A single regulator moved the ground under the search business this week, and it did so twice in one day. On July 16, 2026, the European Commission adopted two binding decisions ordering Alphabet's Google to open parts of the Android operating system to competing artificial intelligence assistants and to share anonymised Google Search data with rival search engines, remedies the Commission says became necessary after Google's own compliance proposals proved ineffective. The details reached the trade press over the weekend, and by Monday morning they framed almost everything else on the agenda: a Swedish damages award tied to the same self-preferencing question, a United States appeals court redrawing the line between pricing and coercion, and a measurement upheaval landing weeks before the autumn television season. The connective thread runs through market power and the question of who gets to check it.

Two rulings, one law, and a 90 percent market

Both decisions rest on the Digital Markets Act, the regulation that imposes behavioural obligations on the largest platforms, the ones Brussels calls gatekeepers. Google has held that designation since September 6, 2023, when the Commission named eight of its services, among them Search, Play, Maps, YouTube, Android, Chrome, Shopping and its advertising business, as core platform services, with full compliance mandatory from March 7, 2024. According to the Commission's guidance cited by PPC Land, the two July 16 decisions draw on separate provisions: Article 6(7), which requires operating-system gatekeepers to grant developers effective interoperability with hardware and software features, and Article 6(11), which requires search gatekeepers to share anonymised data with eligible rivals on fair, reasonable and non-discriminatory terms, the standard shortened to FRAND.

Why act now, and why through this particular mechanism? Both cases opened on January 27, 2026, when the Commission began what the law calls specification proceedings, a formal route that lets regulators define compliance in granular technical terms rather than leaving a gatekeeper to interpret a broad duty on its own. The Commission published preliminary findings on April 16 in the search-data case and April 27 in the Android matter, then adopted final versions three months later. One point shapes how the news should be read: specification proceedings are separate from non-compliance investigations and carry no financial penalty, a sharp contrast with the parallel cases that do.

The scale behind the search-data order is the reason it exists. Google Search has held above 90 percent of the EU market for decades, according to the Commission's guidance, a figure echoed by a University of Antwerp analysis finding that DMA remedies shifted roughly six million EU users to Firefox while leaving Google near 90 percent dominant. Article 6(11) already obliged Google to share query, click, ranking and view data. According to the Commission, Alphabet's first compliance offer stripped out between 90 and 100 percent of unique search queries and excluded AI chatbots from the list of eligible recipients, producing what regulators described bluntly as no meaningful uptake. PPC Land had reported the Commission's preliminary findings in that case in April, and before that the open letter from 18 European trade groups in March urging Brussels to act on Google's search compliance.

What actually changes, and when

The mechanics reward close reading, because the calendar stretches well past the headlines. On the Android side, the decision names 11 features that must become interoperable, grouped into four categories: invocation, which covers how a person starts an interaction with an assistant; context, an assistant's ability to draw on app, sensor or screen data; actions on apps and the operating system; and access to resources, including Google's own on-device AI models. In practice, a person could summon a rival assistant by voice much as they now say "Hey Google," delegate a task such as booking a taxi, or receive a proactive suggestion such as a flight number pulled from an email. The asymmetry being corrected is concrete. Gemini, Google's own assistant, has enjoyed comparatively unrestricted access to these functions, an imbalance PPC Land documented in July 2025 when Gemini gained automatic Android access to phone dialling and messaging data that competitors could not obtain by equivalent means.

The deadlines run into 2028. Google must build these capabilities into Android 18 and finish by August 1, 2027, while a more demanding requirement, concurrent hotword detection, which lets several assistants listen for their own wake words at once, carries a longer runway of Android 19 by August 1, 2028. For especially sensitive functions such as screen automation and centralised on-device data access, Google may impose objective eligibility conditions verified by independent certifiers; it must publish draft terms by February 1, 2027, and begin accepting applications from May 1, with each assessment completed within four weeks.

The search-data schedule is tighter and, for advertisers, more immediately consequential. The final measures require Google to share ranking, query, click and view data from both free and paid results, covering the queries typed, metadata such as language and device type, the URLs viewed, how users interacted with results, and where each result sat on the page. Several categories are excluded to protect individuals: no account information or search histories, no precise timestamps, no rare-word queries that could identify a searcher, generalised rather than exact location, and no URLs tied to paid results, meaning advertising clicks are stripped from the shared set. Recipients face use limits too. The data cannot train general-purpose AI models, build advertising or consumer-profiling tools, or systematically reproduce Google's results. Eligibility is confined to companies running a genuine EU search engine for at least two consecutive years, or founded within two years but having raised more than 50 million euros in capital alongside at least 50,000 monthly EU users. Pricing follows a cost-recovery formula rather than a market rate, and the rollout is fixed: an application form and information page by the end of August 2026, template licences and cost estimates by September, the finalised dataset by November, and final pricing by January 2027, the same month data sharing is due to begin.

Google's response arrived the same day. According to the article, Kent Walker, President of Global Affairs at Google and Alphabet, wrote that the decisions "risk undermining vital privacy and security guardrails for millions of Europeans," and warned on the search order that "Europeans' private searches would be exposed to unfamiliar companies, without adequate anonymisation." Teresa Ribera, the Commission's Executive Vice-President for the Clean, Just and Competitive Transition, framed the package differently, saying the goal was to help "smaller competitors, search engines, or AI assistants, to compete and provide that choice, while protecting the user's privacy." Both decisions state they remain open to review by EU courts, an avenue Google has used before.

There is a practical wrinkle worth naming. For search advertisers, the live question is which companies benefit first rather than which are legally entitled to. According to the analysis PPC Land tracked, DMA contestability measures have so far done more for Microsoft's Bing and Mozilla's Firefox than for European challengers, and that pattern could repeat with data access, since Bing and DuckDuckGo already have the infrastructure to exploit new rights immediately while Ecosia and Qwant would need to build it. The Android decision also sits beside a related precedent: in June 2026 the Commission used interim antitrust measures to order Meta to restore free access to the WhatsApp Business API for competing AI assistants, a sign that regulators increasingly treat reach through a dominant assistant as a competition concern in its own right.

Stockholm puts a number on a decade-old question

If Brussels defined the remedy this week, a Stockholm courtroom tested whether an earlier remedy ever worked, and concluded it did not. On July 1, 2026, the Patent and Market Court ordered Google and Alphabet to pay Klarna Technologies damages for diverting search traffic away from the comparison-shopping service formerly known as PriceRunner, ruling that the changes Google made in 2017 to satisfy a Commission order never actually ended the abuse the Commission had condemned. The judgment in Case PMT 1860-22 fixes principal damages at 950,000,000 British pounds for the United Kingdom, 675,000,000 Danish kroner for Denmark, and 1,142,000,000 SEK for Sweden, with Google and its Swedish arm jointly liable and Alphabet liable alongside them for harm from October 2, 2015 onward. Accrued interest through October 31, 2025 adds substantially to the bill; on the British portion alone the court set interest at 316,226,468 pounds, lifting the pounds-denominated award past 1.26 billion.

The case turns on the Shopping Unit, the boxed carousel at the top of the results page. Google argued the box was not its comparison-shopping service at all, so letting rivals bid within it satisfied equal treatment. According to the judgment, the court rejected that reasoning, holding that the Shopping Unit as it functioned after the 2017 changes was to be regarded as Google's comparison-shopping service, which made its favourable placement continued self-preferencing. The ruling devotes close attention to two ranking systems, Algorithm A, built by Google's web-spam team and launched in 2004, and the Panda algorithm, which reached the United Kingdom on April 11, 2011. Both were designed to address low-quality content rather than to target comparison sites, and the court accepted that Google did not build them to demote rivals. Demotion was nonetheless the effect, since comparison services compile offers already published elsewhere and generate many similar pages, precisely the traits the algorithms penalise.

How was the loss measured? According to the article, the court drew on StatCounter data showing Google's Swedish search share never fell below 85 percent on desktop and 98 percent on mobile between 2017 and 2023, then relied on a 2019 Nielsen Norman Group eye-tracking study finding that 28 percent of clicks on organic results go to the first result, 19 percent to the second, and 12 percent to the third. The award still fell well short of the claim. PriceRunner sought principal running to 3,171,900,000 pounds for the United Kingdom alone; the court granted roughly a quarter of the total claimed, having found neither side's economic modelling fully persuasive and estimated the damages itself. The judgment connects directly to the regulatory record PPC Land has followed, from the EU Court of Justice upholding the original 2.4 billion euro Shopping fine in September 2024 to the coalition of more than 20 comparison sites arguing Google's November 2024 changes still failed to comply and the still-pending record DMA fine tied to the same self-preferencing question. Read together with the Brussels decisions and the 4.125 billion euro Android fine the Court of Justice confirmed on July 2, the week amounts to a sustained stress test of whether Google's past fixes did what they claimed.

A New York court says a price can be a tie

The measurement business absorbed its own antitrust shock. On July 13, 2026, the United States Court of Appeals for the Second Circuit affirmed an order blocking Nielsen from enforcing a radio-ratings policy that tied its national data product to local data purchases, finding the measurement firm exploited a monopoly to coerce Cumulus Media into buying products it did not want. The three-judge panel, in an opinion by Judge Alison Nathan, upheld a preliminary injunction issued by District Judge Jeannette A. Vargas against what Cumulus calls the Network Policy. For the first time at the appellate level in this circuit, according to the article, the court held that a supplier can violate antitrust law not only through an explicit tie but through pricing designed to achieve the same effect, a theory the panel called constructive tying.

The dispute concerns two products. Nielsen sells local radio data measuring listenership in specific areas, and national data compiling all local figures into a single nationwide report; according to court documents, Nielsen holds a 100 percent share of national radio data. Under the Network Policy adopted in 2024, a broadcast network operating a local station in a Nielsen-measured market and not subscribing to local data there could not buy the Nationwide product with that market included. The result is a degraded report Nielsen's own executives, quoted in the opinion, likened to "Swiss cheese," one that for Cumulus would have excluded San Francisco, Los Angeles and New York City. The policy applies only to the 12 customers running both a national network and local stations, a group that includes the three largest radio companies and together controls roughly a third of all radio advertising spend.

The pricing is where the case turns. According to the appellate opinion, once talks reached impasse in August 2025 and Cumulus threatened suit, Nielsen exempted the broadcaster and offered a standalone Nationwide price at least 150 percent more than any other network paid and ten times what Cumulus was paying under its existing contract. The court found the figure so high as to make standalone purchase economically unfeasible, presenting what the panel called an illusory choice: accept the tied offer, or pay 1.2 million dollars more to buy local data from the cheaper rival Eastlan. Tracing the theory to a 1962 Supreme Court decision and a 1967 Second Circuit case, the panel concluded that tying can occur when a seller de facto ties two products through exorbitant prices, leaving a buyer with only one rational choice. The holding hands buyers across advertising and measurement markets a new argument against dominant vendors that drop formal bundling while preserving a bundle's economics.

That argument arrives at a pointed moment, because Nielsen simultaneously reset the numbers underpinning television deals. According to a July 14, 2026 explainer from the Video Advertising Bureau, Nielsen committed to seven methodological changes to its National Big Data plus Panel product, all deploying as currency on August 31, 2026. The seven, named in the explainer, are Latency Adjusted DASH, the Household Demographic Assignment Model, Integrated Weighting, ACR Monitored Tuning, Hispanic Universe Estimates, Co-Viewing, and Provider B Householding, and they are tied to Nielsen retaining its 2024 to 2026 accreditation from the Media Rating Council. The timing carries weight: the recalculation lands weeks before the autumn season, Nielsen's most commercially important measurement window, and the trade body documenting the changes spent the spring accusing Nielsen of manipulating figures that shape how budgets flow between linear television and streaming. The Household Demographic Assignment Model, a machine-learning system that predicts demographics where data is missing, is meant to correct a bias toward older households that had left younger viewers underrepresented, a shift that sits close to the commercial core of television buying, since advertisers transact against specific age targets rather than undifferentiated totals.

Agents keep hitting walls the web built

The same automation the Android decision seeks to unleash keeps running into a duller obstacle: web pages built for humans. A benchmark published June 18, 2026, and covered by PPC Land on July 20, found that AI agents researching software pricing on 100 leading B2B products hit access errors in nearly a third of all runs, forcing many sessions to pull pricing from third-party blogs rather than the vendor's own site. According to the analysis by Siteline founder David Kaufman, a Claude Sonnet 4.6 agent was given a single repeatable instruction across 534 simulated sessions: find the monthly pricing for all publicly listed plans and list each plan's top features. The median session took about 32 seconds, made three tool calls, and cost 0.24 dollars, but that median hid a wide spread. The fastest tenth of sessions finished in 22.9 seconds at 0.142 dollars; the slowest tenth took 50.4 seconds and cost 0.589 dollars, a 4.2-times price gap.

The cause is technical and specific. According to the report, with the exception of Google, agents built by Anthropic and OpenAI do not execute JavaScript, so any pricing table rendered client-side is effectively invisible to them. Two named cases anchor the finding. Linear, an engineering project-management tool, finished in 16.9 seconds using two tool calls and parsed four plans in a single pass from its own site. Zendesk took 53.1 seconds across six tool calls because its pricing table rendered in JavaScript, leaving no data visible; the agent then pivoted entirely to two external blogs. The consequence compounds: sessions that hit at least one access error pulled 58 percent of their eventual content from third-party sources, against just 12 percent for friction-free sessions, and in 5 percent of runs the agent sourced its whole answer from outside pages. The concern is not merely efficiency, according to Kaufman's framing, because third-party blogs often carry stale pricing, so an agent that defaults to them risks recommending whichever vendor publishes more reliably rather than whichever offers better value. The finding lines up with wider reporting PPC Land has tracked, including Google Search Advocate Gary Illyes describing the web's accumulated JavaScript complexity as an obstacle for AI agents retrieving page content. The recommended fix is comparatively plain: render content server-side, put the important information near the top since agents reliably process only the first 15,000 to 20,000 tokens, and publish actual prices.

That friction takes on a competitive dimension in a market growing more concentrated. According to France's competition watchdog, tested this month, OpenAI, Google and Anthropic together hold about 84 percent of the market for AI agents, and its trials of ChatGPT and Gemini shopping answers found sharply different sources cited by each, raising portability questions for merchants whose visibility now depends on which model a customer happens to use. When a handful of assistants mediate discovery and most of them cannot read a JavaScript-rendered price, the decision about how a page is published stops being a design preference and becomes a distribution question.

Also noted