AI Overviews Now Appear in Nearly Half of All Google Searches

Two independent datasets converge on a striking finding: AI Overviews now dominate the Google SERP. According to Search Engine Roundtable, Similarweb puts the rate at ~43% of all queries, while Semrush places it even higher at ~48% — with informational queries pushing the figure higher still. The acceleration is significant enough that SEOs can no longer treat AI Overviews as an edge-case feature.

The implications for traffic are severe. An eight-week analysis reported by Search Engine Journal found that the 'Generative AI' filter in Google Search Console inflates impression counts and artificially elevates position-one rankings, masking real traffic losses. The core problem: AI Overview impressions are counted even when no click occurs, and the averaging methodology distorts position data. The report warns that marketers treating this dataset at face value risk being blind to genuine organic performance declines.

Adding to the complexity, Google has clarified — via Lumar's July digest — that only visible links inside AI Overviews count as impressions, not brand mentions. This means many publishers seeing their brand referenced in AI answers are receiving zero impression credit.

Google is now embedding Top Stories carousels directly inside AI Overviews, and a new Search Console setting allows publishers to opt out of AI training — but the two features are directly linked. As Search Engine Journal reports, publishers who exercise the AI opt-out risk losing their Top Stories placement entirely, since that placement now lives inside the AI Overview surface.

The strategic calculus is difficult. Opting out protects content from being used to train Google's AI models, but removes one of the most visible organic traffic drivers for news publishers. Search Engine Journal's follow-up piece breaks down what the opt-out specifically controls and which metrics publishers should benchmark before making the call. The window to monitor the trade-off before making a decision is narrow, as the feature is already rolling out.

The week's most data-heavy content story came from Ahrefs, which studied 331,000 pages to examine how Google treats AI-generated content. The finding: Google does not penalize content for being AI-generated; it penalizes content that is low quality, regardless of how it was produced. The study aligns with Google's own stated position that content is evaluated on quality signals, not origin.

However, a separate but related story from Search Engine Journal complicates the picture. Google appears to be penalizing AI-generated content under an expanded definition of 'thin content' — meaning that while the signal is quality-based, AI production methods frequently produce outputs that trigger thin-content classifiers. This is not a contradiction of the Ahrefs finding so much as a practical corollary: AI content that lacks depth, originality, or substantive value is being caught by quality filters.

Additional data from Search Engine Journal reinforces the pattern: pages scoring high on AI content detectors tend to rank lower and appear less frequently in Google's index, though heavily flagged pages are not categorically excluded from the top 10. The correlation exists without being a hard cutoff.

Following the initial launch covered in our July 6th issue — which introduced Platform Properties for Instagram, TikTok, X, and YouTube and noted the early `/ai` URL trick for accessing generative AI reports — Google has now completed the global rollout of Platform Properties to all users worldwide, confirmed simultaneously by Google's Search Central Blog, Search Engine Journal, and Search Engine Roundtable. Alongside the full availability announcement, Google published a new social and video performance guide to help users interpret the data. For SEOs who claimed platform properties early, the focus now shifts from access to interpretation of the expanded dataset.

Two ongoing bugs in Google Search Console are affecting SEO workflows. First, the Page Indexing report has been stuck showing weekly data groupings instead of daily breakdowns since June 11th, according to Search Engine Roundtable. Google has applied partial fixes multiple times, but the issue keeps recurring — meaning daily indexing trend analysis is unreliable.

Second, the 'Request Indexing' feature for robots.txt files is not functioning correctly for all sites. Barry Schwartz notes the feature works for some but fails for others, leaving SEOs who rely on GSC to trigger robots.txt recrawls in an uncertain position until a confirmed fix arrives.

Search Engine Roundtable's Barry Schwartz reported significant Google search ranking volatility beginning Thursday, July 23rd, with turbulence intensifying through Friday, July 24th, and continuing into the weekend. The signals point to an unconfirmed algorithm update — Google has not officially acknowledged or named the event. Tracking tools registered elevated fluctuation across multiple sectors, though the exact scope and nature of the update remains unclear. SEOs observing ranking shifts during this window should factor in the timing when diagnosing causes.

Google updated its review snippet structured data documentation to explicitly prohibit fake or undisclosed incentivized reviews, and — critically — added language warning that violations can now trigger manual actions. Both Search Engine Journal and Search Engine Roundtable covered the update, with Search Engine Roundtable noting Google's direct quote: *"Don't include fake or undisclosed incentivized reviews on your page or in your structured data markup."*

While the underlying policy has long existed in spirit, the explicit documentation of manual action enforcement is new. Site owners collecting or displaying reviews via schema markup should audit their review acquisition practices immediately.

A real-world incident this week provided a textbook illustration of a foundational but frequently misunderstood SEO principle. Shared Claude chat pages began appearing in Google Search results despite having a noindex header in place — because that header was sitting behind a robots.txt Disallow rule, meaning Googlebot never fetched the page and therefore never saw the noindex directive.

As Search Engine Journal explains, blocking crawling via robots.txt and preventing indexing via noindex are two separate mechanisms. If Googlebot is blocked from crawling, it cannot read the noindex tag. The Claude incident serves as a live case study for any site using robots.txt as a proxy for noindex — a practice that does not reliably prevent indexation.

A controlled SEO experiment reported by Search Engine Journal reveals the mechanics of Google's Web Rendering Service (WRS) when it hits its five-second rendering limit. The key finding: Google does not abandon the render when time runs out — it effectively pauses a virtual clock, allowing rendering to continue beyond the five-second window rather than cutting off the process entirely.

This has meaningful implications for JavaScript-heavy sites. Content that loads after five seconds is not necessarily lost to Googlebot; the rendering continues, but the behavior is more nuanced than a hard cutoff. Technical SEOs managing client-side rendering should factor this into their rendering architecture decisions.

Google's John Mueller was unusually active with technical guidance this week. On conflicting metadata — such as a 301 redirect pointing one way while a canonical points another — Mueller stated clearly that Google will not publish its order of precedence for conflicting signals, confirming to Search Engine Roundtable that this information simply won't be made available. His advice, per Search Engine Journal: fix the conflict rather than testing which signal wins.

On CMS-injected URLs, Mueller clarified the SEO impact of random URLs added to HTML by CMS platforms — guidance useful for sites running popular CMSes that may unknowingly add extraneous links. On internal search pages, Mueller noted in the Search Off the Record podcast that blocking internal search result pages is no longer explicitly required by Google Search Essentials, but strongly recommends it anyway due to server load risk and potential quality signal contamination. Separately, Google signaled it may treat spammed internal search box pages as a site quality issue, potentially classifying them similarly to hacked content.

Two pieces this week flagged a practical pitfall with Google's `unavailable_after` meta attribute. Search Engine Roundtable explains the core problem: if a publisher updates an expiry date after the original date has been set, Google may not recrawl the page before that date passes, leaving the page incorrectly removed from search results with no mechanism to recover it in time.

Google's Gary Illyes, per Search Engine Journal, acknowledged uncertainty about whether continuously rolling forward the `unavailable_after` date on page renewals could create indexing issues — noting he would need to check. For publishers using this directive for time-sensitive content management, the recommendation is to avoid relying on last-minute date updates.

Building on the Semrush analysis of 1,094 ChatGPT topic categories first reported in our July 20th issue — which found only 15.2% had a clear brand owner — new data this week adds two important dimensions. First, Search Engine Journal clarifies that citations are a poor predictor of which brand actually gets named in AI responses, sharpening the distinction between being referenced and being recommended. Second, a new Victorious study reveals that AI models can accurately describe 96% of brands but rarely mention them unprompted in generated responses. Together, these findings reframe the opportunity: the gap between AI recognition and AI citation is large, and the window to claim category ownership before competitive consolidation is still open.

Two pieces this week offered rare insight into the internal mechanics of AI search citation. On Perplexity, an SEJ contributor analyzed the live answer stream (rather than final outputs) and found that Perplexity always queries the live web, making every query potentially winnable for any site — unlike LLMs that rely on frozen training data. The stream analysis also revealed how video and local results factor into Perplexity's source selection.

On ChatGPT, new research shows that citation rates vary sharply by topic: travel content is cited far more often than education content. This means the opportunity to earn ChatGPT citations is not uniform across industries, and content teams in high-citation categories like travel have a structural advantage in AI visibility.

New Similarweb data covered by Search Engine Journal challenges the dominant narrative that AI search is displacing Google. Instead, users are adopting AI tools as additive behavior rather than a direct substitution — a meaningful nuance given the 9% traditional search drop associated with ChatGPT access documented in our July 13th issue. The new data also reveals that ChatGPT distributes outbound clicks to a very narrow set of publishers, meaning AI referral traffic is highly concentrated rather than broadly distributed — an important caveat for brands assuming AI visibility translates to proportional traffic gains. SearchPilot reinforces the grounding point: Google search performance remains the foundation for AI search visibility, and teams pursuing GEO tactics without first addressing their Google presence are likely misallocating resources.

Kevin Indig's Growth Memo H1 2026 halftime report — sent to 28,367 subscribers — synthesizes the defining AI and search developments of the first half of the year. The memo frames trust as an emerging ranking signal, documents shifts in search behavior, and tracks the broader economic disruption AI is causing across software stocks, token budgets, and job markets. Indig notes that the pace of development is rapid enough that he's considering moving from semi-annual to monthly analyses — a telling signal about how fast the landscape is shifting.

Google announced version 0.2 of its Open Knowledge Format (OKF), adding five new trust signals to the structured data framework, according to Search Engine Journal. The OKF is designed to help Google evaluate entity trustworthiness and authority, and this update extends its signal set. While still a relatively early-stage initiative, the expansion signals Google's continued investment in entity-based trust evaluation — a direction directly relevant to E-E-A-T and Knowledge Graph optimization strategies.

Moz published a multi-engine study examining whether schema markup is a direct ranking factor or simply best practice for SERP feature eligibility. The research covers structured data's relationship with search visibility and specifically addresses local SEO implications. While the schema-as-ranking-factor debate is long-running, the multi-engine scope — extending beyond Google — adds a fresh angle to the analysis.

Google Ads made structural changes to automated bidding this week. Target CPA and Target ROAS are now available as explicit bidding strategy options in the Google Ads interface, previewed in Ads Editor 2.13. More urgently, an August 17th update will change how budget-limited Target CPA and Target ROAS campaigns behave — pushing them closer to stated targets, but at potentially higher costs. Search Engine Journal advises advertisers to audit and adjust bids before the update takes effect to avoid unexpected cost increases.

Following three consecutive weeks of rapid ChatGPT Ads platform development — global expansion and AI creative generation (July 6th), Sales ROAS metrics and geo-location controls (July 13th), and conversion value attribution columns (July 20th) — OpenAI this week rolled out another significant batch of updates: daily budget mechanics, budget pacing, geo-exclusions, conversion-optimized campaigns, API features, and a new product ad format, per Search Engine Roundtable. The most strategically notable addition is a new 'Agent' campaign type that routes commercial-intent searchers into a Business Agent chat experience rather than a traditional landing page — a potentially significant departure from standard click-to-site advertising logic that marketers building AI search presence should evaluate.

Following our July 20th coverage of the SerpApi lawsuit dismissal — where the court rejected Google's DMCA claims, finding that Google's SearchGuard tool protected ad revenue rather than copyright — Search Engine Journal has now drawn out the broader implication for all publishers: the same legal reasoning that defeated Google weakens the copyright-based defenses any website can mount against AI crawlers and data scrapers. Publishers and SEOs who rely on legal mechanisms to control how their content enters AI training pipelines now have fewer enforceable options, and need to reassess both technical and contractual content protection strategies in light of the ruling.

Two competitive landscape data points emerged this week. First, Amazon has not returned to U.S. Google Shopping in the year since it withdrew in July 2025, per Search Engine Journal analysis of auction data. The absence has reshaped CPCs and auction dynamics for the retailers who remain, creating both risks and opportunities depending on product category.

Second, Microsoft's Q4 2026 earnings showed Bing search and advertising revenue up 10% year-over-year — but this marks a deceleration from 12% the prior quarter, and well below the 21% peak recorded a few quarters earlier. The slowdown may indicate cooling momentum for Bing's AI-powered search growth narrative.