Google AI Overviews: come funziona e tutela dei dati

AI Overviews is Google Search’s feature that synthesizes responses at the top of the SERP, combining generative models (Gemini), structured knowledge and relevant online content. The goal is to reduce the steps needed to navigate complex queries, offering an immediate overview with links for deeper exploration. This shift of the “first answer” toward a generated overview opens reflections on publishers’ rights, algorithmic transparency, competitive dynamics and personal data protection, topics now at the center of institutional discussion in Italy and Europe.

What AI Overviews is and how it works

AI Overviews generates an overview that precedes organic results, presenting a coherent and contextual text reconstructed from multiple information signals. The variant indicated as AI Mode applies a query decomposition strategy, similar to “query fan-out”: the query is divided into sub-questions and investigated in parallel, with final recomposition into a unified output that integrates definitions, operational steps and references to sources. In practice, search becomes an interaction that “speaks” and guides toward further exploration, maintaining links to sources on the page.

Key points of the process:

Query decomposition: the system identifies salient aspects and generates simultaneous micro-searches to cover relevant sub-topics.
Evidence selection: information is integrated from indexed resources and signals of relevance, quality and authority.
Generative synthesis: the model reorganizes content into a unified output, with follow-up suggestions that guide deeper exploration.
SERP presentation: the overview occupies a prominent position, while links to sources appear in less central areas of the page.

This transition from “list of links” to a “composed answer” is a discontinuity: the ability to satisfy informational intent already in the first block reduces necessary clicks and modifies traffic dynamics toward original sites.

Impacts on the information ecosystem and traffic

Industry analyses in the 2024-2025 biennium signal an increase in searches that conclude without clicks to external sites (zero-click) and a reduction in organic click-through rate in various verticals. The most recurring estimates place zero-click between 60% and 70%, with average CTR reductions between 15% and 25% depending on the sector, indicating greater attention to synthetic responses that often satisfy basic informational needs. These dynamics affect organic results visibility, traffic-linked advertising revenues and the sustainability of editorial models.

Operational implications:

Organic visibility: a generative overview at the top of the SERP reduces the competitive space of traditional results.
Monetization models: fewer accesses can impact advertising KPIs, pushing toward more targeted editorial strategies.
Content design: increases the value of clear, citable information blocks with quality signals, capable of being referenced in the overview.

The FIEG complaint and legal profiles in focus

On October 15, 2025, FIEG submitted a complaint to Agcom, as National Coordinator under art. 49 of the Digital Services Act (DSA), to examine AI Overviews’ framework in light of related rights, algorithmic transparency and competition. The issue concerns the balance between rapid access to information and proper recognition of publications’ value, in a context that requires attention to information pluralism.

Three main axes:

Copyright and fair compensation: art. 43-bis L.d.A. (implementation of art. 15 Dir. 2019/790) recognizes publishers’ compensation for online use of publications; only “very brief extracts” that don’t dispense with integral reading are excluded. According to Agcom, the criterion is qualitative and case-by-case: evaluation focuses on how much the synthesis satisfies user intent compared to the original text.
Transparency and systemic risks (DSA): Very Large Online Search Engines are required to provide accessible information on recommendation systems’ functioning and periodic risk assessments, including disinformation, fundamental rights and media pluralism.
Competition and results presentation: analyzing the effect of a primary-position response on SERP neutrality and external sources exposure, a topic that dialogues with the Digital Markets Act and antitrust discipline, with interpretative questions still open on generative syntheses.

Precedents and institutional architecture

Italy has already adopted regulatory solutions to rebalance relationships between platforms and publishers. In the GEDI-Microsoft (Bing) and GEDI-Meta cases, Agcom determined compensation and indemnification for journalistic content use in intermediation contexts. In the new scenario, Agcom can receive complaints, acquire data and interface with the European Commission for DSA profiles; the Commission has competence over VLOP/VLOSE (Very Large Online Platforms/Very Large Online Search Engines), while ordinary courts remain the primary forum for copyright.

Elements to remember:

Deliberations 278/24/CONS and 180/25/CONS established parameters for compensation and indemnification in digital contexts.
Multi-level coordination: DSA for transparency and risk mitigation; DMA and antitrust for competition; L.d.A./Copyright Directive for related rights.
AI Overviews verification requires data and tests that isolate effects on traffic, visibility and pluralism, with attention to replicable measurements.

How to structure “citable” content in overviews

For publishers and brands, the pragmatic response consists of redesigning content to be useful for synthesis without losing the value of in-depth analysis. Clear structures, quality signals and verifiable references increase the probability of being referenced in the overview and obtaining qualified clicks toward complete content.

Operational guidelines:

Granular intents: segment pages into blocks that answer specific sub-questions (definitions, checklists, steps, examples).
Evidence and sources: include verifiable data and citations to facilitate correct attribution.
Frequently asked questions and markup: facilitate capture of typical “fan-out” sub-questions with FAQ sections and supporting semantic structure.
Quality signals: cultivate experience, expertise, authoritativeness and trustworthiness to strengthen selection criteria.
In-depth pages: design clear paths for those arriving from the overview, with advanced content and regulatory references.

Data security and right to be forgotten: why precautions are needed

The integration of generative models in search brings the theme of security and data governance back to the center, with particular reference to the right to be forgotten. The protection provided by art. 17 GDPR and European jurisprudence is based on link deindexing; a paraphrased synthesis could represent personal information not immediately traceable to a specific URL. It’s not proven that AI Overviews draws from already deindexed content; it remains essential to distinguish between data used for training (historical) and data retrieved in real-time (subject to policies and filters), adopting ex ante precautions.

Governance recommendations:

Privacy by design and by default (art. 25 GDPR): design preventive output controls to avoid irrelevant inclusions of personal data.
Reporting channels: establish procedures for intervention requests on syntheses that include unwanted personal information.
Internal input and output policies: avoid entering sensitive data in queries; periodic audits when overviews are incorporated into business touchpoints.
Flow distinction: map processes and legal bases when employing generative systems in processes involving personal data processing.

For Data Protection Officers and security managers, monitoring the emergence of new case studies is fundamental, also to promptly update information notices, registers and response plans.

Next steps: what to monitor in the EU

The European framework offers tools for balanced management: DSA for transparency and systemic risk mitigation, GDPR for data protection, Copyright Directive and L.d.A. for publications’ value, antitrust and DMA for competition. The challenge is to make these levels dialogue coherently, with measures proportionate to syntheses’ impact on information consumption and sources’ sustainability.

For organizations:

Measure impact: observe impression rates, overview citations, CTR and post-click conversions, comparing pre/post historical series.
Focus on quality: produce original, well-structured and updated content, with verifiable evidence and clear attribution.
Ensure compliance: integrate privacy by design principles, define reporting and review processes, and provide training for editorial and marketing teams.
Evaluate agreements/licenses: where relevant, explore agreements that clarify usage and valorization methods for content in AI products.

Toward more useful and reliable search

AI Overviews accelerates access to information by transforming the SERP into a reasoned “gateway,” capable of responding with greater immediacy and suggesting paths for deeper exploration. For this model to work to the advantage of users, publishers and businesses, transparency on synthesis criteria, source recognition and rigorous attention to personal data security throughout the entire cycle of response generation and presentation are needed. The institutional path initiated in Italy and Europe aims to ensure that search innovation is accompanied by rights protection, information quality and voice plurality: necessary conditions for lasting trust in the AI-enhanced search experience.

Sources

AI Overviews: quando l’innovazione rischia di diventare abuso (AI4Business) https://www.ai4business.it/intelligenza-artificiale/google-ai-overviews-quando-linnovazione-rischia-di-diventare-abuso/
AI Overview: come funziona la novità con cui Google sfida OpenAI (AI4Business) https://www.ai4business.it/intelligenza-artificiale/ai-overview-come-funziona-la-novita-con-cui-google-sfida-openai/

Want to bring transparency, citability and data security to your search processes? Contact us for consultation and discover how to apply them to your use cases with enterprise AI solutions.