I decided to open this newsletter with an analysis that, in some ways, represents a double starting point.
It is one for me, because it inaugurates a space in which I will try to share observations, scenarios, and signals of change in the world of digital publishing — looking Inside the Algorithm. But it is also, and above all, a starting point for the industry itself: we are entering a new phase, one that is more selective, more technical, more algorithmic.
The article that follows attempts to describe this transition, beginning with a key patent published by Google in 2023, whose effects are now manifesting in tangible ways. From the era of MFA (Made for AdSense) sites to the advent of artificial intelligence, digital publishing has undergone radical change.
A vision that Francesco Apicella — among the leading experts in the media world and digital strategies, founder of Evolution Group, a tech company that supports hundreds of publishers in adapting to new scenarios, and my mentor (and boss) — had already mapped out with clarity years ago. As soon as generative AI began to seriously impact the online content ecosystem, Apicella started raising awareness within the sector about the urgency of evolving from “site” to “brand”: a transition that appears simple but is in fact crucial, because it represented the only way to be recognized by Google as an authoritative subject amid the speculative noise of MFA sites. That vision anticipated what has today become obvious. Those who started down that path early now enjoy a real competitive advantage. Those who stayed put, by contrast, find themselves forced to make an even larger leap — because today, to stay in the game, being a brand is no longer enough. You need to become a source.
From Traffic to Brand: the First Great Evolution
The recent history of the web has been marked by the proliferation of MFA (Made for AdSense) sites, designed to maximize advertising revenue with no real attention to informational quality. Google tried to counter this dynamic through various core updates aimed at rewarding what it called “quality.”
However, the mass introduction of generative AI further complicated the landscape. Core updates such as those of November 2023 and March 2024 attempted to stem the wave of automatically generated, low-quality content, but with mixed results. In many cases, the updates also hit established publishers hard, penalizing sites that (and here we open Pandora’s box — or rather we will open it) apparently complied with the guidelines (see E-E-A-T, for instance) and had built an editorial reputation over time. The paradox was plain: while MFA sites proliferated, some of the best editorial brands saw their advertising revenues collapse.
In recent years, Google also tried — not without difficulty — to build perimeters around quality, in an attempt to distinguish authentic journalism from the background noise of speculative content. A significant example was Showcase, a platform launched in 2020 to give publishers a more curated editorial space where they can exercise direct control over how content is presented and distributed. It is not just a showcase, but a genuine global licensing program: Google compensates participating publishers, incentivizing the production of quality journalism and fostering stronger relationships with audiences. Inside Showcase, editorial choice becomes central again: selected articles, timelines, in-depth features, and even free access to some paywalled content. It is a logic radically different from Discover or the traditional SERP: here the publisher decides what to show, not the algorithm. In an age where AI synthesizes everything, platforms like Showcase are a reminder that quality needs context, structure, and recognizability. But also that Google, when it matters, knows how to identify quality, value it, and pay for it.
The reaction to these scenarios pushed many publishers toward a clear strategic direction: becoming recognizable brands. Being a brand means having a consistent editorial line, an identifiable tone, a direct relationship with one’s community. Evolving into a brand also means thinking and designing content according to channels and user behaviors: social media presence, vertical newsletters, push notifications designed to drive return visits, content crafted to increase average session time, alongside PR activities, event participation, and marketing. Every touchpoint becomes an opportunity to build habit and community.
Google offers valuable tools such as News Consumer Insights (NCI), which allows publishers to analyze reader behavior. The data speaks clearly: compared with those who do not use this information, publishers who know their readers in depth record +20% in direct revenue, +20% in ad value, and +40% in CTR on published content.
Editorial branding, therefore, is not a stylistic exercise. It is an operational strategy, supported by people, data, technology, and attention to reader behavior. And it becomes, today more than ever, the baseline from which to evolve into a source. Because without a relationship, no content is truly relevant — neither for the user nor for the algorithm.
From Brand to Authoritative Source: the New Evolutionary Threshold
But the new horizon taking shape is becoming a “source” in the most technical and systemic sense of the term: a data and content origin that a search engine — or more precisely, an LLM — considers reliable, structured, accessible, and processable.
Patent US11769017B1, published September 26, 2023 and titled “Generative Summaries for Search Results,” outlines with great clarity the new cognitive architecture of the SERP. The system no longer merely returns a list of links; it uses a Large Language Model (LLM) to generate a natural-language (NL) summary based on selected documents (SRD — Search Result Documents) as a response to the query.
The Technical Phases of the Process Described in the Patent
- Query reception: the user’s request is acquired, either as direct input or as an automatically generated prompt (e.g., context or history).
- SRD identification: the system selects a set of documents that form the corpus from which information is extracted.
- LLM processing: text snippets or fragments are processed by an LLM, which generates a preliminary linguistic output.
- NL synthesis: from this output, a coherent natural-language summary is generated, summarizing the information at a balanced level of detail (avoiding both over-generalization and hyper-specificity).
Dynamic Selection of the Generative Model
One of the most sophisticated elements described in the patent is the dynamic model selection. The system can choose among different LLMs (or versions of them) based on:
- query characteristics
- type and quality of content in the SRDs
- user context
This ensures an optimal combination of semantic accuracy, relevance, and computational scalability.
Verifiability and Traceability: the New Standard
The patent also introduces the ability to associate sections of the NL summary with specific SRD documents via direct links. This not only allows for point-by-point verification of information, but also increases trust toward the ecosystem of selected sources. It is an evolved form of automated citation system, where content must be structured to facilitate that mapping.
Confidence Levels
The system can also assign different confidence levels to text portions of the generated summary. This could manifest visually through the use of distinct colors or other UI/UX signals, making content transparency available to the end user in terms of reliability.
Implications for Digital Publishing
Being a source in the context of patent US11769017B1 does not simply mean being authoritative in the conventional sense; it also means:
- having content written in a formally processable way
- using structured signals: schema.org, structured data, microdata
- providing granular, segmentable, and verifiable information
- constantly updating static content
Those who do not structure their content to be read, interpreted, and synthesized by an LLM risk disappearing from the new generative SERP.
Is the era of traffic over? Is the era of the brand at its peak? Perhaps — but the near future belongs to sources. And in this future, those who win will be those who not only know how to create valuable content, but can write it to be read by an algorithm, synthesized by an LLM, and recognized as reliable by an automated artificial intelligence system.
There are many strategies that we have long recommended to newsrooms, starting from reading the signals Google itself already offers in the SERP, or from the targeted use of advanced analytics platforms. The methodology we share with the publishers who ask for our support is based on an integrated approach: reading signals, building awareness, transforming the editorial mindset. It is not just about applying techniques, but about spontaneously — and structurally — evolving the way content is conceived and produced. All of this is supported by technologies and technical frameworks that, today, can no longer be optional. An excellent piece of content published on a technically inadequate platform will always struggle to surface compared to mediocre content hosted on an infrastructure built for digital publishing. Quality, by now, also runs through code — that is no longer a secret.
I have also been changing my own approach to online search for some time. The queries I run on Google have dropped drastically. Not out of disaffection, but because I know that I would often be confronted with an expanse of junk content, or at best would have to spend too much time manually sorting through reliable sources among fake news and clickbait attempts. That is why I prefer platforms that do a first filter for me: ChatGPT Search, for example, where AI composes a synthetic, pre-filtered response — though it remains essential to understand the criteria by which it selects sources, because errors are always lurking. Or YouTube, which for tutorials is unbeatable: I can see who is speaking, assess the quality of the content, and quickly decide who to trust.
Platforms that simplify the choice, that reduce noise and increase perceived quality, are winning today. Google, by contrast, continues to offer a vast undifferentiated sea in which the user must swim in search of a pearl. People are starting to tire of it.
For the first time since 2015, indeed, Google has fallen below a 90% share of the online search market, settling at 89.7% according to Statcounter (December 2024). It is a historic data point, marking the beginning of a new era: users are starting to search elsewhere. Bing (4%), Yandex (2.6%), and Yahoo (1.3%) are also recording small gains, capitalizing on the ongoing fragmentation. Meanwhile, many people already use AI to plan trips, shop, or look up practical information — activities once dominated by Google. We are still talking about crumbs, of course.
But perhaps this collective disenchantment is precisely what is pushing Google to change course more decisively. Perhaps it is the beginning of a new cycle, where quality — real, measurable, and structured quality — will return to center stage.
History teaches us that nothing lasts forever, and the question that remains open is: is Google still in time to restore order? Or has it lost control of its own ecosystem?
Frequently asked questions
