Google AI Mode, AI Overviews and Search Console: how to prepare your website in 2026

In 2026 Google Search has accelerated its shift toward a more conversational, generative, and multimodal experience. Google says AI Mode has surpassed one billion monthly users, AI Overviews reaches more than 2.5 billion, and the company is introducing new controls and reporting for website owners. For a team running WordPress, WooCommerce, or a corporate site, the useful question is not whether SEO should be abandoned in favor of “GEO.” The useful question is how to adapt a strong SEO foundation to a search environment where people ask more complex questions, continue conversations, use images, and receive answers assembled from multiple sources. The right strategy preserves technical and editorial fundamentals, adds structures that improve understanding and retrieval, and above all measures real results before multiplying pages.

1. What actually changed in Google Search during 2026

In May Google introduced a new phase of Search built around AI Mode, a more conversational search box, and agents that can work across information from the web. The company says AI Mode has surpassed one billion monthly users and that queries have more than doubled each quarter since launch. This does not mean traditional results have disappeared. It means a single user can move from a short query into a conversation, refine constraints, request comparisons, and open specific sources from a generative response.

For site owners, the biggest change is the amount of context a page may need to contribute. A URL that only repeats a keyword and generic commercial proposition is less useful when the query asks for details, limits, steps, examples, or comparisons. Pages that explain a topic clearly, expose recognizable entities, and connect to complementary resources are more likely to be useful both in classic results and generative experiences. Editorial strategy should reflect that reality without overoptimizing for one interface that will keep evolving.

• More complex and follow-up queries.
• Greater need for context and self-contained answers.
• Links to the web remain part of generative search experiences.

2. GEO does not replace SEO: Search infrastructure still matters

One of the most common 2026 misconceptions is presenting GEO as a replacement for SEO. Google's documentation does not support that interpretation. Generative features are part of Search and remain subject to quality systems, spam policies, preview controls, crawling, and indexing requirements. A page blocked by robots, marked noindex, served with errors, or canonicalized elsewhere does not gain an advantage because it adds an agent-oriented layer.

A better way to understand GEO is as an extension of the discipline. SEO addresses how to make a page crawlable, indexable, relevant, and competitive. AI-search readiness places additional emphasis on explicit definitions, entity relationships, concise answers, semantic structure, sources, machine-friendly versions where appropriate, and content capable of resolving longer queries. All of that works better on a stable technical foundation. Serious work does not start with a new file; it starts by ensuring the primary page deserves to exist and can be processed correctly by Google.

3. Design pages to answer complete intents, not repeat keywords

Conversational queries expose thin content quickly. If someone asks how to prepare WooCommerce for AI search, they do not need ten pages repeating “AI SEO WooCommerce” with small variations. They need to understand what can be controlled, what should be reviewed, which technical signals matter, how products should be structured, which limitations exist, and how results can be measured. A page that covers that intent well can include many semantic variants without turning every variant into a separate URL.

This changes how clusters should be planned. Start by researching a broad keyword universe across products, problems, features, industries, roles, integrations, questions, and geography. Then group terms by intent. Create a new URL only when there is a clearly different need. This reduces cannibalization and doorway pages, improves internal linking, and makes maintenance easier. It also helps generative systems because each URL has an understandable role within the site instead of belonging to a collection of nearly identical pages.

4. Entities, structured data, and visible consistency

Structured data remains a tool for helping Google understand specific information on a page. It does not guarantee rich results and should not describe content that does not exist. The practical rule is simple: create a useful visible page first, then mark up what is actually there. For a software landing page, Organization, WebPage, BreadcrumbList, and SoftwareApplication may be relevant. For an article, BlogPosting and BreadcrumbList can describe author, dates, title, and relationship to the site.

Consistency matters as much as the schema type. If JSON-LD names a product, the visible page should use the same product name clearly. If an updated date changes, the content should genuinely have been reviewed. If an entity has a canonical URL, internal links should point to that reference instead of splitting signals across routes. Kairoseth's reusable engine centralizes these elements so every new publication inherits the same semantics rather than depending on copied markup.

5. Internal links: turn the site into an understandable graph

Internal links serve two important functions. For users, they provide a natural path to more context, a product, or an action. For crawlers, they help discover URLs and understand relationships. In an AI-search-oriented architecture, that graph should be deliberate: a commercial landing connects to the product, its supporting article, the relevant hub, and other resources in the same cluster. The article links back to the landing while explaining the problem in greater depth.

Anchor text also matters. “Click here” provides less context than “AI Search Optimizer for WordPress” or “2026 Google AI Mode guide.” There is no need to force exact-match anchors everywhere; the text only needs to describe the destination naturally. This discipline prevents orphan pages and allows a system to understand that product, problem, solution, guide, and geography belong to the same topic without mechanically repeating one keyword.

6. On September 24 Google added multimodal reporting: what changes for SEO

Google Search Central announced dedicated multimodal search reporting in Search Console today. The new filter makes it possible to analyze traffic from Lens, Circle to Search, images uploaded to Google, and searches initiated from an image in Chrome. This matters because it makes a previously difficult-to-isolate part of search behavior measurable. For ecommerce, retail, travel, interiors, fashion, and other visual sectors, the data can reveal new discovery paths.

The answer is not to add more images without purpose. A multimodal strategy needs photography or graphics that clearly represent what the page explains, reasonable filenames, useful alt text for accessibility, correct dimensions, good performance, and semantic alignment with the content. On product pages, image, name, price, description, and structured data should describe the same object. Search Console can then show which pages gain multimodal visibility so teams improve where real demand exists instead of optimizing blindly.

7. Search Console should decide what content to expand

Publishing does not finish the job. A new page first needs to be discovered, crawled, and indexed. Then impressions and queries begin to show whether Google associates it with the intended search need. Teams should review indexing, queries, CTR, position, countries, devices, and now generative or multimodal surfaces when reporting is available. These data points turn a hypothesis-based strategy into an evidence-based one.

For example, if a Barcelona AI Search landing starts receiving impressions for WooCommerce-related queries, a specific product article may be justified. If impressions come from “llms.txt WordPress,” the relevant section can be expanded or a separate guide can be created if the intent is clearly different. If a keyword receives no visibility, it should not automatically be multiplied into ten geographic pages. First determine whether authority is missing, the intent is poorly addressed, or there simply is not enough demand.

8. Where llms.txt v2 fits—and where it does not

llms.txt emerged as a proposal for helping models and agents find relevant resources and readable versions of a site. The v2 revision supports path-scoped indexes and discovery relationships to Markdown alternatives. Kairoseth uses that architecture because it reduces friction for consumers that understand the convention: HTML remains canonical while each content item can expose a Markdown version derived from the same source plus a curated language-specific index.

Google clarified in June, however, that llms.txt is not required for Search and does not positively or negatively affect rankings. It therefore should not be used as a “guaranteed ranking” argument. Its role is complementary: make retrieval easier for other systems without manually duplicating content. Success in Google still depends on canonical URLs having quality, accessibility, strong linking, and clear intent satisfaction. Maintaining both layers is reasonable as long as their purposes are not confused.

9. What this means for a business in Barcelona or Spain

AI Mode is available in Spain, and local businesses increasingly compete within experiences where commercial intent and contextual information are mixed. A Barcelona clinic, agency, ecommerce company, B2B software provider, or travel business may receive questions combining location, budget, integration, language, or specific constraints. Content should be able to answer those combinations without creating an almost identical landing page for every imaginable phrase.

Useful geographic work adds real context: service availability, languages, applicable regulation, local ecosystem, logistics, relevant industries, examples, or specific processes. A “Barcelona” page should explain why Barcelona changes the decision. The same applies to Catalonia, Spain, or Europe. This discipline allows clusters to scale in an organized way while avoiding doorway pages. For multilingual companies, hreflang, localized slugs, and semantic parity across languages belong in the architecture rather than being patched later.

10. A practical 30-day AI Search readiness plan

During the first week, review the technical foundation: robots, sitemap, indexing, canonical, speed, 4xx/5xx errors, metadata, and structured data. The second week can focus on editorial architecture: map products, primary pages, intents, and overlapping content. During the third week, improve high-potential pages with concise answers, substantive sections, internal links, sources, images, and coherent structured data. The fourth week should focus on measurement and identifying which hypotheses deserve expansion.

On WordPress and WooCommerce this cycle can repeat monthly. AI Search Optimizer helps review public signals, select content, and maintain a verifiable machine-friendly layer. Kairoseth's editorial engine applies the same idea to publishing: one template generates canonical, hreflang, schema, sitemap, llms.txt v2, Markdown, and linking while the editor focuses on intent, quality, and evidence. The ultimate goal is not to produce more pages but to give each page a clear purpose and let it evolve with data.

11. Mistakes to avoid when optimizing for AI search

The first mistake is mass-publishing AI-generated content without review and turning every keyword into a separate page. The second is hiding text “for bots” that provides no visible user value. The third is adding structured data that describes services, ratings, or questions not actually present on the page. The fourth is assuming llms.txt replaces sitemap, robots, or internal architecture. The fifth is measuring success only by the number of published URLs rather than indexing, queries, clicks, and conversions.

It is also important to avoid claims that cannot be verified, such as “guaranteed to appear in ChatGPT” or “number one in AI Overviews.” Generative systems change, select sources dynamically, and do not give publishers total control. A serious value proposition focuses on improving technical and editorial readiness, reducing ambiguity, facilitating discovery, and preserving evidence of what was published. That position may sound less dramatic than an absolute promise, but it is far more sustainable for a company building visibility over years.