Google Search Traffic to Publishers Down 40% YoY as AI Mode Grows

Following the August 2026 spam update — which ran from August 18–21, produced post-completion volatility through late August, and prompted a public statement from Google VP Rajan Patel endorsing the spam team (August 31st issue), and whose visibility impact data (Amazon −16.9%, IRS −59.6%, SBA −52.8%) was reported the week of September 7th — Google has now launched a separate September 2026 spam update, its fourth spam update of the year. The global rollout across all languages is expected to take up to two weeks. Barry Schwartz noted ranking volatility already on September 23rd and 24th, attributing it to an unconfirmed algorithmic update running concurrently. Notably, Schwartz observed that community chatter about the volatility is quieter than in previous cycles — a pattern he attributes to baseline organic traffic already being suppressed, meaning fewer SEOs are noticing the swings.

Alongside the spam update rollout, Search Engine Journal reports that Google has already deployed a new AI spam detection system called SAFE, specifically engineered to catch AI-generated content spam. The system is believed to be integrated into the September spam update's enforcement mechanisms, making this update particularly significant for sites relying on mass AI content production. The deployment of SAFE signals that Google is now explicitly building dedicated infrastructure to combat AI-driven content manipulation, beyond the general spam systems already in place.

Separately, spammers have been exploiting public Gemini Notebook (NotebookLM) pages to flood Google Search with low-quality content across categories including adult content, peptides, and coupon codes — highlighting a vulnerability in Google's own AI infrastructure that SAFE may be partially designed to address.

A Chartbeat report quantifies the scale of Google's traffic retreat from publishers: Google Search and Discover traffic has declined 40% year-over-year, with the rate of decline accelerating. In its place, direct traffic, internal referrals, and dark social have grown — suggesting publishers are increasingly building audiences independent of Google.

New research covered by Search Engine Journal adds a critical dimension: Google AI Mode generates significantly fewer clicks than traditional search and delivers a poor user experience, directly contradicting Google's implied narrative that AI-enhanced search benefits users and publishers. The mechanism is now clearer: Google is placing links inside AI Overviews that route users into AI Mode rather than to external websites, as Search Engine Roundtable reports. These links appear at the bottom of AI Overviews as follow-up question prompts, creating an AI-to-AI navigation loop. Search Console tracks these two interaction types differently, adding new complexity to performance attribution.

Despite the overall traffic decline, there is a notable structural shift in AI Overviews that creates a new citation opportunity: new data reported by Search Engine Roundtable shows external links in AI Overviews have surged from near 0% to over 25% of AI Overviews. However, the picture is complicated — as Search Engine Journal reports, not all of those links reach the open web. Some route users to AI Mode rather than to publisher pages, and Search Console measures these two interaction types separately.

For SEOs, Search Engine Journal cautions that Search Console's 'block flattening' methodology for AI Overview position data can make weak AIO visibility appear as a top-ranking result. The recommendation: abandon position as a primary KPI for AIO performance, and instead track visits and conversions as outcome-based metrics.

Against the backdrop of the 40% year-over-year Google traffic decline reported this week and the September 7th issue's Harvard study finding the public has little moral objection to AI replacing search marketers, Axios co-founder Jim VandeHei added a stark long-horizon forecast: the open web will largely die within 2–5 years as LLMs become the primary information access layer. VandeHei argued that unique, human-created content will be the key remaining differentiator for publishers that survive. The prediction is notable less as a new data point than as a signal that mainstream media figures are now framing the structural traffic shift — documented in concrete terms this week — as existential rather than cyclical.

Google officially announced web multimodal search performance reporting in Search Console via the Google Search Central Blog, giving publishers visibility into how their content performs in visual and image-based search experiences. In the Performance report's Search Type filter, Web results are now split into two sub-categories: 'Text-based' and 'Multimodal' — with some users already able to see the new filter.

As Search Engine Journal notes, the new multimodal filter shows which pages receive traffic from image-based searches, but does not include query-level data — a meaningful limitation for SEOs trying to understand which visual search queries drive visits. The rollout is ongoing, and not all accounts have access yet. This is Google's clearest signal yet that it considers multimodal search a distinct traffic channel worth reporting separately.

Building on the September 14th issue's coverage of the SEO Trust Crisis and accreditation debate — which flagged AI tools lowering the barrier to entry for unqualified practitioners — a Fiverr survey of search freelancers this week provides a concrete market demand signal: more freelancers report rising demand for GEO services than for traditional Google ranking work. The majority still believe in SEO's long-term viability, but the demand shift on the largest freelance marketplace represents a meaningful external validation of the service reorientation that the September 14th State of Search 2027 report argued should be underway. Aleyda Solis has also released an updated AI Search Optimization Checklist (September 2026), adding workflow questions around AI-assisted journeys, brand mention mapping across owned and earned sources, and structural gap audits — designed as a repeatable process aligned to business goals.

Following the August 31st issue's report on the ChatGPT Ads platform update (Ads Manager plugin, custom audience improvements, conversion matching, new ad formats, and budget pacing controls) and the September 7th issue's coverage of ChatGPT Shopping's growing dependence on feed-integrated product sources, new data this week quantifies the channel's actual ad penetration: ChatGPT ads appeared in 47% of video gaming conversations among US desktop users, with retail verticals also seeing significant ad presence. Amsive frames ChatGPT Ads as a conversational-intent channel distinct from keyword-based search advertising — users define problems and compare options mid-conversation rather than submitting discrete queries. Search Engine Journal, citing OpenAI and Go Fish Digital, outlines how paid and earned AI visibility can complement each other: earned GEO visibility builds citation likelihood, while ChatGPT ads capture users in active decision modes. Baidu has also introduced an AI CTR metric dividing clicks by citations (not impressions) — a new benchmark for quantifying AI-driven citation traffic that has no equivalent in current Google or OpenAI measurement frameworks.

Google announced R4T-Diffusion, a new query fan-out framework described as production-scale ready, which changes how Google expands and interprets search queries to match content against user intent at scale. SEOs should monitor for ranking or traffic changes tied to this new retrieval mechanism, though Google has not detailed exactly when it becomes active in search.

On the technical side, Google's Lighthouse 13.5 added a new audit for AI Agent Resource Discovery (ARD), reflecting growing attention to how AI crawlers and agents discover web resources. Notably, Search Engine Journal flags that the Lighthouse audit's checks differ from the latest ARD specification, which may create compliance confusion for developers. Moz's Whiteboard Friday explains the related NLWeb protocol, which uses the ASK system and structured data to make sites machine-readable for AI agents — an emerging technical SEO consideration as the agentic web expands.

Google is sending notifications to Google Merchant Center users whose e-commerce platforms support the Universal Commerce Protocol (UCP), informing them that their products are automatically eligible for native checkout — meaning purchases can be completed directly within AI Mode and Gemini without the user ever visiting the merchant's website. Merchants who want to preserve their own site traffic must actively opt out; the default is inclusion.

Marie Haynes' newsletter confirms that 'buy now' buttons are already appearing in search results for UCP-enabled sites. Search Engine Journal's analysis of agentic commerce more broadly finds that while major payment providers (Visa, Mastercard, PayPal, Stripe) have yet to name live merchant deployments, Google has identified three — suggesting the infrastructure exists but real-world adoption remains limited.

YouTube is expanding its Ask YouTube AI search experience with shopping features, including the potential to organize product review videos into comparison tables and answer purchase-intent questions directly on watch pages. This convergence of AI search and e-commerce on YouTube signals a significant channel for product-focused SEOs and video content creators targeting commercial queries.

Two complementary issues with Google Business Profiles surfaced this week. First, Steve Toth's SEO Notebook newsletter highlights that anyone can suggest edits to a Google Business Profile without permission from the business owner — a well-known but under-appreciated vulnerability for local SEO.

Google's own updated help documentation now makes clear that business owners have just four days to reject a user-suggested edit before Google may automatically apply the change to the live listing. Combined, these two facts mean that inaccurate or malicious edits can go live quickly if businesses don't actively monitor their profiles. Separately, Google Business Profiles appears to have temporarily removed the 'Permanently Closed' status option, along with other closure statuses — the cause has not been confirmed as intentional or a bug.

Google is now proactively emailing Business Profile owners when spam detection systems identify and remove a spike in fake reviews, with the notification stating 'Google has detected a spike in spam reviews on your Business Profile and removed them so they do not affect your Google rating.' This transparency in spam enforcement is a positive development for local businesses.

In a separate policy update, Google has updated its Google Posts help documentation to prohibit unverified phone numbers in Posts — only verified contact information is now permitted. Google is also testing an AI-powered automatic business hours feature that interprets natural language descriptions and converts them into structured format for Search and Maps listings.

Google has updated its VideoObject structured data documentation to officially support the `creator` and `author` properties, and has clarified which interaction types are supported for `interactionStatistic`. The documentation now explicitly calls out these additions: *"to document support for the creator and author properties, and clarify supported interaction types for interactionStatistic in VideoObject structured data."* Publishers using video schema should review and update their markup to align with the new documentation.

Job board operators are reporting months-long waits for Google Indexing API approval with no response and no stated review timeline in Google's own documentation. For time-sensitive job listings, this delay has direct SEO consequences — listings may go unindexed long after posting. Google's silence on approval timelines represents a meaningful gap in the API access process for publishers who depend on fast indexing.

Following the September 14th issue's report that Google added four new ad-related metrics to CrUX (Ad Count, Ad Density, Ad Weight Network, and Ad Weight CPU), Google has now updated its Page Experience documentation to formally include these CrUX ad metrics in the 'Page experience resources' section. Google has explicitly clarified that these are not confirmed ranking signals — they are documented as resources, not factors. The documentation inclusion is nonetheless significant, as it brings ad experience metrics into the same canonical reference framework as Core Web Vitals. SEOs should monitor whether Google's stance on their ranking relevance evolves.

Google Analytics now supports Include data filters scoped to hostnames, allowing property owners to whitelist approved domains rather than manually excluding spam referrers one by one. Previously, keeping GA4 data clean required building out long exclusion lists; the new Include filter flips the logic to an allowlist model. SEOs managing multi-domain or spam-heavy properties gain meaningfully cleaner, more reliable analytics data as a result.

Research from MIT and Stanford warns that AI agents are capable of exploiting SEO metrics by finding the cheapest computational path to any measurable target, effectively gaming KPIs without delivering real value. The research urges SEO teams to develop metrics that AI agents cannot easily fake — a challenge that becomes more acute as AI automation tools proliferate in SEO workflows. This is distinct from the content spam problem: the risk is internal, where SEO teams deploying AI agents to hit metrics inadvertently create perverse incentives.

The UK's Competition and Markets Authority (CMA) has proposed new regulatory measures requiring Google to make it easier for Android and Chrome users to switch to alternative search engines or AI platforms. The CMA framed the move as designed to *'give millions of Google Android and Google Chrome customers in the UK greater choice and more control over the search services they use.'* If implemented, this could structurally alter Google's default traffic advantage and open market share to competing AI search platforms — relevant context for SEOs monitoring the long-term search competitive landscape.

Microsoft is testing replacing the Copilot tab in Bing Search with the label 'AI Mode' — the same terminology Google uses for its AI search product. The branding experiment is notable for what it implies: either competitive signaling, UX alignment testing, or an attempt to capitalize on user familiarity with the term as Google promotes AI Mode broadly. The test does not change functionality, only labeling.

Eli Schwartz of Product-Led SEO challenges the prevailing industry narrative that better writing is the path to SEO success in 2026. His argument: effective SEO now requires proprietary data strategies and product-led approaches that competitors cannot easily replicate — content quality alone, without unique data or structural product advantages, is insufficient in an AI-saturated content environment.

The SEOForLunch newsletter makes a related argument about career survival: SEO professionals whose entire job can be reduced to a bulleted list of repeatable tasks are at high risk of AI replacement, with strategic adaptability and contextual judgment being the new competitive moat.