, VP client partner at the measurement firm Measured, something else: "After August 17, 2026, Google will treat that number as an instruction". Measured sells incrementality testing and media mix modelling, so its warning and its sales pitch point the same way, but the arithmetic does not depend on who sells what.
The asymmetry is the part that matters. On August 12, Greg Finn of Cypress North asked Ginny Marvin, Google's Ads Product Liaison, on the Marketing O'Clock podcast whether the change would help a campaign with a $100 Target CPA that was actually converting at $125. "Yeah. No, it won't," Marvin replied. The correction applies only when three conditions hold together: the campaign uses a target, is limited by budget, and is beating that target. Campaigns that overshoot on cost are left alone. Campaigns that undershoot on cost are pushed up. The same interview confirmed that targets set at ad group level fall inside the change.
Google's stated rationale is that targets should mean what they say. "We want advertisers to set targets that actually mean something to their business," Marvin wrote in July, and on August 5 she published a Q&A arguing that in budget-constrained campaigns "the target becomes the efficiency lever". An advertiser who resets the target to current performance, she wrote, should see similar conversion volume after a brief recalibration. One who does nothing may find the system entering different auctions from those it competed in before.
There is a coherent case here, and some practitioners accept it. Craig Graham, who works with ecommerce, lead generation and healthcare brands, wrote that he could see the value in matching target and performance more closely. On Reddit, a commenter described the old setup as "a hack that google patched out": a capped campaign with a low target spent its limited money on the cheapest, highest-return auctions available, so the target worked as a floor on auction quality rather than a goal.
But a Google sales slide, which Mike Ryan, head of ecommerce insights at Smarter Ecommerce, shared on LinkedIn on August 5, told the story from the other side. Its title read "Effect #2: Campaign can appear to overachieve (even while missing high quality demand)." Budget-constrained campaigns, the slide said, lower bids to stretch budgets and lose auction competitiveness. Read plainly, that is Google telling advertisers that a $5 cost per acquisition against a $10 target is a symptom of under-spending. Julie Friedman Bacchini, commenting in the same discussion, said Google's documentation kept steering advertisers towards a bigger budget, which she called a misunderstanding of how budget-limited campaigns work.
Here is where the two revenue lines separate. On the day the change lands, Google's revenue from a capped campaign does not move: the budget is the budget. What moves is what that budget buys. If a $5 conversion becomes a $10 conversion, the same spend returns half the conversions, and the advertiser who wants the old volume back has one obvious route, which is to raise the budget. Greg Finn predicted the next step in a July 2 recording with Barry Schwartz of Search Engine Roundtable, describing the change as arriving "under the guise of predictability and optimization" and saying "we are going to see increased CPCs." Marvin's answer, quoted in Search Engine Roundtable's first detailed report that same day, was that a capped campaign would not spend more unless its budget was increased. Both turned out to be right. On September 14, Mike Ryan posted the first charts, and Search Engine Roundtable summarised them the next day: cost per click was rising for budget-limited campaigns, while "a sizable pool of cheap clicks" had been released to campaigns that were not limited by budget. In Ryan's reading, Google had previously suppressed bids on capped campaigns to keep them inside their budgets, and August 17 ended the suppression. The capped advertiser now pays more per click for the same money. Its uncapped rival inherits the cheap clicks. Google, which sold the same budget either way, gains the strongest possible argument for the next budget increase.
Toni Poulain, senior vice president of media operations at Goodway Group, framed it as a principal-agent problem in an AdExchanger column on August 31. The change, he wrote, makes clear that "the entered target is not a ceiling that the platform will try to beat." A platform's automated agent balances advertiser returns, publisher yield and shareholder growth, and "it earns more when advertisers spend more." Poulain was careful to add that Google genuinely believes consistency helps advertisers, and that "the algorithm is not malfunctioning; it is doing exactly what it is being asked to do." That is precisely the issue. The question is who does the asking.
The rollout began on Monday, August 17, when Google started forcing overperforming CPAs up across five campaign types, and finished ten days later. Advertisers learned the end date only on September 15, when an updated Google help page was reported saying that "this global rollout was completed on August 27, 2026." Hana Kobzová at PPC News Feed spotted the edit, noting that campaigns were now trending towards their original targets if left unadjusted, and Barry Schwartz reported it on Search Engine Roundtable. A change that reprices conversions across five campaign types finished without an announcement, and its completion date surfaced through a help document.
The most vivid accounts from advertisers arrived in October. On October 5, PPC Land reported a Reddit thread titled "August 17 DESTROYED our Google Shopping performance!!!!" Its author ran a Shopping campaign with a 300% target ROAS that had been returning between 3,000% and 4,000%, ten times the target or more. After the change, sales fell. A Google representative suggested moving the target closer to actual delivery. Another advertiser, scottylebot, had two non-brand Shopping campaigns at 325% and 400% that had returned more than ten times; after raising targets on August 17, the best of them settled at 2.2 times on a 600% target. A third kept a 415% target, raised the budget by about 40%, and got higher spend, higher cost per click and roughly half the usual conversions. Sceptics in the thread pointed out that branded search may have inflated the original figures. Nobody disputed the direction.
Microsoft's answer to the same question is instructive, and not because Microsoft is a charity. In August, Navah Hopkins, Microsoft's Ads Liaison, said Microsoft Advertising "continues to allow campaigns to over achieve on TCPA/TROAS regardless of budget limited status". The statement came in a discussion of Microsoft removing Max CPC from new non-portfolio campaigns from October 1, a reminder that every platform is pushing buyers towards automated bidding. What differs is the treatment of the gap between a target and a result. One platform lets the advertiser keep it. The other takes it back. Microsoft's own search advertising revenue excluding traffic acquisition costs grew 10% in its fiscal fourth quarter to June 30, 2026, the second deceleration in three quarters, so it has every commercial reason to make that distinction visible.
The price of a click, and a court that looked inside the auction
A sceptic might say all this is anecdote, and that the real test of whether Google's interest and the advertiser's interest have parted is price. On that test the record is long, and some of it was written under oath.
Start with the benchmarks. WordStream's 2026 figures, drawn from 13,474 US search campaigns on Google and Microsoft between April 2025 and March 2026, put the median cost per click at $5.42, against $2.32 in 2016, an increase of about 134% in a decade. The same data carries the strongest counter-argument in this piece. Click-through and conversion rates each roughly tripled over the period, and cost per lead fell to $66.69, the first year-on-year decline in about five years. Advertisers who track conversions properly have been paying more per click without paying proportionally more per lead. A year earlier, LocaliQ's 2025 benchmark had found cost per click up 12.88% to $5.26, with 87% of industries paying more, and Katia Hausman, LocaliQ's vice president of paid media products, said the sharper increases came on campaigns using smart bidding. In April 2024, MediaPost reported Tinuiti's first-quarter benchmark, which found Google search cost per click up 13% year on year while average order values rose about 1% and clicks grew 4%. The visit cost more; the basket at the end of it barely changed.
European ecommerce looks harsher. Channable's benchmark, built on 1.38 billion euros of Google Ads spend by more than 10,000 merchants and published on July 12, found that between June 2025 and June 2026 the cost per click on Shopping and Performance Max rose 15% while return on ad spend fell 46% on Performance Max and 43% on standard Shopping. Channable is a feed management vendor, and its data excludes Search. Another vendor, Billy Grace, found the opposite trend in a matched cohort of several hundred European advertisers: median cross-channel cost per click fell 10.8% in the first half of 2026, while Google's share of their spend dropped from 62.1% to 57.3%. Tinuiti's aggregate data, reported by MediaPost on July 16, points the same way at headline level: in the second quarter of 2026, spend on Google Search grew nearly 14% while total Google Search cost-per-click growth was 1%, against 19% on Microsoft. Google's search revenue growth, on that evidence, is coming more from volume, new queries and campaign types than from a broad rise in the price of a click. Prices do not only go up, and they move with who turns up to bid. When Amazon left Google Shopping auctions in July 2025, PPC Land reported cost-per-click falls of 25% to 30%. When JD.com's Joybuy began buying Shopping ads in Europe in March 2026, it was competing against more than 10% of 510 monitored accounts within 31 days, and by the end of the month it was being outranked less than half the time, down from a median of 80%. The auctioneer collects either way.
Benchmarks cannot say why prices move. A courtroom can. In September 2023, Jerry Dischler, then vice president and general manager of Google Ads, testified in the Justice Department's search case that Google had tuned its auctions in ways that raised prices, at times by 5% to 10%, to meet revenue targets. "We tend not to tell advertisers about pricing changes," he said, as Search Engine Roundtable reported on September 20, 2023. "My goal was to get creative so we could meet our quota." MediaPost reported the testimony a day earlier: auction and reserve price adjustments of as much as 5% on average and up to 10% for some queries, and a May 2019 effort by Dischler's team to meet quarterly revenue targets set by the chief financial officer, Ruth Porat. Digiday quoted his 2019 email: "How badly do we want to hit our numbers this quarter?" On the stand, Dischler said a 15% increase would be "a dangerous thing to do," with Meta and TikTok waiting. Buyers interviewed by Adweek dismissed the 5% figure as too low. AdExchanger's daily roundup on September 20 ran under the headline "Google Admits To Search Auction Gouging," and noted that Dischler cited Amazon and TikTok gaining share as Google's main worry, which AdExchanger read as the only brake on further increases. Google's defence then and since has been consistent: prices are set by a real-time auction and no advertiser pays more than its maximum bid.
Judge Amit Mehta's August 2024 liability ruling found that Google had exercised its monopoly power by "charging supracompetitive prices for general search text ads", and that this had allowed it to earn monopoly profits. AdExchanger's reading of the 286-page opinion picked out the line that "Google has increased text ads prices without fear of losing advertisers," and the court's view that the only apparent limits were "potential advertiser outcry and bad publicity." Advertisers testified that they move money in search of return, yet none said they had significantly shifted spend away from search. Mehta's 230-page remedies opinion of September 2, 2025 went further on mechanics. It found that Google had raised text ad prices through what PPC Land described as "pricing knobs", in increments small enough to sit within the ordinary noise of the auction, so that advertisers who noticed prices rising did not attribute the increases to Google. "Google took care to avoid blowback from advertisers," the court wrote. The remedy was disclosure: Google must publicly disclose material changes to its ad auctions. Digiday's list of what was in and out summed it up as Google having to disclose major auction changes to limit hidden cost increases. What the court refused matters as much. Mehta declined to restore a true exact match option, which Google discontinued in 2014, and declined to order query-level data for advertisers. The industry tried a voluntary route as well. When the Media Rating Council finalised standards for ad auction transparency in January 2026, covering search, social and retail among other formats, MediaPost noted the omission of Google's participation.
The final judgment, entered on December 5, 2025 for six years, requires Google to disclose changes to its search advertising algorithms 31 days before it implements them. Google appealed on January 16, 2026, arguing that the notice period constrains its ability to respond quickly, and on May 22 it asked the appeals court to throw out its entire search monopoly loss. The appeal is pending. That context makes the August 17 change interesting in a way that has nothing to do with whether it is good or bad bidding policy. It was disclosed on June 15, about nine weeks before it took effect, comfortably more than 31 days. Whether or not it counts as a material auction change under the decree, the disclosure came inside an announcement that led with other features, and the completion arrived as a help page edit. Disclosure, it turns out, is not the same thing as attention, and a change can be published without being noticed.
The commercial and regulatory pressure has not waited for the appeal. On September 12, 2025, the Federal Trade Commission opened inquiries into Amazon and Google over whether they misled advertisers about ad pricing and terms, with the Google side covering its internal pricing processes and whether it raised costs without proper disclosure. In February 2026, AdExchanger reported a new EU antitrust probe into allegations that Google had rigged search ad auctions by "artificially increasing the clearing price"; a Google spokesperson replied that ad prices are determined by a real-time auction. And on May 11, 2026, the Chicago firm Keller Postman launched a mass arbitration campaign on behalf of thousands of US businesses that bought Google search or display ads since August 2016, citing about $728 billion of spend and a damages theory under which a 10% overcharge, trebled, would return 30% of spend. Advertisers must bring those claims one by one because Google's own contracts bar class actions. Three weeks later, PPC News Feed reported that Google had updated its Google Ads Terms of Service for the first time in eight years, and that the arbitration agreement now allows batch arbitration for similar claims, alongside adjusted liability caps and new fee disclosures. The overcharge claims themselves remain unproven.
Amazon's case shows what such an inquiry can become. On August 31, 2026, the FTC and 22 states sued Amazon in Seattle, alleging that it ran a generalised second-price auction and then applied an undisclosed "soft reserve" on top of the result, so that the share of clicks charged at the advertiser's own bid rose from 4% in late 2020 to 79.1% in 2024. Among the names the complaint says Amazon considered for what it alleges was a surcharge was "profit extraction lever"; the team settled on "performance premium." Amazon's reply was that "in no scenario does an advertiser pay more than their bid," that Sponsored Products cost per click stayed flat in inflation-adjusted terms between 2019 and 2024, and that relevance-weighted auctions saved advertisers more than $8 billion between 2021 and 2025. Both Amazon and Google rest their defence on the same sentence about maximum bids. It is true, and it is beside the point. A maximum bid is a ceiling, and the argument has always been about how close to the ceiling the auctioneer chooses to charge.
Defaults that drift towards more inventory
Bidding is the most visible dial. It is not the only one, and the pattern repeats with uncomfortable regularity: a control the advertiser once set becomes a suggestion the system may override, and the default points towards more reach, more matching and more inventory.
Take match types. On August 5, Google emailed advertisers to say that Search campaigns using automatically created assets or the campaign-level broad match setting would be upgraded to AI Max for Search on September 1, with search term matching switched on by default. The email was signed by the "Google Ads Team", with no named executive and no blog post. Search Engine Roundtable published the same notice on August 7. Avoiding the upgrade meant switching off the legacy features before the date. One Reddit user called it "basically just a rebrand and a shove into a new bucket."
What that matching does to an account is now measurable, at least by people who are not Google. Mike Ryan's analysis of 383 million impressions from EMEA ecommerce Search campaigns between January 2025 and July 2026 found that exact match lost about one impression in four to AI Max: by July 2026, roughly 29% of what advertisers had bought as exact match was being broadened by AI Max search term matching, and about two-thirds of that expansion happened in the final four months. Smarter Ecommerce's earlier study of more than 250 retail campaigns, published in November 2025, found AI Max delivering roughly 35% lower return on ad spend than traditional match types. Xavier Mantica's four-month test put AI Max at $100.37 per conversion, against $52.69 for exact and $43.97 for phrase. Ezra Sackett, director of paid search at Monks, found that 99% of AI Max impressions across about 30,000 search terms produced no conversions in August 2025. A year after launch, buyers told Digiday that AI Max had pushed up search budgets and costs: Mediaplus clients had seen cost per click rise 10% to 15% on average, and Collective Measures reported increases around 10%, with some clients hit by as much as 25%. Part of that is competition rather than pricing, since Adthena estimated the number of advertisers in search auctions had grown 35% year on year. "The CPC pain is real," said Ashley Fletcher, Adthena's chief marketing officer.
Google's own figures have moved in the meantime. AI Max launched on May 6, 2025 with a claim of 14% more conversions or conversion value at similar cost. By the April 2026 general availability announcement the figure was 7%, on Google's internal 2026 data excluding retail. A Google holiday guide later used 15%, under a different definition. On Alphabet's July 22 earnings call, chief business officer Philipp Schindler cited a 15% average uplift for AI-powered campaigns, prompting Lucas Mota, a senior paid media specialist, to write that "15% with the same ROAS is a crazy commercial claim to make."
The evidence does not all point one way. Etsy reported a 36% year-on-year increase in gross merchandise sales during the holidays, partly credited to AI Max, together with a 21% rise in new buyers. That figure came from Google's own retail holiday guide, however, and PPC Land observed that none of the three case studies in it carried a measurement window, a control condition or a statement of incrementality method. A success story written by the seller is still a success story. It is not a test.
Broader matching also brings traffic nobody wants to pay for. Lunio, which sells invalid-traffic detection and drew on its own client data, examined more than 414 million clicks and found that invalid traffic in retail Search campaigns using AI Max rose from 2.46% to 5.28% between the fourth quarter of 2025 and the second quarter of 2026, while standard retail Search eased from 3.72% to 3.07%. AI Max accounted for 68% of the invalid clicks Lunio detected. The study is a vendor's benchmark and its revenue arithmetic does not fully reconcile, but it adds to a body of independent evidence that does not flatter AI Max. Who decides what counts as invalid, and what gets refunded? Google does. Its Invalid Activity Credit Report, documented for the first time in June 2026, covers only Search and Performance Max. Display, Video and Demand Gen are excluded, the detection criteria are unpublished, and credits are issued "where appropriate and possible."
The same drift shows up in audiences. In February 2026, PPC Land reported that Google would turn Demand Gen Lookalike segments into suggestions from March. The narrow, balanced and broad tiers had capped reach at about 2.5%, 5% and 10% of users in a location; afterwards, the system "may go beyond those parameters if its models predict better conversion performance." Restoring the hard limit required a form, with removal taking about a week. Adriaan Dekker, a Google Ads consultant, read the form as an admission: Google knows most advertisers would not choose full expansion by default.
It shows up in reporting, too. Since May, the search terms report has shown, for queries from Lens, AI Mode, AI Overviews and autocomplete, Google's "best approximation of the user's intent" rather than the words a person typed. Ginny Marvin had explained an earlier AI Max reporting anomaly found by Brad Geddes, co-founder of Adalysis, by saying Google was "increasingly determining relevance by inferred intent". The system that chooses the query now also describes it.
And it shows up in creative. From October 12, Google Ads will scrape discount offers from advertisers' own websites and attach them as promotion assets to some Search and Performance Max campaigns, "without requiring any manual setup." The switch is on by default and sits at account level. PPC Land pointed out that a discount claim in an ad is a legal representation as well as a marketing one. Liability, as it happens, has been moving in the opposite direction from control. When Google's fully automated tools generate and serve creative, Google applies an AI label that "cannot be overwritten or removed," yet its documentation leaves compliance with rules such as Article 50 of the EU AI Act, with penalties of up to 15 million euros or 3% of global turnover, with the advertiser.
Some controls do come back, usually under pressure. In July, Digiday reported that Google had quietly given ground on Performance Max controls, letting selected advertisers in an alpha deselect the Search Partner Network and the Display Network, both of which had been mandatory. Buyers at three agencies were refused access by their Google representatives. "Search advertisers don't like the Display Network. We don't want to opt in," Kyle Rovinski, associate director of search at Duncan Channon, told Digiday. PPC Land's earlier coverage had put Search Partner Network return on ad spend 37% below Google Search proper. The network's scale has long been disputed. In November 2023, Adalytics estimated that the Search Partner Network generates roughly $10.5 billion a year for Google and found its search embed code on more than 80,000 sites, over 20,000 of them parked domains, where Google's documentation spoke of hundreds of non-Google websites. Dan Taylor, a Google Ads vice president, said Adalytics had a track record of publishing inaccurate reports. Six weeks earlier, one buyer told Adweek that around 80% of a Performance Max campaign's impressions had run on open-web inventory rather than on Google's own properties. The default still includes both networks.
When the rep calls
Defaults do much of the work. People do the rest. In November 2024, Ian Harris convened 50 PPC agency leaders who documented Google representatives contacting clients directly, threatening to go around agencies, and pressing agencies to "Switch on broad match" and "Increase spend in Performance Max." Andy Youngs, co-founder of The PPC People, described a representative rewriting ad copy, changing pinning and switching a bidding strategy without authorisation, in changes that did not appear in the account's change history. Pete Fairburn summarised the mood in a sentence: "Google's driving force is to make money for Google." Marvin acknowledged that Google had not properly followed its processes for approvals. In August 2025, a growth marketer who had moved every client off Performance Max told Digiday that the first incrementality test on one large brand came back below 10%, that return on ad spend rose about 20% after clients moved back to standard Shopping, and that a Google representative had admitted: "There were bonuses for the staff who could get their clients on PMax." Google replied that more than a million advertisers use Performance Max and that it shipped more than 90 quality improvements in 2024.
The friction has since been formalised rather than removed. In February 2026, a checkbox on Google's support contact form was found to authorise a Google specialist to change an account with no further confirmation, while the advertiser remains "solely responsible" for the impact on performance and spend. In May, Emmanuel Flossie, a Google Shopping specialist and Google Ads Diamond Product Expert, accused Google's sales representatives of sending commercial email with no opt-out, in breach, he argued, of the CAN-SPAM Act of 2003. "Compliance with a 20-year-old federal law shouldn't be optional," he wrote.
None of this is new in spirit. In November 2022, Search Engine Roundtable reported that Google was offering some advertisers a $100 credit to enrol in automatically applied recommendations. The recommendations themselves remain unpersuasive to many buyers. A StackAdapt survey of 500 marketers, run by NewtonX between May 19 and June 8, 2026, found that only 6% act on in-platform AI recommendations almost always, and 12% said recommendations feel timed to drive spend rather than performance. StackAdapt sells an AI advertising platform that competes with in-platform tools, which is worth weighing. So is the direction of Google's own tools. In September, PPC News Feed reported that Google's Recommended Investment Strategy had gained a Growth mode that "strictly adds budget to constrained campaigns", next to a Holistic mode that reallocates budgets and adds weekly spend, with an "Apply all" option that puts changes live at once.
The pattern extends beyond Search. From October 1, Local Services Ads began charging for missed business-hours calls where the caller stays on the line for more than 20 seconds. Google framed the change as "rewarding businesses that provide excellent responsiveness." Some practitioners agreed that speed to lead matters. Others, such as Blastoff Ads, pointed out that many missed calls roll into voicemail because the business is already on the phone, and PPC Land noted that the policy said nothing on voicemail handling, dispute steps or geographic scope. A missed call is, by definition, a lead that produced no conversation. From October 1 it also produces a charge.
Who marks the homework
An advertiser's revenue is measured in its own books. Google's view of that revenue is measured in Google's tools, and the gap between the two is where much of the argument now sits. Measured put it bluntly in its August analysis: "Platform-attributed conversions and incremental business impact are not the same thing." The firm sells incrementality testing, as noted, yet the sentence would survive any audit. Matt Prohaska of Prohaska Consulting put the industry's frustration more sharply when advertisers pushed for auction standards in 2024, describing the status quo to Digiday as "letting three to five companies grade their homework incorrectly and inappropriately."
Consider what happened to the counting itself during 2026. At Google Marketing Live on May 20, Google introduced Qualified Future Conversions, which PPC Land described in September as a metric that counts conversions up to 180 days after an ad click. A qualifying action within seven days of a click, such as an attributed branded search, a site visit or an add-to-cart, licenses the counting of conversions over 30, 90 or 180 days. John Chen, Google's senior director of product management for ads measurement, said of the older lookback logic: "We think that's probably a little bit too arbitrary." In one example Google showed at the launch, the new metric produced 24,680 qualified future conversions against 20,854 standard conversions. The same day, Google placed its open-source marketing mix model, Meridian, inside Analytics 360. PPC Land noted that feeding Gemini signals drawn from Google Ads data into a model that also evaluates Google Ads spend "raises questions about circularity and what measurement professionals call prior contamination," and that Meridian is built by the company that also sells a large share of the media it evaluates.
The concern is not abstract. In August, a preprint by Niklas Heusch, whose address points to Zalando, used a simulated retailer over 156 weeks to show that a standard marketing mix model overstated paid search returns by about 2.5 times. It read 10.61 times return on ad spend where the true figure was 4.20. Changing the priors moved the estimate to 11.38 times; a structural method built on geo-experiments brought it back to between 4.14 and 4.31. The results are simulated and the paper does not test Google's product, but PPC Land's summary of the error is hard to improve on: "The observational estimates are wrong in the same direction, which is the direction that costs money."
Sometimes the value is assigned outright. In September 2025, PPC News Feed reported that Google would switch on store visits as primary conversions from October 8, with values set by Google, an arrangement critics said was making reports show inflated conversions without real sales values. Several of Google's 2026 measurement launches share a feature. They ask advertisers to hand over more data, and then report the value of that data with a number only Google can compute. The Data Strength Uplift metric, announced on September 10, calculates the conversions an advertiser's first-party data setup recovers, with no formula or baseline published. PPC Land called it "a number an advertiser cannot independently audit." Elsewhere, the plumbing quietly drops what does not fit. Google's daily attribution engine processes only the last seven days, so offline conversions uploaded later are bypassed by the attribution modelling, even though standard columns still count them. Because data-driven attribution feeds Smart Bidding, a late B2B upload never reaches the model that sets the bids.
Billing has its own version of the problem. Display impressions in Campaign Manager 360 and Display and Video 360 have been counted on an "on-download" trigger, which fires once the ad code runs even if the creative never loads. In February 2027, Google will move billing to "begin to render", the standard the Media Rating Council's display guidelines required in October 2017. Google expects "a slight decrease in overall billed impression volume" and has not put a figure on it. An August 2025 notice had scheduled the same switch for September 2025.
To be fair to Google, some of its measurement moves have made independent checking cheaper. At Google Marketing Live in May 2025 it cut the minimum budget for incrementality experiments to $5,000, from thresholds previously near $100,000. That is a genuine concession. The test still runs on Google's platform, with Google's priors, and is read by Google's software. An incrementality test designed by the seller is better than no test. It is not the same as an audit.
Fewer clicks, and the inventory nobody can switch off
All of this is happening while the shape of the results page changes underneath it. Ads began appearing inside AI Overviews on October 3, 2024, for mobile users in the United States, and existing Search and Shopping campaigns became eligible automatically. In May 2025, Google's own help document, as Search Engine Roundtable reported, stated that advertisers could not opt out of serving ads in AI Overviews and that Google Ads did not offer segmented reporting for them. When the format reached 11 more countries in December 2025, it did so through a help page update, and those ads are reported as "Top Ads" with no separate segment. AI Mode followed the same path. Shopping ads there were formally announced on February 11, 2026, and in September PPC Land found that Google Ads offers no segment that isolates AI Mode delivery; impressions and clicks land in standard campaign rows. "Clearly, Google really wants searchers to click on these ads in AI Mode," Barry Schwartz observed of an undocumented test.
What do those placements do to click-through? Seer Interactive's study of 3,119 informational queries across 42 organisations, published in November 2025, found that paid click-through rate on queries with an AI Overview fell from 19.7% to 6.34%, a drop of 68%. The paid sample was small and the study measured clicks, not conversions or cost. Google's position, set out by Philipp Schindler in April 2025, is that ads in AI Overviews monetise "at approximately the same rate" as ordinary search ads.
On Google's side of the ledger, that position looks well founded. Alphabet's second-quarter results on July 22 showed Search and other revenue up 17% to $63.27 billion, against $54.19 billion a year earlier. Total Google advertising revenue rose 14% to $81.6 billion, Alphabet's operating margin reached 34%, and Alphabet raised its 2026 capital expenditure guidance to between $195 billion and $205 billion. Google Network revenue, the part that pays publishers, fell 1% to $7.30 billion. Schindler credited Search growth to "very, very deep Gemini integration." Revenue rising 17% while paid click-through on AI Overview queries falls by two-thirds is not a contradiction. It means each remaining click, or each impression, is earning more, or that volume elsewhere is making up the difference. Either way, the price per unit of attention has not been falling on Google's side.
On the advertiser's side the picture is murkier. In May, Google confirmed at Marketing Live that eligibility for the new AI Mode ad formats, Conversational Discovery ads, Highlighted Answers and Direct Offers among them, requires AI Max for Search or Performance Max with text customisation. In September, Google began a small test serving exact and phrase match keywords in AI Mode, but AI Max and Performance Max kept exclusive access to Highlighted Answers. The newest, scarcest inventory is reserved for the campaign types that give Google the most latitude over queries, creative and landing pages, and its results are blended into reports that cannot separate them.
Scarcity is the right word. A study by Productrise, which sells tracking tools, compared more than two million product listings in the US and UK in August and found that AI Mode showed 3.9 products per response against 27.8 in traditional search. Only 1.28% of products in traditional results also appeared in AI Mode on the same day, and AI Mode's lead offer was 21.6% more expensive on matched products. Fewer slots on a page means fewer winners per query, and an auction with fewer winners tends to clear at a higher price. For the merchant that wins, that may be fine. For the two dozen or so merchants who used to appear and no longer do, the alternative is to pay for a sponsored slot or to vanish.
The same playbook, at every auction house
It would be convenient to treat all of this as a Google story. It is not. The incentive belongs to the business model, and every company that runs an ad auction and sells automation on top of it has been moving in a similar direction during 2026.
Meta is the closest parallel. In late August, Bram Van der Hallen of Edge.be reported that the placements control was disappearing from some Ads Manager accounts. Its substitute, value rules, can cut bids by up to 90% but cannot block a placement outright, which matters to any brand that wants nothing to do with Audience Network. The change followed the removal of detailed targeting exclusions in January 2025, a default from October 2025 allowing up to 5% of budget to reach excluded placements, and an Advantage+ structure that has barred campaign-level placement exclusions since February 2026. Meta says Advantage+ placements deliver an 11.7% lower cost per acquisition on average, without disclosing the sample. Its own results show where the gains land: in the second quarter of 2026, ad revenue rose 27% to $59.36 billion and the average price per ad rose 12%, as it had in the first quarter. In August 2025, a former Meta product manager, Samujjal Purkayastha, alleged in a UK employment tribunal complaint that internal reviews had found Shops ads return on ad spend inflated by 17% to 19% because it counted shipping and taxes merchants never received. Meta declined to comment, and the tribunal has made no finding.
The newest entrant copied the defaults within months. OpenAI's ChatGPT ads, which launched as a pilot on February 9, 2026, reached a $1 billion annualised advertising run-rate in under 200 days. By August, new ad groups had "Maximize results" preselected, with no cost-per-acquisition, cost-per-click or return guarantee, and the only hard ceiling, a manual maximum bid, had to be chosen ad group by ad group. From August 17, the same day Google's bidding change began, OpenAI switched on automatic advanced matching for existing pixels, giving advertisers about ten days' notice to opt out of sending hashed form data with conversion events. Reporting, meanwhile, still splits results into two device buckets while targeting offers three platforms.
Apple's auction is opaque in a different way. Mike Rhodes of ConsultMyApp scored 627 keyword-app observations from UK Apple Ads auctions for relevance and found that the most relevant app held the top position on only 43.9% of keywords, and that relevance order matched auction order on 7.6%. Apple does not publish its ranking formula. On one keyword, iTranslate, scored 100 out of 100 for "translator," sat in fifth place.
On the open web, the toll is set by the exchange. Judge Leonie Brinkema found in April 2025 that Google had monopolised the publisher ad server and ad exchange markets, and PPC Land reported that the court considered AdX's fee of around 20% supracompetitive for more than a decade. Her remedies decision, unsealed in September 2026, imposed a six-year worldwide decree that bans first look and last look and opens AdX to rival ad servers, but left the 20% take rate, against roughly 10% at rivals, without a cap. Scott Messer of Messer Media counted 16 of Google's own 19 proposed fixes adopted as written. Regulators rewrote the rules of the auction. The price of running through it stayed where it was.
Microsoft is the useful exception, partly because it is also the challenger. In December 2025, Navah Hopkins said an exact match keyword in a Microsoft Search campaign will always beat Performance Max for the same query, regardless of Ad Rank. A month earlier, an exact match keyword in Google, "best hypoallergenic food for dogs," had matched 13 queries that did not contain the word "hypoallergenic", all labelled as close variants. Brian Lasonde, founder of PPC Boost, posted the example. An anonymous commenter added: "We all know who benefits from this." Microsoft is still removing manual controls of its own, as its Max CPC change shows, and it is no longer the cheap option: Tinuiti's second-quarter data put its cost per click up 19% year on year. But its choices on exact match and on target over-achievement demonstrate that the alternative is a product decision, not a law of machine learning.
What the other side of the auction is optimising
If Google optimises for Google, what are advertisers doing about it? The answer, on the evidence of 2026, is that they are rebuilding the controls the platforms removed, one workaround at a time, and paying third parties to do it.
Some of the responses are mechanical, and the Reddit thread on the August 17 change catalogues most of them. One agency moved clients to their actual return on ad spend in phases. Another user, scottylebot, raised targets and added a low-priority Maximize Conversion Value campaign to pick up the auctions the older campaigns skipped. A third, flimflambam, switched to Maximize Conversion Value and reported the account was "back to crushing." The original poster tried both Maximize strategies and still found cost per click and Shopping visibility below where they had been before August 17. Each of those moves is an adjustment to Google's terms rather than an escape from them. Others are moving money. Billy Grace's European cohort, the same one that showed Google's share of spend falling from 62.1% to 57.3% in the first half of 2026, also showed spend on Reddit up 650%, on TikTok up 65% and on Bing up 34%. Mitch Voskuilen, its co-founder and chief executive, said the advertisers who did best "judged results on the full picture rather than whichever number a platform hands back."
A whole layer of tools now exists to stand between the advertiser and the platform's judgment. Adthena's Decision Intelligence, available from September 16, combines an advertiser's own account data with Adthena's auction tracking to rank options for cutting rising search costs, and stresses that every account change is made by the marketer. Fluency, which manages about $3 billion in annual ad spend, built its system so that large language models never touch live budget execution; only deterministic agents, running rules the advertiser sets in advance, can move money. Incrementality vendors, invalid-traffic filters and independent mix models make up the rest. Each of them has something to sell. Each of them also exists because the party that sells the media, runs the auction, writes the bidding software and grades the results has a stake in the answer.
Then there is the slower route. The FTC's inquiry into Google's ad pricing, opened in September 2025, had produced no filing against Google when PPC Land reported the Amazon suit on August 31. Keller Postman's arbitration campaign is gathering claimants one contract at a time. The 31-day notice requirement in Mehta's judgment is under appeal. None of these will reprice a click this quarter.
And so the argument returns to Daniel Toledo's line. Quality Score, in 2005, was Google's way of saying that the best ad, not merely the richest bidder, should win. It was never a pure meritocracy, and Google always decided what "best" meant. But the principle pointed in a direction advertisers could share: make a better ad, pay less. The August 17 change points somewhere else. A campaign that delivers a $5 acquisition against a $10 target is no longer treated as efficient; Google's own sales material describes it as missing demand. The fix offered is a higher target or a bigger budget. When Ginny Marvin was asked whether the same logic would help a campaign missing its target, the answer was four words long.
That does not make Google uniquely cynical. Amazon is defending a lawsuit over what it charged on top of its auction. Meta is converting placement controls into bid adjustments. OpenAI preselected automated bidding before its ad platform was a year old. Microsoft lets its campaigns keep the gap between target and result, and is removing Max CPC anyway. Every auction house is optimising for its own revenue, as it is entitled to do. The advertiser's problem is that the house also owns the gavel, the catalogue and the ledger, and in 2026 it has been taking the remaining dials off the bidder's side of the table.
Mike Ryan had promised data 30 to 60 days after the change, and his first charts arrived on September 14. They showed what the change was designed to do and what its critics feared, both at once: dearer clicks for capped campaigns, cheaper ones for everybody else, and the same budget flowing to the same auctioneer. Val Riley, vice president of marketing and strategy at Unbounce and Insightly, said it plainly in a MediaPost interview about AI Max in August 2025: "It's up to the marketer to judge the performance of the platform." Google optimises its revenue. Advertisers have to optimise theirs, and increasingly they have to do it with instruments Google does not make.
By the numbers
- $63.27 billion: Google Search and other revenue in the second quarter of 2026, a 17% rise on a year earlier. PPC Land
- 46%: Drop in Performance Max return on ad spend between June 2025 and June 2026, across 1.38 billion euros. PPC Land
- 383 million: EMEA Search impressions studied by smec; by July 2026, AI Max was broadening about 29% of exact match. PPC Land
- August 17, 2026: Day Google began pulling budget-capped campaigns that beat their targets up towards those targets. PPC Land
- 79.1%: Amazon clicks billed at the advertiser's own bid in 2024, against 4% in late 2020, the FTC alleges. PPC Land
Discussion