sigmoid.social is one of the many independent Mastodon servers you can use to participate in the fediverse.
A social space for people researching, working with, or just interested in AI!

Server stats:

579
active users

#aioverviews

5 posts5 participants0 posts today

#PenskeMedia, owner of #RollingStone and #Billboard, sued #Google, alleging its #AIsummaries use their journalism #withoutconsent and reduce website #traffic. The lawsuit claims Google leverages its search dominance to impose terms on publishers, impacting advertising and subscription revenue. Google argues that #AIoverviews benefit users and drive traffic to a wider variety of websites. reuters.com/sustainability/boa #tech #media #news

🧠 In questo esempio, si vede in azione una nuova versione del workflow multi-agente dedicato all'ottimizzazione dei contenuti per #AIOverviews e i sistemi ibridi (SE + #LLM).

⚙️ Supporto a #GPT5, integrazione di semantic-ranker di Google per calcolare la rilevanza contestuale tra la query e le risposte dei risultati nella SERP di Google.  
👉 I dettagli: linkedin.com/posts/alessiopoma

___ 
✉️ 𝗦𝗲 𝘃𝘂𝗼𝗶 𝗿𝗶𝗺𝗮𝗻𝗲𝗿𝗲 𝗮𝗴𝗴𝗶𝗼𝗿𝗻𝗮𝘁𝗼/𝗮 𝘀𝘂 𝗾𝘂𝗲𝘀𝘁𝗲 𝘁𝗲𝗺𝗮𝘁𝗶𝗰𝗵𝗲, 𝗶𝘀𝗰𝗿𝗶𝘃𝗶𝘁𝗶 𝗮𝗹𝗹𝗮 𝗺𝗶𝗮 𝗻𝗲𝘄𝘀𝗹𝗲𝘁𝘁𝗲𝗿: bit.ly/newsletter-alessiopomaro 

www.linkedin.com#aioverviews #llm #gpt5 #langgraph #prompt #react #ai #genai #generativeai #intelligenzaartificiale #llm #seo | Alessio Pomaro🧠 In questo esempio, si vede in azione una nuova versione del workflow multi-agente che ho sviluppato, dedicato all'ottimizzazione dei contenuti per #AIOverviews e i sistemi ibridi (SE + #LLM). ⚙️ È stato aggiunto il supporto a #GPT5, e integrato un nuovo sistema per calcolare la rilevanza contestuale tra la query e le risposte dei risultati nella SERP di Google. 👉 La rilevanza viene misurata da un modello Reranker di Google (semantic-ranker, disponibile su Vertex AI), che un agente utilizza come tool via API. 🤖 L'interazione tra gli agenti, inoltre, è stata rafforzata migliorando i dati nello stato di #LangGraph e i #prompt dinamici che ne derivano. 🛠️ Il flusso di lavoro è basato su un'architettura LangGraph con diversi agenti che usano il modello selezionato nelle opzioni. Nel video, Gemini 2.5 Pro. ⚙️ Flusso di funzionamento.. 👉 In base alla query di ricerca indicata, il sistema analizza l'AI Overview estraendo i contenuti dalle fonti e anche dagli altri risultati nella SERP di Google. 👉 Misura la rilevanza contestuale delle risposte fornite delle fonti, e, in base ai dati, crea delle risposte più pertinenti in modo iterativo, considerano quella con indice di rilevanza maggiore. 👉 Infine, crea un piano d'azione per l'ottimizzazione della pagina di interesse, elaborando tutti i dati a disposizione. ✨ Nell'interazione tra gli agenti (che si vede nella sidebar di sinistra) applico una forma del paradigma #ReAct, in cui un agente produce un'analisi e un'altro (l'esecutore) applica le indicazioni dell'analisi per ottimizzare la risposta. 👉 Ne parlerò nella prossima edizione dell'Advanced SEO Tool, e nelle prossime Accademie Search On Media Group. ___ ✉️ 𝗦𝗲 𝘃𝘂𝗼𝗶 𝗿𝗶𝗺𝗮𝗻𝗲𝗿𝗲 𝗮𝗴𝗴𝗶𝗼𝗿𝗻𝗮𝘁𝗼/𝗮 𝘀𝘂 𝗾𝘂𝗲𝘀𝘁𝗲 𝘁𝗲𝗺𝗮𝘁𝗶𝗰𝗵𝗲, 𝗶𝘀𝗰𝗿𝗶𝘃𝗶𝘁𝗶 𝗮𝗹𝗹𝗮 𝗺𝗶𝗮 𝗻𝗲𝘄𝘀𝗹𝗲𝘁𝘁𝗲𝗿: https://lnkd.in/dN-bwBrV #AI #GenAI #GenerativeAI #IntelligenzaArtificiale #LLM #SEO

🧠 In ambito di Search (es. motori di ricerca o sistemi #RAG) si parla spesso di "pertinenza". 

🔍 Negli ultimi mesi ho studiato l'argomento, e condivido alcuni spunti che mi hanno affascinato: linkedin.com/posts/alessiopoma

___ 
✉️ 𝗦𝗲 𝘃𝘂𝗼𝗶 𝗿𝗶𝗺𝗮𝗻𝗲𝗿𝗲 𝗮𝗴𝗴𝗶𝗼𝗿𝗻𝗮𝘁𝗼/𝗮 𝘀𝘂 𝗾𝘂𝗲𝘀𝘁𝗲 𝘁𝗲𝗺𝗮𝘁𝗶𝗰𝗵𝗲, 𝗶𝘀𝗰𝗿𝗶𝘃𝗶𝘁𝗶 𝗮𝗹𝗹𝗮 𝗺𝗶𝗮 𝗻𝗲𝘄𝘀𝗹𝗲𝘁𝘁𝗲𝗿: bit.ly/newsletter-alessiopomaro 

"When Google unveiled AI Overviews last year, online publishers worried that the AI-generated blurbs in the top spot of search results would spark precipitous declines in traffic and gut the business model of vast reaches of the web.

Now, there is growing evidence validating those fears.

New research shows the web traffic that publishers have long relied on is significantly slowing, thanks to AI-generated summaries and the rise of AI chatbots.

Traffic to CNN's website has dropped about 30% from a year ago. Business Insider's and HuffPost's sites saw traffic plunges around 40% in the same period, according to figures from digital market data firm Similarweb.

While many factors influence traffic fluctuations, publishers say the introduction of Google's AI Overviews in May 2024 has packed a punch.

Helen Havlak, the publisher of The Verge, a tech news site and Vox Media's most visited homepage, said when people see AI summaries, they visit sites for information less often."

npr.org/2025/07/31/nx-s1-54841

"Sixteen months have passed since we showed you how Google’s algorithm was killing independent websites by favoring big media publishers that were abusing their reputation to sell you bad products.

Since then, Google enforced a new spam guideline called “site reputation abuse” by removing entire sections of major websites from their search index… at least for a few months.

While this drama was unfolding, Google rolled out AI Overviews under the slogan, “Let Google do the searching for you.”

But are AI Overviews leading you to the best results on the web? Or is this just a new prime spot for selling you products you will regret buying? And how does this change affect the websites that made Google’s AI possible in the first place?

To find out, we spent weeks conducting research into air purifiers we reviewed (including models that don’t exist) to determine whether Google Search results and AI Overviews are leading consumers to good advice.

It turns out this rabbit hole is deeper than we thought, with ramifications that span major Reddit communities and could be leading you to scams and defamatory hallucinations.

I hope you’re ready for a long read because this article is packed with examples, fresh data and way too many screenshots."

housefresh.com/beware-of-the-g

Beware of the Google AI salesman and its cronies
HouseFresh · Beware of the Google AI salesman and its croniesExposing the overly salesy AI Overviews that will push you to buy bad products and exploring the system making it possible.

🧠 Studio e ottimizzazione dei contenuti di #AIOverviews: come si può lavorarci? 
✨ Questo software è un piccolo esempio. Come funziona: linkedin.com/posts/alessiopoma

___ 
✉️ 𝗦𝗲 𝘃𝘂𝗼𝗶 𝗿𝗶𝗺𝗮𝗻𝗲𝗿𝗲 𝗮𝗴𝗴𝗶𝗼𝗿𝗻𝗮𝘁𝗼/𝗮 𝘀𝘂 𝗾𝘂𝗲𝘀𝘁𝗲 𝘁𝗲𝗺𝗮𝘁𝗶𝗰𝗵𝗲, 𝗶𝘀𝗰𝗿𝗶𝘃𝗶𝘁𝗶 𝗮𝗹𝗹𝗮 𝗺𝗶𝗮 𝗻𝗲𝘄𝘀𝗹𝗲𝘁𝘁𝗲𝗿: bit.ly/newsletter-alessiopomar 

"Search With Stateful Chat" patent (Cf. patents.google.com/patent/US20 ) - appears to describe the Gemini app for smartphones.

"Method for Text Ranking with Pairwise Ranking Prompting" (Cf. patents.google.com/patent/US20 ) - documents an experimental process described in this research paper titled "Large Language Models are Effective Text Rankers with Pairwise Ranking Prompting" (Cf. arxiv.org/pdf/2306.17563 ). There is no indication this was introduced into a live agentic system like Gemini.

"User Embedding Models for Personalization of Sequence Processing Models" (Cf. patents.google.com/patent/WO20 ) - documents an experimental process for improving recommender (sub-)systems (like movie searches) that incorporate large language models. The process is described in this research paper titled "User Embedding Model for Personalized Language Prompting" (Cf. arxiv.org/pdf/2401.04858 ).

"Systems and methods for prompt-based query generation for diverse retrieval" (Cf. patents.google.com/patent/WO20 ) - updates a 2022 patent for a process named PROMPTAGATOR that generates queries more efficiently based on a small number of examples, as described in this research paper titled "Promptagator - Few-shot Dense Retrieval from 8 Examples" (Cf. arxiv.org/pdf/2209.11755 ). This could be used to generate query fan-outs (but query fan-out has been used in multiple systems at least since the 1990s, so there are many implementations).

"Instruction Fine-Tuning Machine-Learned Models Using Intermediate Reasoning Steps" (Cf. patents.google.com/patent/US20 ) - documents an older method for fine-tuning instructions submitted to LLMs, as described in this 2022 research paper titled "Scaling Instruction-Finetuned Language Models" (Cf. jmlr.org/papers/volume25/23-08 ). The work has been superseded by this paper titled "Mixture-of-Experts Meets Instruction Tuning: A Winning Combination for Large Language Models" (Cf. arxiv.org/pdf/2305.14705 ).

This is the AI Overviews patent, titled "Generative summaries for search results" (Cf. patents.google.com/patent/US11 )

patents.google.comUS20240289407A1 - Search with stateful chat - Google Patents Implementations are described herein for augmenting a traditional search session with stateful chat—via what will be referred to as a “generative companion”—to facilitate more interactive searching. In various implementations, a query may be received, e.g., from a client device operated by a user. Contextual information associated with the user or the client device may be retrieved. Generative model (GM) output may be generated based on processing, using a generative model, data indicative of the query and the contextual information. Synthetic queries may be generated using the GM output, and search result documents (SRDs) may be selected. State data indicative of: the query, contextual information, one or more of the synthetic queries, and the set of search result documents, may be processed to identify a classification of the query. Based on the classification downstream GM(s) may be selected and used to generate one or more additional GM outputs.

🧠 Mentre fino alla settimana scorsa #AI Mode di Google presentava le fonti con link senza parametri e con attributo "noreferrer", oggi la situazione è già cambiata.
👉 Iniziano a comparire i text fragment ("#:~:text=" - come su #AIOverviews) e i link non sono più "noreferrer".
✨ Vedremo quale sarà la versione definitiva. 

___ 
✉️ 𝗦𝗲 𝘃𝘂𝗼𝗶 𝗿𝗶𝗺𝗮𝗻𝗲𝗿𝗲 𝗮𝗴𝗴𝗶𝗼𝗿𝗻𝗮𝘁𝗼/𝗮 𝘀𝘂 𝗾𝘂𝗲𝘀𝘁𝗲 𝘁𝗲𝗺𝗮𝘁𝗶𝗰𝗵𝗲, 𝗶𝘀𝗰𝗿𝗶𝘃𝗶𝘁𝗶 𝗮𝗹𝗹𝗮 𝗺𝗶𝗮 𝗻𝗲𝘄𝘀𝗹𝗲𝘁𝘁𝗲𝗿: bit.ly/newsletter-alessiopomar