What Happened This Week
Two regulators, an ocean apart, reached different verdicts this period on how much structural change Google’s search and advertising business should be forced to absorb. In the US, a federal judge let Google keep its ad exchange intact, accepting behavioral remedies instead of the breakup the Department of Justice wanted. In the EU, the opposite instinct produced a literal redesign: Google rebuilt part of its search results page for EEA users only, and dropped a travel feature entirely rather than restructure it, its response to a €460 million fine the European Commission issued in July for self-preferencing its own travel and shopping results. That EU move is also a direct continuation of something this briefing flagged as worth watching last issue, whether Google would extend geography-specific enforcement beyond a single spam policy. It just did, in a bigger way than a policy carve-out. Underneath both regulatory stories, the mechanics of search kept drifting further from anything a keyword rank tracker can measure: Google began testing ads inside AI Mode using match types advertisers already know, in a small unscoped experiment, OpenAI kept building out its own ad infrastructure while its geographic footprint actually grew slightly earlier than reported, and two independent studies this period both found that a synthesized AI answer and the data or rankings underneath it can diverge substantially, in pricing and in which sources get credited.
The Findings
Did the DOJ actually break up Google’s ad business, or just rearrange the furniture?
On September 2, 2026, Judge Leonie Brinkema of the US District Court for the Eastern District of Virginia issued her remedies ruling in the Department of Justice’s ad tech antitrust case, more than a year after finding Google had illegally monopolized the publisher ad server and ad exchange markets. She rejected all three of the DOJ’s structural remedies: the divestiture of Google’s AdX exchange, the open-sourcing of DFP’s auction logic, and the contingent divestiture of DFP Remainder. Instead, she accepted most of the parties’ proposed behavioral remedies, aimed at increasing interoperability between Google’s ad tools and rivals’.
What was not confirmed: the specific behavioral obligations. Brinkema’s full opinion was filed under seal, with 14 days for the parties to request redactions and 30 days to submit a jointly proposed final judgment, meaning the operative terms were not yet public as of this writing. Reporting describes the general direction (real-time bid access for rivals on AdX, changes to Unified Pricing Rules) based on Google’s own prior proposals, not the sealed opinion itself.
Practitioner implication: nothing changes in your ad accounts this month. Google keeps AdX and DFP under one roof, and the take rate is unchanged for now. The concrete terms won’t be public until the redaction window closes; treat any specific “Google must do X” claim before then as inference from pre-trial proposals, not the ruling itself.
Did Google just extend its EU-only search split, the way this briefing said to watch for?
On September 8, 2026, Google Search Central published documentation for two new EEA-only result units: an Aggregator Unit for vertical search services (comparison shopping sites, online travel agencies, metasearch engines, and directories) covering hotels, flights, long-distance transport, and retail products, and a Supplier Unit for direct providers like airlines and individual hotels. Only one Aggregator Unit renders per query, with the top-ranked approved provider expanded by default; a Supplier Unit only appears alongside an Aggregator Unit. Entry to the Aggregator Unit requires an application and, depending on category, a structured data feed or live API. The same week, Google removed its dedicated vacation-rentals unit from EEA search entirely, telling the travel trade press it could not currently display the unit while making the DMA-related changes and that it hopes to restore some version of it later.
The redesign follows a €460 million ($525 million) fine the European Commission issued Google on July 23, 2026, for self-preferencing its own services, including shopping, hotels, transport, and sports results, in Search, part of an €890 million combined penalty (a separate €430 million targeted Google Play’s anti-steering rules). Google has 60 days from that decision to show compliance or face daily penalties of up to 5% of global turnover.
What was not confirmed: Google’s own documentation gives no launch date, never references the Digital Markets Act by name, and states availability as EEA-only with no mention of extending to other regions. Reuters reported Google’s own estimate that earlier DMA compliance changes cut free direct-booking traffic to European businesses by 30%, but that figure describes the earlier changes, not this specific rollout. Reporting has also pointed to a September 23, 2026 Commission checkpoint on whether Google’s response is adequate, though the Commission has not published that date itself.
Practitioner implication: this is the geography-specific enforcement pattern this briefing flagged last issue (the site reputation policy split, in effect since August 30) showing up again, and in a more structural form: an actual results-page redesign, not just a change in how a penalty applies. If you operate in the affected commercial categories in the EEA, the practical decision is whether to pursue Aggregator Unit approval and the feed work it requires, or rely on the automatic Supplier Unit, which needs no data submission but only appears when an Aggregator Unit does. Watch whether this EEA-specific architecture pattern extends to other categories or other regulators; that is now a live and recurring possibility, not a one-off.
Can you advertise in AI Mode without switching to Google’s automated AI Max?
On September 4, 2026, Google Ads Liaison Ginny Marvin confirmed that Google has begun testing exact match and phrase match keywords from standard Search campaigns as text ads inside AI Mode, on queries where Google identifies “explicit and direct user intent.” Marvin’s own description matters here: “a small experiment that recently kicked off,” not a policy change. Before this test, the only routes into AI Mode ad placements were AI Max, Performance Max for Search, or broad match with Smart Bidding. Google has named no countries, given no end date, and released no performance data.
This follows Google’s September 1 mandatory migration of Search campaigns using broad match or Automatically Created Assets into AI Max, though the deprecation of legacy Dynamic Search Ads campaigns was pushed to February 2027 to avoid disrupting advertisers ahead of Q4 retail.
What was not confirmed: whether this experiment will expand, hold at its current scope, or quietly end. Google has published no targeting mechanics beyond the match-type eligibility and the “explicit and direct user intent” condition, and no dedicated Google Ads API reporting fields have been announced for it, unlike the reporting views shipped alongside the AI Max migration itself.
Practitioner implication: this is not yet a stable second path into AI Mode’s ad inventory. It’s a limited, unscoped experiment restricted to a subset of queries Google hasn’t defined publicly. If your account is inside it, treat any AI Mode delivery from exact or phrase match campaigns as provisional, not a new lever to build a strategy around, and watch whether Google expands or ends the test before committing budget to it.
Did OpenAI’s global ChatGPT Ads expansion actually happen this period?
Search Engine Land reported on September 4, 2026, that OpenAI had expanded its ChatGPT Ads stack: more flexible custom audience management (mixing identifier types, exceeding 5 million members, lower restrictions on exclusion audiences), additional identifiers (including phone numbers, names, and Android GAID) that its existing Measurement Pixel and Conversions API can use for conversion matching, carousel-level reporting for product-feed advertisers, and an Ads Manager plugin for building campaigns directly inside ChatGPT and Codex using natural language. What was not accurately timed in some coverage: the actual self-service geographic launch across India, Europe, the Middle East, and North Africa happened on August 31, 2026, per OpenAI’s own announcement, one day before this briefing’s coverage window opens. OpenAI said the ads product had reached a $1 billion annualized revenue run rate in under 200 days and was used by tens of thousands of advertisers, with ads restricted to Free and Go subscription tiers and Plus, Pro, and Enterprise tiers remaining ad-free.
What was not confirmed: how the additional identifier support changes match-rate accuracy in practice, and whether the natural-language campaign creation tools in Ads Manager materially change campaign quality versus simply lowering the setup barrier.
Practitioner implication: if you’re planning ChatGPT Ads for international markets, the actual geographic access opened slightly earlier than most coverage this period implied. What’s genuinely new this period is measurement and workflow tooling, not a new attribution product and not the footprint itself.
Do AI Mode’s product picks and AI citation patterns actually track the underlying data, or have they quietly decoupled?
Two separate studies published September 1, 2026, both point the same direction. Productrise, an e-commerce visibility tracking firm, compared Google AI Mode’s product recommendations against the standard “Popular Products” Shopping carousel across more than 100,000 queries and 2 million listings in the US and UK from August 9 to 31. The same product appeared on both surfaces only 1.28% of the time per day. Where it did, the first-listed seller differed 49.6% of the time and the price differed 38.1% of the time, with AI Mode showing the higher price in 68.4% of those cases (a 22.2% median gap when it was higher). Google told Futurism that both surfaces draw from the same Shopping Graph, but has not independently verified Productrise’s findings.
Separately, a Wellows dataset covering 596,723 prompts answered by two or more of ChatGPT, Gemini, Perplexity, Google AI Overviews, and Google AI Mode, reported by TechTimes, found only 10.2% of cited URLs appeared on more than one engine for the same prompt, rising to 17.9% at the domain level, while brand-name overlap in the answer text reached 67.4%.
Older data adds a related, separate data point: an Ahrefs analysis of roughly 540,000 query pairs from September 2025, the most recent independent breakdown specifically comparing AI Overviews against AI Mode, found the two Google surfaces shared cited URLs only 13.7% of the time, which Ahrefs attributed to AI Mode’s query fan-out mechanism decomposing prompts into multiple sub-queries. That figure is a year old and describes Google’s two surfaces against each other, not a finding from the Wellows dataset above.
What was not confirmed: Productrise’s methodology only analyzed the free “popular_products” carousel against AI Mode’s product cards, not all shopping surfaces, and the queries reflect Productrise’s own monitored set rather than a random sample of all Google searches. The Wellows dataset’s underlying collection methodology beyond the headline overlap figures was not independently verified for this briefing.
Practitioner implication: ranking well in a shopping carousel or being cited on one AI surface tells you very little about your standing on another. Entity-level authority across sources, not URL-level rank tracking, is the more relevant signal to build toward.
Are publishers still restructuring around the platforms, and did Apple clarify its own AI crawler rules?
USA Today’s parent company confirmed on September 3, 2026, a restructuring of its audience operations into three divisions (a Central Production Desk, a Content Pillars Audience Desk, and a Strategic Platforms Desk), alongside layoffs, with senior vice president Monica Richardson telling staff that “search traffic is under pressure, and platforms are increasingly keeping user experiences to themselves instead of sending them back to us.” Separately, Apple updated its Applebot support documentation around September 4-7 to state explicitly that “site rules for Applebot-Extended are not considered in ranking for Search,” clarifying that disallowing Apple’s AI-training crawler in robots.txt has no effect on inclusion in Apple’s search products, mirroring the split Google and OpenAI already maintain between their indexing and training crawlers.
What was not confirmed: USA Today did not disclose headcount figures, and Apple’s documentation does not address whether the training opt-out affects ranking inside Siri or Apple Intelligence answers specifically, only “Search” generally.
Practitioner implication: the Apple clarification removes one specific piece of uncertainty. Blocking Applebot-Extended costs you AI-training inclusion, not search visibility. The USA Today restructuring is one more data point in a pattern this briefing has tracked for several issues: publishers reorganizing around off-platform distribution rather than betting on search referrals recovering.
What This Means For You
The regulatory story this period is really one story told twice, with opposite outcomes. The same underlying question, how much structural change should platforms with dominant search and ad positions be forced to accept, produced a US court accepting behavioral fixes while keeping Google’s ad business whole, and an EU regulatory process that keeps producing geography-specific product redesigns, now backed by a nine-figure fine, instead. That divergence is now a pattern, not a one-off: last issue’s site reputation policy split and this issue’s Aggregator and Supplier units are the same enforcement instinct applied twice in a matter of weeks. If you operate across both jurisdictions, you’re increasingly managing two different search products, not one product with a regional setting. Meanwhile, the monetization layer of conversational search kept building out on both sides of the Google-OpenAI rivalry, though Google’s own move is still an unscoped test, not a settled product, and the two studies published this period are a concrete reminder that an AI-synthesized answer, whether it’s a shopping recommendation or a citation, is not a transparent window onto the underlying rankings or catalog. It’s a separate output with its own selection logic, and increasingly its own commercial incentives layered on top.
What to Watch Next
- The DOJ v. Google AdX remedies, once unsealed. In progress. The specific behavioral obligations remain sealed pending a 14-day redaction window and a 30-day joint final judgment filing; the real operational impact on ad accounts won’t be clear until then.
- Whether the EEA-specific search architecture pattern spreads further. Live and recurring. Two geography-specific changes (the site reputation policy split, the Aggregator/Supplier units) in a matter of weeks makes a third plausible; watch whether other commercial categories or other regulators adopt the same approach.
- Whether the Commission accepts Google’s EEA redesign as adequate compliance. Open. Google has 60 days from the July 23 fine to demonstrate compliance; some reporting points to a September 23 checkpoint, though the Commission has not confirmed that date directly.
- Whether Google’s AI Mode ad experiment expands, holds, or ends. Unscoped. No countries, end date, or performance data have been published; treat it as provisional until Google says more.
- Search Console’s AI performance report adding click or query data. Still impressions-only. Google has given no timeline for adding the metric practitioners most want.
- Whether Google’s August spam update specifically targeted AI-generated content. Unconfirmed. Independent case studies (Glenn Gabe/GSQi) show AI-generated content implicated in some of the largest ranking losses, but Search Engine Journal’s own review of the “AI-targeting” theory found it resting on thin, anecdotal evidence rather than a Google confirmation.
- The SerpApi lawsuit’s next hearing, September 29. Scheduled, outside this issue’s window. SerpApi has asked the court to dismiss Google’s amended complaint with prejudice; no connection to the goto redirect rollout has been confirmed by either party.
Did the DOJ force Google to sell its ad exchange?
No. Judge Leonie Brinkema rejected the divestiture of AdX and related structural remedies on September 2, 2026, accepting behavioral remedies instead. The specific terms remain under seal.
Is Google removing search features in Europe?
Yes, partially. Google introduced new Aggregator and Supplier result units for the European Economic Area on September 8, 2026, following a €460 million DMA fine for self-preferencing, and separately removed its dedicated vacation-rentals unit from EEA search entirely, saying it hopes to restore some version of it later.
Can I advertise in Google’s AI Mode without using AI Max?
Only inside a small, unscoped experiment. Google confirmed on September 4, 2026, that exact match and phrase match keywords from standard Search campaigns can serve as text ads in AI Mode on queries with “explicit and direct user intent,” but it’s described as a test, not a general policy, with no countries, end date, or performance data published.
Does AI Mode show the same product prices as Google’s Shopping carousel?
Rarely, and not consistently. A Productrise study of over 100,000 queries found the same product appeared on both surfaces only 1.28% of the time, and when it did, prices differed 38.1% of the time, with AI Mode showing the higher price in most of those cases.
Does blocking Apple’s Applebot-Extended hurt my search ranking?
No, according to Apple’s own documentation, updated in early September 2026. Disallowing Applebot-Extended opts a site out of AI-training use only; pages remain eligible for inclusion in Apple’s search products.
