For many years, the model was relatively simple:

Query ⇨ Search results page ⇨ Click ⇨ Website

Today, this journey can no longer be taken for granted.

Google, Bing, and emerging AI-powered search platforms are increasingly able to understand complex questions, perform multiple searches, compare sources, and generate answers directly for users.

Google has introduced and progressively expanded AI Overviews and AI Mode. Microsoft is integrating search with Bing and Copilot, while systems such as ChatGPT Search are further transforming the concept of a search engine by combining web search with AI-generated answers.

The result is a significant shift in user behaviour: for many informational queries, users no longer need to visit multiple websites to build an answer themselves.

Search engines are therefore evolving from lists of documents into systems for retrieving, understanding, and synthesising information.

This does not mean that SEO is dead.

It means that its purpose is expanding.

We should no longer ask only:

“How can I rank this page?”

but also:

“How can I make this page a reliable source that a search engine or AI system can find, understand, retrieve, and cite?”

From SEO to AEO and GEO

In recent years, terms such as SEO, AEO, and GEO have become increasingly common.

SEO: Search Engine Optimization

Optimising websites and content to gain visibility in traditional search engines.

AEO: Answer Engine Optimization

Optimising content so that it can directly contribute to the answer provided to the user.

GEO: Generative Engine Optimization

Optimising a website’s presence within AI-powered and generative search experiences.

However, these should not be considered three completely separate disciplines.

SEO remains the foundation.

Google has clarified that AI Overviews and AI Mode continue to rely on its core Search ranking and quality systems and that no special technical requirements are needed to appear in generative search features.

Microsoft is also increasingly using the term GEO to describe how content participates in AI experiences and is selected as a source for generated answers.

We can therefore represent this evolution as:

SEO ⇨ Discovery

AEO ⇨ Answer

GEO ⇨ Citation & Generative Visibility

SEO enables content to be discovered.

AEO and GEO extend this concept towards making content easier to understand, retrieve, and use within generative answers.

Types of Queries and the Risk of Traffic Loss

The impact of generative search is not the same for every type of query.

Informational Queries: HIGH RISK

These are queries where the user is primarily looking for an explanation.

For example:

• What is the OAuth2 protocol?
• What does canonical URL mean?
• How does a CDN work?
• What is the difference between SEO and SEM?

For these queries, a generative answer can often provide a sufficiently complete explanation immediately.

The consequence is an increase in zero-click searches: searches that end without the user visiting an external website.

Comparative and Research Queries: MEDIUM/HIGH RISK

AI platforms are particularly effective at comparisons.

A question such as:

“Which CMS is better suited to an international healthcare portal: Drupal or WordPress?”

may require the system to compare:

• security;
• scalability;
• multi-country management;
• editorial workflows;
• SSO;
• privacy;
• integrations;
• costs;
• maintenance.

Google AI Mode and other generative systems can perform multiple searches to build this comparison directly.

Clicks remain important when users want to verify a source, explore a technical aspect in greater depth, read about real-world experience, or analyse more detailed data.

Transactional Queries: LOWER RISK

When users need to complete an action, websites continue to play a fundamental role.

For example:

• purchasing a product;
• booking a hotel;
• requesting a quote;
• signing up for a service;
• downloading software;
• contacting a company.

However, this area is evolving as well.

AI Agents are progressively transforming artificial intelligence from a system that simply provides answers into a technology capable of interacting with services and applications.

The evolution could therefore be represented as:

Search ⇨ Answer ⇨ Conversation ⇨ Action

How Generative Search Works

Not every search engine uses the same architecture, and we do not know every detail of how their systems work.

However, we can identify a general model.

1. Understanding Intent

The system interprets the request and attempts to understand:

• meaning;
• context;
• entities;
• intent;
• relationships between concepts.

As a result, an exact match between the keyword entered by the user and the text on a page is becoming less important.

2. Query Fan-Out

One of the most interesting features of generative search is query fan-out.

Google has officially confirmed that AI Overviews and AI Mode can perform multiple related searches across different subtopics and data sources.

A complex query can therefore be broken down into several secondary searches.

For example:

“What is the best CMS platform for a European HCP portal?”

could generate searches related to:

healthcare CMS

Drupal security

WordPress healthcare

GDPR CMS

multi-country CMS

HCP authentication

SSO healthcare portal

Content therefore does not necessarily need to match the exact wording of the user’s original query to become a relevant source.

Retrieval, Grounding, and RAG

One of the fundamental concepts behind AI-powered search is grounding.

A Large Language Model contains knowledge acquired during training, but when answering questions that require current information, it can use external sources.

One approach used for this is Retrieval-Augmented Generation (RAG).

In simplified terms:

Query
⇩
Retrieval
⇩
Source Selection
⇩
Information Extraction
⇩
Answer Generation
⇩
Citations / Links

Google explicitly confirms the use of RAG and grounding techniques within its generative search features.

Microsoft also considers grounding a central component of the AI-powered web: web sources allow generative systems to connect their answers to up-to-date and verifiable information.

For SEO and content professionals, this means something important:

before content can be used in an AI-generated answer, it needs to be discoverable, understandable, and considered sufficiently relevant.

This is precisely where traditional SEO continues to play a fundamental role.

Google, Bing, and the New Search Landscape

This transformation is not limited to Google.

Google

Google primarily operates through:

• traditional Search;
• AI Overviews;
• AI Mode.

AI Overviews synthesise information directly within the SERP when Google determines that doing so adds value to the search experience.

AI Mode extends this model further by enabling more complex searches, comparisons, and conversational follow-up questions.

Both can use query fan-out to discover sources that may differ from those surfaced by a single traditional search query.

Bing and Microsoft Copilot

Microsoft is following a similar direction through Bing and Copilot.

One particularly interesting development is the introduction of AI Performance in Bing Webmaster Tools.

The report allows publishers to analyse metrics such as:

• total citations;
• cited pages;
• grounding queries;
• citation trends over time.

Microsoft is therefore beginning to distinguish more explicitly between traditional SEO visibility and visibility within AI-generated answers.

This may provide an early indication of how SEO analytics tools will evolve over the coming years.

ChatGPT Search and Emerging AI Search Engines

Alongside traditional search engines, we are seeing the growth of systems where search is integrated directly into an AI assistant.

ChatGPT Search, for example, can search the web and use external pages as sources for its answers.

For publishers, this introduces another dimension of organic visibility.

There is no longer only:

Google ranking

but rather an ecosystem consisting of:

Search visibility

AI visibility

Citation visibility

Brand visibility

A website can therefore generate value even when its content is used as a source without necessarily generating an immediate click.

From Keywords to Topics, Intent, Entities, and Context

Keywords are not disappearing.

Their strategic role, however, is changing.

The traditional approach:

One keyword ⇨ One page

is becoming less effective when applied too rigidly.

A modern approach should instead consider:

Keyword ⇨ Topic ⇨ Intent ⇨ Entity ⇨ Context

Google has also clarified that there is no need to create dozens of pages simply to cover every possible long-tail variation.

Modern search systems can understand synonyms, concepts, and entities even when there is no literal match with the user’s query.

Building comprehensive semantic coverage of a topic is therefore becoming increasingly important.

How to Write Content for AI-Powered Search

There is no magic formula for getting cited.

Nor is there an ideal paragraph length that guarantees inclusion in an AI-generated answer.

The objective should be to make content easy to understand for both people and information retrieval systems.

A good structure can include:

• descriptive headings;
• clear paragraphs;
• direct answers when appropriate;
• tables for comparing data;
• lists when they improve readability;
• precise definitions;
• examples;
• verifiable sources;
• connections between related concepts.

The old idea of an Atomic Answer that must always be 40-60 words long should therefore not be considered an SEO rule.

The answer simply needs to be clear, complete, and proportionate to the question.

The Real Competitive Advantage: Non-Commodity Content

Artificial intelligence makes it extremely easy to produce generic content.

And that is precisely why generic content is losing value.

Google uses the concept of non-commodity content to describe content that adds something beyond information that is already widely available.

This may include:

• first-hand experience;
• tests;
• benchmarks;
• proprietary data;
• original research;
• case studies;
• screenshots;
• real-world examples;
• well-supported expert opinions;
• internally developed methodologies.

An article that simply repeats:

“10 benefits of cloud computing”

can easily be summarised by an AI system.

An article presenting:

“How we reduced response times by 38% by migrating this architecture to Azure”

contains information that the model is unlikely to find elsewhere.

This distinction will become increasingly important.

E-E-A-T in AI-Powered Search

Experience, Expertise, Authoritativeness, and Trustworthiness continue to represent important concepts for assessing content quality.

However, they should not be interpreted as a single “AI ranking factor”.

The fundamental principle remains demonstrating that the content is supported by genuine experience, expertise, and verifiable information.

Experience

Demonstrate real-world experience through:

• tests;
• real cases;
• screenshots;
• original photos;
• problems encountered;
• results achieved.

Expertise

Demonstrate expertise through:

• technical insights;
• appropriate terminology;
• sources;
• data;
• accurate explanations;
• clear author profiles.

Authoritativeness

Build authority through:

• citations;
• backlinks;
• mentions;
• publications;
• industry reputation.

Trustworthiness

Strengthen trust through:

• verifiable sources;
• identifiable authors;
• editorial transparency;
• publication and update dates;
• clear company information;
• website security.

Structured Data: Important, but Without the Myths

Schema.org and structured data remain important SEO tools.

They help search engines understand certain types of information and may enable specific rich results.

However, they should not be considered a requirement for being used in AI-generated answers.

Google has clarified that there is no special Schema.org markup for AI Overviews or AI Mode.

Structured data should therefore be implemented when it accurately represents the content on the page and provides genuine value to Search.

For example:

• Article;
• Product;
• Organization;
• Person;
• Event;
• BreadcrumbList;
• LocalBusiness.

The principle remains simple:

Use structured markup when it adds meaning; do not create artificial markup simply to “convince the AI”.

Images, Video, and Multimodal Content

Search is becoming increasingly multimodal.

Answers can include:

• text;
• images;
• video;
• products;
• maps;
• structured information;
• interactive elements.

Google explicitly recommends complementing textual content with high-quality images and videos where appropriate.

It is therefore important to consider:

• image quality;
• alt text;
• file names;
• surrounding text;
• captions;
• video SEO;
• image/video sitemaps where appropriate;
• consistency between visual and textual content.

An original diagram or technical screenshot can provide significantly greater informational value than a generic stock image.

Topic Clusters and Internal Linking

Generative search further reinforces the importance of topical authority.

A single isolated article is unlikely to demonstrate the same depth as a connected content ecosystem.

An effective structure might look like this:

Pillar Page

Deep Dive A

Deep Dive B

Case Study

FAQ

Tutorial

Comparison

Internal linking helps search engines understand:

• relationships between content;
• hierarchy;
• context;
• topical depth.

It also improves crawling and content discovery.

Technical SEO: What Does Not Change

Artificial intelligence does not eliminate the need for a technically solid website.

The fundamentals remain essential:

• crawlability;
• indexability;
• robots.txt;
• XML sitemaps;
• canonical tags;
• correct redirects;
• HTTP status codes;
• performance;
• mobile usability;
• accessibility;
• JavaScript rendering;
• internal linking;
• understandable HTML;
• important content available in textual form.

If a search engine cannot properly retrieve content, it is unlikely to be able to use that content as a source.

The New SEO KPIs

One of the biggest mistakes will be continuing to measure success exclusively through organic traffic.

Traditional KPIs remain important:

• rankings;
• impressions;
• CTR;
• clicks;
• organic sessions;
• conversions.

However, they increasingly need to be complemented by new metrics:

• citations in AI-generated answers;
• cited pages;
• branded searches;
• brand mentions;
• direct traffic;
• referrals from AI platforms;
• engagement;
• conversion rate;
• visibility for strategically relevant queries.

Bing has already started this transition through AI Performance in Bing Webmaster Tools, providing data on citations and grounding queries from Copilot, Bing AI, and selected partner experiences.

Google, on the other hand, currently includes AI Overviews and AI Mode within the overall Web search traffic reported in Search Console.

Measuring AI visibility is therefore still a rapidly evolving area.

SEO Is Not Disappearing: Its Objective Is Changing

For years, the primary SEO objective was:

Achieve a ranking
⇩
Generate a click
⇩
Bring the user to the website

Today, the journey can increasingly become:

Be discovered
⇩
Be considered relevant
⇩
Be retrieved as a source
⇩
Be cited
⇩
Build trust
⇩
Generate a visit or an action

This fundamentally changes how we should assess the value of content.

Content can contribute to brand visibility even before a click occurs.

Conclusion

SEO is not dead.

On the contrary, it remains part of the infrastructure on which much of the new AI-powered search ecosystem is being built.

What is changing is the context.

We are no longer optimising solely for a results page made up of ten blue links.

We are optimising for an ecosystem where several different experiences coexist:

Search Engine

Answer Engine

Generative Engine

AI Assistant

AI Agent

Google AI Overviews, AI Mode, Bing, Microsoft Copilot, and emerging AI platforms are progressively transforming how information is discovered and used.

The future of SEO is therefore not about finding tricks to “please LLMs”.

It is about building technically accessible websites and creating content that has something genuinely valuable to offer:

experience, data, expertise, clarity, authority, and original information.

In the past, the main objective was to get the click.

Today, the objective is also to become a source for the answer.

Tomorrow, it may be about becoming both a source and a service that AI agents can interact with directly.

Primary Sources:
Google Search Central – AI features and your website,Google – Generative AI optimization guide,Bing Webmaster – AI Performance

L'articolo How SEO Is Changing with the Advent of AI Overview, AI Mode, and Generative Search proviene da Giovanni Anfuso.

First came chatbots, then AI-powered coding assistants such as GitHub Copilot, and today we have entered the era of AI Agents-systems capable not only of answering questions but also of interacting with external tools, databases, Git repositories, CRMs, and enterprise applications.

In this rapidly evolving landscape, a new standard is emerging that is likely to become increasingly important: Model Context Protocol (MCP).

Although it is still relatively unknown today, it will likely become one of the fundamental concepts every software developer should understand.

What is Model Context Protocol?

Model Context Protocol, commonly referred to as MCP, is an open protocol designed to enable AI models to communicate with applications, services, and data sources through a standardized interface.

Simply put, it provides a common language between AI models and the tools we use every day.

Let’s look at a simple example.

Imagine asking an AI assistant:

“Show me all the orders placed today by customer Rossi.”

Without MCP, the assistant would need to understand the ERP’s specific APIs, authentication methods, response formats, and custom integration logic.

With MCP, all of this is standardized.

The AI doesn’t need to know how every individual application works internally. Instead, it communicates through a shared protocol that any compatible service can expose.

Why was MCP created?

Until recently, every integration with a Large Language Model required custom development.

Every application exposed different APIs.

Every authentication method was unique.

Every response followed its own format.

The result?

Developers spent a considerable amount of time building integrations that were difficult to reuse.

Model Context Protocol was created to solve exactly this problem.

Just as HTTP standardized communication between web browsers and web servers, MCP aims to standardize communication between AI models and software applications.

How does it work?

The overall architecture is remarkably simple.

On one side, there is the AI model.

On the other, one or more MCP servers expose tools, data, and capabilities.

When a user submits a request, the AI model can:

• query databases;
• search documentation;
• interact with a CRM;
• browse Git repositories;
• read local files;
• call REST APIs;
• retrieve enterprise data;
• execute authorized operations.

The model doesn’t guess how to interact with external systems. It knows which tools are available and uses them through a standardized protocol.

This makes integrations significantly more reliable and predictable than custom-built solutions.

A real-world example

Imagine a company that uses:

• a CRM;
• an ERP;
• a customer database;
• a document management system;
• Git repositories;
• a ticketing platform.

Today, enabling an AI assistant to work with all these systems would require numerous custom integrations.

With MCP, each system can expose its capabilities as standardized tools.

The AI assistant can then:

• retrieve business data;
• compare information across systems;
• generate reports;
• create documentation;
• assist developers and support teams.

All without reinventing the integration logic every time.

Why developers should pay attention

Many people believe AI will reduce the need for software developers.

Personally, I think the reality is different.

What will change is the nature of our work.

We’ll spend less time writing repetitive code and more time designing architectures, defining workflows, integrating systems, and building tools that AI can safely interact with.

In this context, understanding MCP could become a real competitive advantage.

MCP and programming languages

One of MCP’s biggest strengths is that it is language-agnostic.

MCP servers can be implemented using virtually any modern programming language, including:

• PHP;
• C# and .NET;
• Java;
• JavaScript and Node.js;
• Python;
• Go;
• Rust.

This means existing applications can become AI-ready without requiring a complete rewrite.

Possible use cases

The possibilities are virtually endless.

Using technologies I’m most familiar with as examples, a Drupal website could expose capabilities that allow an AI assistant to:

• create content;
• search articles;
• edit pages;
• query Views;
• retrieve analytics.

An ASP.NET application could expose tools to:

• query databases;
• manage users;
• generate reports;
• retrieve documents.

A PHP-based ERP or business application could allow AI to:

• verify customer orders;
• check inventory availability;
• prepare quotations;
• retrieve invoices.

All through the same open standard.

The future of AI Agents

More and more companies are investing in AI Agents, and the difference compared to traditional chatbots is significant.

A chatbot answers questions.

An AI Agent gets work done.

It can use tools, make decisions within predefined limits, retrieve information, and collaborate with multiple business systems.

To do this effectively, however, it needs a standardized way to access external resources.

That is exactly where MCP comes in.

Is it worth learning today?

In my opinion, absolutely.

We’re at a stage that feels very similar to where REST APIs were fifteen years ago.

Back then, they seemed like a niche technology used by only a handful of developers.

Today, REST APIs are at the heart of virtually every modern application.

I believe Model Context Protocol could follow a similar path.

It won’t replace traditional APIs, but it will become an additional layer that makes it much easier for AI models to interact with the software we use every day.

Final thoughts

Artificial Intelligence isn’t just changing the way we write code.

It’s changing the way software systems communicate with one another.

I believe Model Context Protocol represents one of the first serious attempts to establish a shared standard between AI models and software applications.

For this reason, I believe that learning MCP today means preparing for a transformation that has already begun.

Just as cloud computing, Docker, Kubernetes, and REST APIs reshaped software engineering over the past decade, developers who start experimenting with MCP today will likely have a significant advantage when this technology becomes part of everyday software development.

Personally, I believe that in the coming years we’ll hear less about simple chatbots and far more about intelligent applications designed from the ground up to collaborate with AI through open standards such as Model Context Protocol.

L'articolo MCP: The Technology Few Are Talking About That Could Change the Way Developers Work proviene da Giovanni Anfuso.

In recent years, web development has been undergoing a radical transformation. The widespread adoption of Artificial Intelligence tools has changed the way developers write code, design architectures, and solve problems.

In this context, Vibe Coding has emerged: an approach to development where the focus shifts from manual implementation to guiding AI, while maintaining creative and technical control over the project.

But what exactly is Vibe Coding, and why is it becoming increasingly popular among developers?

What Is Vibe Coding

Vibe Coding is a programming approach based on:

• continuous interaction with AI systems
• intent- and goal-driven development
• rapid iteration and experimentation

The developer does not necessarily write every line of code from scratch but instead:

• describes the problem
• evaluates AI-generated solutions
• refines, corrects, and optimizes

Code becomes the result of a human–machine dialogue, rather than simple typing.

The Main AI Systems Used in Vibe Coding

Vibe Coding would not exist without a new generation of intelligent tools. Among the most relevant are:

GitHub Copilot
Probably the most widely used tool in everyday coding.
It suggests code in real time directly inside the editor, based on file and project context.

Ideal for:

• speeding up development
• reducing boilerplate code
• exploring alternative solutions

Main limitation: it does not always understand the project’s overall architecture.

ChatGPT
Commonly used for:

• complex logic design
• refactoring
• code explanation
• conceptual debugging

It is one of the most powerful tools in Vibe Coding because it enables reasoning about code, not just generating it.

Gemini
Highly effective for:

• understanding large contexts
• analyzing complex workflows
• integrating with cloud ecosystems

Often used for technical brainstorming and architectural evaluation.

Other Relevant Tools

• Codeium
• Tabnine
• Cursor
• Sourcegraph Cody

All contribute to an ecosystem where developers focus more on the why rather than the how.

How to Use Vibe Coding Effectively

Vibe Coding truly works only when applied with a structured approach. A healthy workflow includes:

• Clearly defining the problem
• Using AI to generate an initial solution
• Critically analyzing the produced code
• Adapting it to real project requirements
• Testing and validation

AI accelerates development, but it does not replace technical thinking.

How the Web Development Approach Has Changed

Before AI:

• linear development
• focus on syntax
• longer implementation times

With Vibe Coding:

• greater focus on architecture
• more time spent on strategic decisions
• extremely fast prototyping
• continuous experimentation

Developers increasingly become solution orchestrators, not just executors.

Seniority: Why It Still Matters (Maybe Even More)

A frequently underestimated aspect is that Vibe Coding amplifies skill differences.

An experienced developer:

• recognizes incorrect patterns
• identifies security issues
• evaluates performance and scalability impacts
• understands when AI produces fragile code

A less experienced developer risks:

• blindly trusting outputs
• accumulating technical debt
• losing understanding of the codebase

AI does not eliminate seniority — it makes it even more essential.

What to Watch Out for with Vibe Coding

Using Vibe Coding without awareness can create serious problems:

• insecure code
• inefficient solutions
• over-engineering
• dependency on AI
• loss of fundamental skills

The biggest risk is not AI itself, but losing control over the code.

Conclusion

Vibe Coding represents one of the most significant evolutions in modern web development. Tools like ChatGPT, GitHub Copilot, and Gemini are redefining how developers work.

However, one principle remains unchanged:

AI is an extremely powerful tool, not a substitute for expertise.

Those who learn to use it with critical thinking and solid technical foundations will gain a significant competitive advantage.

L'articolo Vibe Coding and Artificial Intelligence: How Web Development Is Evolving proviene da Giovanni Anfuso.

What is Gemini 3 Flash

An AI Model Built for Speed
Gemini 3 Flash is an advanced AI model that combines “frontier-level” reasoning capabilities with extremely low latency. Google’s goal is clear: provide intelligent, contextual responses almost instantly, making the user experience smoother and more natural.

“Gemini 3 Flash: frontier intelligence built for speed”
— Google, official blog

Differences Compared to Previous Models

Compared to Gemini 2.5 Flash, the new version offers:

• Faster response times
• Improved reasoning abilities
• Lower usage costs
• Greater efficiency for high-traffic applications

Key Features of Gemini 3 Flash

Advanced Multimodal Intelligence
Gemini 3 Flash can understand and generate content from:

• Text
• Images
• Audio
• Video

This makes it ideal for virtual assistants, content analysis tools, and interactive AI applications.

Reasoning and Contextual Understanding
The model maintains strong problem-solving, planning, and complex analysis capabilities, proving effective in areas such as:

• Programming
• Decision support
• Complex information processing

Availability of Gemini 3 Flash

Integration in the Gemini App
Google has made Gemini 3 Flash the default model in the Gemini app, improving everyday user experience with faster and more accurate responses.

Gemini 3 Flash on Google Search
The new model is also integrated into AI Mode on Google Search, providing smarter, deeper, and more contextual results for complex queries.

Developer Tools
Gemini 3 Flash is available through:

• Gemini API
• Google AI Studio
• Vertex AI
• Android Studio
• Gemini CLI

This broad availability allows developers to create scalable and high-performance AI applications.

Benefits for Users and Enterprises

For End Users

• Faster responses
• Higher accuracy
• Improved conversational experience

For Developers and Businesses

• Lower costs
• High performance
• Easy integration
• Support for real-time use cases

Conclusion

With Gemini 3 Flash, Google takes a decisive step toward faster, more accessible, and efficient AI. The new model demonstrates how cognitive power and speed can be combined, making AI an increasingly central tool in daily digital life and in the development of innovative solutions.

Source: Google Blog

L'articolo Google Unveils Gemini 3 Flash: Ultra-Fast AI with Advanced Intelligence proviene da Giovanni Anfuso.