GSC Goes Global, Every Search Link Gets Rerouted, and AI Mode Gets Smarter

What Happened This Week

Generative search finished a structural transition this period. The tools for measuring and opting out of it are now global, not pilot-stage. Google completed the worldwide rollout of Search Console’s AI performance reports and opt-out toggle on August 31, closing a rollout that started in the UK back in June. Days earlier, Google shipped its longest and hardest-hitting spam update of the year, reshuffling top-10 rankings at a rate 82% above normal churn. Separately, under direct pressure from the European Commission, Google carved the EU out of its site reputation abuse enforcement entirely, the first time one of its core spam policies has split by geography rather than applying uniformly. Google also began routing search result links through a new server-side redirect that makes large-scale programmatic reading of its results pages meaningfully harder, landing in the same window it amended its ongoing lawsuit against SERP-scraping API SerpApi. And AI Mode picked up its third new model in six weeks, Gemini 3.8 Flash, which shipped with a citation bug serious enough that Google had to publicly confirm a fix was coming. None of this points toward AI search slowing down or consolidating. In our read, it points toward a widening gap between how fast Google ships changes to the assistant layer and how reliably that layer is verified before it reaches users. That’s our interpretation of the pattern, though, not something either company has said outright.

The Findings

Did Search Console’s AI reporting just become universal, or is this still a partial rollout?

Google completed the global rollout of Search Console’s Generative AI performance reports and the “Search generative AI control” toggle on August 31, 2026, across all verified properties worldwide. It’s the same feature set that began as a UK pilot on June 3 and had been progressively widening access since (Issue 5 and 6 both tracked earlier stages of this same rollout). The control sits under Settings > Search generative AI and offers three states: include, exclude, or inherit from parent. Opting out removes a domain’s links and crawled text from AI Overviews, AI Mode, and Discover’s generative previews, with changes propagating through Google’s index within 1-2 days. Google’s documentation confirms opting out carries no ranking penalty. The performance report itself tracks impressions by URL, country, and device back to May 18, 2026, but still excludes clicks, CTR, and average position entirely.

What was not confirmed: Google has not given any timetable for adding click or query-level data to this report, and hasn’t addressed whether opting out affects crawl budget or indexation speed for reasons unrelated to the generative-AI toggle itself.

Practitioner implication: this is now table stakes, not a pilot feature. If you haven’t checked your properties’ access yet, do it now. But treat it as an impressions-only visibility signal, not a substitute for real referral analytics. The opt-out is also still a blunt, sitewide-or-nothing lever: it can’t separate AI Overviews from AI Mode, and it does nothing about Google-Extended’s model training use of your content.

Was this month’s spam update unusually aggressive, or does it just look that way?

Google’s August 2026 spam update ran from August 18 at 9:27am PT to August 21 at 1:49am PT, 2 days 16 hours, the longest of the three spam updates Google has shipped this year (March and June were both faster). Google gave no new policy category and described it only as a routine refresh. SE Ranking’s independent analysis of 100,000 keywords across 20 verticals found the update genuinely was harder-hitting than normal: 16.71% of previously top-10 URLs fell entirely out of the top 100, an 82% increase over baseline churn, making a top-10 ranking roughly 1.8x more likely to disappear completely than during an unperturbed period. Volatility ranged from 74.6% in real estate to 85.6% in fashion and beauty, with YMYL categories showing the least disruption.

What was not confirmed: Google did not say whether this update specifically targeted AI-generated or GEO-manipulated content, and SE Ranking’s own tracking doesn’t break down which spam categories were actually being caught.

Practitioner implication: if a client lost rankings in this window, don’t assume a manual action or a structural problem with the page. Cross-check the date against this rollout first. The scale of churn here means normal-looking ranking noise from mid-to-late August may not be noise at all.

Why did Google just create two different versions of one of its spam policies?

Following direct engagement with the European Commission, Google announced on August 28 that starting August 30, manual actions under its site reputation abuse (parasite SEO) policy would stop affecting search results for users inside the European Economic Area, while continuing to apply everywhere else. Practically: a manual action can still be issued against a site, and non-EEA searchers still see the demotion, but EEA searchers see no ranking effect from it at all. Google’s own Search Central Blog confirms this directly, framed as balancing DMA compliance concerns against the policy’s anti-manipulation purpose.

What was not confirmed: neither Google nor the Commission has published the full terms of whatever understanding produced this change, so it’s unclear whether this is a template for other spam policies or a one-off.

Practitioner implication: this is a genuinely new kind of enforcement fork. Geography-split spam policy is not something Google has done before. If you manage sites with EEA and non-EEA traffic that carry (or are vulnerable to) a site reputation abuse flag, the ranking impact of that flag now depends on where your visitor is, not just whether the flag exists.

Is Google’s new goto redirect a scraper crackdown aimed at AI search, or something else?

Google confirmed on August 26, 2026 that search result links are rolling out behind a google.com/goto?url=[encoded] server-side redirect, replacing the direct destination URL that previously sat in the page’s HTML. For a human searcher, nothing visibly changes: the destination still displays as text under each result, and clicking still lands on the same page. For anything reading the results page programmatically, it’s a different story. A results page that used to expose hundreds of destination URLs directly in its markup now requires following each one’s redirect individually to find out where it goes, which is a much heavier, slower process at scale. Google’s own explanation was general, “a long history of technical measures against evolving forms of abuse,” and named no specific target.

What was not confirmed: Google has not said scrapers, rank-trackers, or AI search platforms are the target, and multiple outlets covering this are explicit that the scraper-targeting read is industry interpretation, not a Google statement. Worth noting for context, not as confirmation of intent: this rollout completed in the same window Google amended its ongoing lawsuit against SerpApi (Issue 6) after losing the core of a DMCA claim against it in July. SerpApi’s rank-tracking and SERP-data API is exactly the kind of large-scale, programmatic result-reading this redirect makes more expensive. Whether the two are connected is our inference from timing, not something either party has stated.

Practitioner implication: if your stack includes anything that reads Google’s results programmatically, a rank tracker, a SERP-scraping API, or an AI product that grounds itself by scraping Google directly rather than using a licensed data source, expect that to get slower and more expensive, not faster. This doesn’t touch rankings, indexing, or what a searcher sees, so it isn’t a visibility issue for site owners; it’s an infrastructure cost issue for anyone downstream of Google’s results.

Did Gemini 3.8 Flash improve AI Mode, or mostly just prove Google is shipping faster than it can verify?

Google’s Robby Stein announced Gemini 3.8 Flash’s rollout into AI Mode on September 2, the third Flash model swap in six weeks (3.5 Flash at I/O in May, 3.7 Flash on August 14, now 3.8), available to AI Pro and Ultra subscribers via the model selector. Google frames it as its “most intelligent workhorse model yet” for agentic tasks, coding, and multi-step reasoning. Within roughly a day of shipping, Google confirmed a real problem: AI Mode responses running on 3.8 Flash had stopped reliably linking or citing sources, especially on top-of-funnel queries. Independent researchers spotted and reported it before Google’s own acknowledgment. Google said a fix is coming but hasn’t given a timeline.

What was not confirmed: whether the citation failure is isolated to 3.8 Flash’s launch configuration or reflects something structural about how newer models integrate with AI Mode’s citation layer; Google also hasn’t said when 3.8 Flash reaches non-paying users or whether it’s headed for the zero-click AI Overviews surface.

Practitioner implication: if you track AI Mode citation rates, don’t read a dip during this window as a change in how Google values your content. Check the date against this rollout first. More broadly, three model swaps in six weeks is a fast cadence even by Google’s standards, and a citation-breaking bug surfacing within a day of one of those swaps is the kind of regression that pace makes more likely to slip through. Google hasn’t said anything about its own QA process; we’re inferring the connection from timing alone.

Is Reddit’s citation collapse in ChatGPT Search a Reddit-specific problem or a broader grounding shift?

Promptwatch, an AI visibility tracking firm, reported on August 19 that Reddit’s share of ChatGPT Search citations collapsed 86% over a four-day window in mid-August, from a stable ~3.8% baseline share down to roughly 0.5%, with the drop concentrated around August 14. Promptwatch ran Google AI Overviews and AI Mode as controls over the same period: both showed real but much more gradual declines in Reddit citation share (roughly 11% and 30% relative declines respectively), not a sudden collapse. OpenAI gave Yahoo Tech a general comment saying it doesn’t set a fixed visibility level for individual sites and that ChatGPT still cites Reddit, but has not addressed the mechanism behind this specific drop.

What was not confirmed: whether this reflects a deliberate change to ChatGPT Search’s retrieval/grounding pipeline, a Reddit-side change (robots.txt, licensing terms, API access), or something else; the tracking methodology only measures citations that render in the UI, not underlying retrieval or indexing behavior. A single source has claimed Reddit itself blocked crawlers via robots.txt around August 20, but that claim traces to one blog aggregator, not a major outlet, so treat it as unconfirmed rather than fact.

Practitioner implication: if Reddit-sourced content is a meaningful part of your AI-citation strategy, this is worth monitoring specifically on ChatGPT Search rather than assuming it’s a platform-wide trend. The Google surfaces moved in the same direction but nowhere near as sharply, which suggests something specific to ChatGPT’s pipeline rather than a broad de-prioritization of Reddit as a source.

What This Means For You

In our read, the throughline this period is a widening gap between shipping speed and verification. That’s our interpretation of the pattern below, though, not a claim either company has made about its own process. Google completed a genuinely useful measurement rollout (GSC’s AI toggle and reporting), ran its most aggressive spam update of the year, and split a core spam policy by geography under regulatory pressure: three structural moves in two weeks. At the same time, it shipped a new AI Mode model with a citation-breaking bug it had to publicly acknowledge within a day, and OpenAI’s retrieval pipeline produced an unexplained sourcing shift that its own control comparisons suggest is real. None of this is catastrophic on its own, but taken together it’s a reasonable case for treating any given week’s AI-search behavior as provisional until you’ve checked it against what Google or OpenAI actually shipped that week. The platforms are moving fast enough that “did something change on their end” should be the default first question, not an afterthought.

What to Watch Next

• Gemini 3.8 Flash’s citation fix. In progress. Google confirmed the bug and promised a fix with no timeline; watch AI Mode citation rates for your own tracked queries until it lands.
• EEA site reputation abuse split. In effect since August 30. Watch whether Google extends geography-specific enforcement to other spam policies, which would be a meaningfully bigger structural change than this one policy.
• Search Console AI reporting’s next expansion. Impressions-only for now. Google has given no timetable for adding click or query-level data; this is the single biggest gap practitioners keep flagging.
• ChatGPT Search’s Reddit citation share. Unexplained as of this writing. Watch whether it stabilizes at the post-collapse ~0.5-1.5% range or continues declining, and whether OpenAI ever comments on the cause.
• Whether the August spam update targeted AI-generated content specifically. Unconfirmed. SE Ranking’s volatility data is real but doesn’t isolate spam categories; watch for follow-up reporting that does.
• Google’s goto redirect rollout and the SerpApi lawsuit. Both unfolding in parallel. Neither Google nor SerpApi has confirmed a connection between the redirect rollout and the litigation; watch for whether the amended lawsuit’s next filing references it, or whether Google issues any more specific statement about goto’s purpose.

Did Search Console’s new AI reporting features just launch?

Not exactly. The Generative AI performance report and opt-out toggle finished a global rollout on August 31, 2026, that began as a UK pilot back in June. It’s now available on every verified property worldwide, not a new feature.

Does opting out of AI Overviews hurt my rankings?

No. Google’s own documentation confirms the opt-out control carries no ranking penalty for standard organic search; it only removes your content from AI-generated answer surfaces.

Is Google’s site reputation abuse policy still enforced in the EU?

Only partially. As of August 30, 2026, manual actions under the policy no longer affect what EEA-based searchers see, though the same manual action still fully applies to searchers outside the EEA.

Is Google’s new goto redirect targeting AI search tools?

Not confirmed. Google says only that it’s a general technical measure against abuse and hasn’t named a target. It’s rolling out in the same window Google amended its lawsuit against SERP-scraping API SerpApi, which is worth watching, but no direct connection has been confirmed by either party.

Did Gemini 3.8 Flash break AI Mode’s source citations?

Yes, temporarily. Shortly after its September 2 launch, Google confirmed AI Mode responses running on Gemini 3.8 Flash had stopped reliably citing sources on some queries, and said a fix was coming without giving a timeline.

Why did Reddit’s citations in ChatGPT Search suddenly drop?

It’s not yet fully explained. Promptwatch tracking found an 86% collapse in Reddit’s citation share over four days in mid-August. OpenAI told Yahoo Tech in general terms that it doesn’t set a fixed visibility level for individual sites and that ChatGPT still cites Reddit, but hasn’t addressed the specific mechanism behind this drop, and it’s unclear whether it’s a ChatGPT-side change or something Reddit did.