A client opens Analytics on a Monday morning and finds a 23% drop in organic traffic compared to the previous month. Rankings stable, content intact, no penalties in Search Console. Users arrive, read Google's answer, and no longer click.
Google AI Mode arrived in Italy in March 2025 and in twelve months has reshaped the SERP more deeply than any algorithmic update of recent years.
What is Google AI mode
The direct evolution of AI Overviews. While those were blocks of synthetic responses appearing above organic results for informational queries, AI Mode goes much further: it transforms the entire search interface into a conversation. The user asks a question, Google answers using Gemini, and links to sites become optionally cited sources in the margin.
For queries where AI Mode activates, the classic ten-result layout practically disappears from the initial viewport. The click becomes an exception. That said, AI Mode doesn't affect everything equally. Transactional queries still hold up well: "buy Nike shoes size 42" or "book restaurant Florence centre" still generate clicks because they require an action Google can't complete on the user's behalf. What collapses are generic informational queries, the ones on which much of the blog and content site traffic was built.
Who is losing traffic and by how much
Data from the first months post-launch in Italy shows highly differentiated drops depending on content type.
| Content type | Estimated CTR impact |
|---|---|
| Generic informational guides | -35% / -60% |
| FAQs and frequently asked questions | -40% / -70% |
| "What is" and "how does it work" articles | -30% / -55% |
| Product reviews and comparisons | -10% / -25% |
| Local content with specific intent | -5% / -15% |
| Transactional pages (e-commerce, contacts) | Stable or slight decline |
Agencies and professionals who bet on generic informational content are seeing the steepest drops. Those with a strong local positioning or niche technical content are holding up much better.
There's also an effect that few are monitoring: sites cited in AI Overviews as a trusted source often see an increase in branded searches in the following weeks. Users read Google's response, notice the source, and then search for it directly. High-quality traffic, even if in lower volume.
How Google chooses who to cite
Google doesn't cite sources randomly. From analyzing the pages that appear as references in AI Overviews, a fairly consistent set of signals emerges.
The most relevant concerns page structure. Gemini doesn't read between the lines — it follows declared structures. Pages with correctly implemented JSON-LD schema (Article, FAQPage, HowTo) are extracted with significantly higher frequency than pages without markup. We covered this in detail in the article on schema markup for AI Overviews.
Topical authority also matters. A generalist blog publishing on everything tends not to be cited. A vertically focused site on a specific topic, even with less overall traffic, is preferred. E-E-A-T signals (Experience, Expertise, Authoritativeness, Trustworthiness) have become much more important than they were in traditional SEO.
Freshness matters, but only if declared explicitly. The dateModified field in JSON-LD for anything related to tools, prices, regulations and statistics makes a difference. Pages with a recent update date climb in citations for queries where recency is relevant.
Finally, there's text structure. Pages that appear most often in citations are those that contain a clear answer to the main question within the first three or four paragraphs. The in-depth content serves those who click and want to know more, but Gemini doesn't wait for paragraph six.
What to do now
Before touching anything, you need to understand which pages are actually losing traffic and why. Search Console shows the impression vs. click trend: if impressions stay stable but clicks drop, AI Mode is intercepting the query before the user reaches the organic results. If both drop, the problem is something else.
A quick tool for an initial technical overview is PerSeo Insights: analysis in a few minutes on structured data structure, indexability issues and critical signals. From there, you can see where it's worth intervening first.
On the content front, informational articles that have lost traffic need to be restructured. The concrete question to ask is: does a reader who arrives on the page find the answer to their question within the first or second paragraph, or do they have to scroll through several blocks? If they have to scroll, the page is optimized for a human reader with time to spare, not for an automated extraction system.
For structured data, every blog article should have at least BlogPosting with headline, datePublished, dateModified, author with a linked profile and publisher with logo. Pages with questions and answers gain a lot from adding FAQPage. For practical tutorials, HowTo.
The other direction to explore is content that AI Mode can't replicate well: case studies with original data, comparisons based on direct testing, hyper-specific local content. "Best accountants in north Florence 2026" is a query Gemini handles poorly because it requires up-to-date data and on-the-ground verification. An internal analysis based on data collected from real clients produces something that doesn't exist anywhere else.
How things really stand
Generic informational traffic is dropping for almost everyone. Sites with a strong local positioning, those with deep vertical content, and those with original data are holding up — in some cases growing. Those losing the most are sites built entirely on content that answers generic questions: "what is SEO", "how to make a quote", "difference between X and Y". That type of content has become the ideal input for AI models, and Google aggregates it and serves it directly.
Those who have invested in specificity, topical authority and correct technical structure have an advantage that will be difficult to erode in the short term.