Google Issues Clear 2026 Guidance: No Separate SEO for AI Overviews or AI Mode
Generative search has spawned a new SEO industry faster than search engines could explain the rules. Site owners are being advised to create llms.txt files, rewrite articles for neural networks, split text into special blocks, publish hundreds of pages targeting long questions, and buy brand mentions on forums. Some recommendations are harmless, others useless, while certain tactics risk pushing sites into the same territory Google has spent years trying to clear of doorway pages and mass low-quality content.
By August 2026 Google’s position became much clearer. The company released separate guidance on optimization for generative search functions and explicitly stated that no dedicated SEO for AI Overviews and AI Mode exists. From Google’s perspective, visibility work in generative search remains ordinary search optimization. A page must be reachable by the crawler, enter the index, match user intent, and contain information Google considers useful and sufficiently trustworthy.
Yet the conclusion “nothing has changed” would also be incorrect. Classic search mainly selected documents and showed them to the user. Generative search first selects documents, then extracts facts, compares information, and assembles its own answer. A site now competes at several stages simultaneously: it must be found, chosen among candidates, used as a source, and ideally receive a link the user actually clicks.
How Google Collects AI Answers and Why This Changes SEO
AI Overviews and AI Mode rely on Google’s search index and core ranking systems. The generative system receives relevant pages from search, extracts information, and uses the material when constructing responses. Consequently, no separate “AI index” exists that site owners can target with special techniques.
A second key mechanism Google calls query fan-out. The model can take one complex question and run multiple additional searches in parallel. A user asking which backup system a small company should choose after a ransomware attack may trigger separate searches on recovery speed, copy isolation, cost, cloud storage, compatibility, and protection of administrative accounts.
The old notion that “one query equals one SEO page” begins to break down. Site owners cannot and should not guess every additional question the model will formulate behind the scenes. It is far more effective to build a thematic content system where the search engine can reliably find good answers to different parts of the user’s task.
Do Not Turn Query Fan-Out into a Landing-Page Factory
Generative search unexpectedly returns practitioners to rules that have existed for more than a decade. Google has long fought doorway pages. Modern versions look more polished: instead of thousands of pages titled “laptop repair Moscow,” “laptop repair Khimki,” and “laptop repair Mytishchi,” creators are now encouraged to publish thousands of pages targeting questions an AI might supposedly generate. The mechanics remain unchanged.
Google’s spam policies specifically address scaled content abuse. It does not matter whether pages are written by humans, templates, or generative models. The problem arises when a site produces large volumes of low-originality content primarily to manipulate search coverage.
Landing Pages Are Not Dead — Poor Landing Pages Simply Became Weaker
Good landing pages solve the user’s task directly rather than serving as doors to other pages. The typical commercial landing page suffers from informational emptiness: vague promises, identical benefits, hidden pricing, and missing limitations. For a generative system such a document is almost useless.
Google recommends creating non-commodity content — information that cannot be easily replaced by dozens of similar publications. Own experiments, statistics, interviews, product tests, and detailed real-world implementation analyses are far harder to substitute.
Technical Optimization Starts with Basics
Before experimenting with GEO techniques, site owners should open Search Console and perform a standard technical audit: verify indexing of important URLs, check robots.txt and CDN protections, review noindex and canonical tags, ensure correct server response codes, maintain an up-to-date XML sitemap, and confirm mobile usability. No special technical requirements have been introduced for generative features.
Structured Data Remains Useful but No AI-Specific Schema Exists
Google states that structured data are not a separate requirement for AI Overviews or AI Mode. Standard Schema.org markup for Product, Person, Article, Organization, and LocalBusiness continues to help the engine understand content and participate in rich results.
Measurement and Future Outlook
Google has begun rolling out dedicated reporting in Search Console for generative visibility. Site owners are advised to track indexing, generative impressions, ordinary search performance, traffic from AI services, and downstream conversions rather than relying on a single metric. The next evolution involves AI agents that not only answer questions but also perform actions, making semantic HTML, accessible forms, and clearly structured data even more valuable.
Related articles
Wildberries and Ozon to Accept Digital Rubles from September 1, 2026
Major Russian marketplaces Wildberries and Ozon will begin accepting digital rubles for purchases starting September 1, 2026. Ozon plans to introduce the new payment option first in a testing phase alongside bank cards and other methods before gradually rolling it out to all customers. Wildberries confirmed it will comply with requirements from the Bank of Russia and existing regulations without providing further details. From the same date, major banks must enable clients to open digital wallets, transfer funds, and pay for goods through their regular banking apps connected to the Central Bank platform. Operations with digital rubles will remain free for citizens. The rollout is mandatory for large merchants with annual revenue exceeding 120 million rubles that are served by major banks, with smaller businesses facing obligations in subsequent years based on revenue thresholds.
Positive Technologies Wraps Up Third Positive Hack Camp with 81 Students from 20 Countries
Positive Technologies has completed the third edition of its international Positive Hack Camp cybersecurity training program. More than 80 students from 20 countries traveled to Moscow at their own expense to receive intensive instruction in ethical hacking and information security. The two-week program featured 30 practical sessions delivered by experts from Positive Technologies and invited instructors from Lebanon. Participants studied attack modeling, network traffic analysis, privilege escalation in web applications, operating system security, and reverse engineering of complex devices. The camp also included cultural activities and concluded with examinations where all students passed and received diplomas. The initiative aims to strengthen global cyber resilience and build an international community of white hackers, with previous participants already contributing to national cybersecurity efforts in Algeria.
Rubytech Tests Russian LLM Cotype 3 on Chinese GPUs Matching NVIDIA H100 Performance
Rubytech has successfully tested the Russian large language model Cotype 3 developed by MWS AI on Chinese graphics accelerators instead of NVIDIA hardware. The trials took place on the Skala^r AI machine software-hardware complex and demonstrated stable operation with inference performance comparable to configurations using NVIDIA H100 GPUs. Eight Chinese GPUs were used in the test system, delivering an average first-token generation time of about 8 seconds and inter-token latency between 111 ms when processing 27,000-token contexts. Engineers adapted drivers, software environments, and orchestration tools, achieving up to 2-2.2 times faster performance in certain scenarios compared to the standard Chinese GPU setup. The results indicate that corporate AI infrastructure can be built without mandatory reliance on NVIDIA accelerators, which is particularly relevant for government agencies, state-owned companies, and critical information infrastructure organizations requiring localized model deployment and controlled scaling. Rubytech also noted potential reductions in total cost of ownership with Chinese GPUs, although the company emphasized this is an additional option rather than a complete replacement for NVIDIA solutions.
Can the Moon Be DDoSed? Exploring Future Cybersecurity Challenges for Lunar Bases
A speculative analysis examines whether DDoS attacks could target a lunar base in 2036, highlighting how Earth-based internet protocols would fail over the 384,000 km distance. The piece details NASA’s LunaNet architecture and Delay/Disruption Tolerant Networking (DTN) that uses store-and-forward mechanisms to handle long delays and link disruptions. It explains why traditional volumetric attacks would be limited by narrow space communication channels while resource-exhaustion attacks on orbital relays and DTN nodes could prove more effective. The analysis contrasts Earth’s redundant routing with the Moon’s few irreplaceable links and emphasizes that attacking ground control centers might be more practical than flooding lunar hardware. It concludes that any real lunar network will inherit familiar terrestrial bottlenecks despite its exotic setting.