#ai-overview — Public Fediverse posts
Live and recent posts from across the Fediverse tagged #ai-overview, aggregated by home.social.
-
Google's AI-powered search summaries could be generating vast numbers of inaccuracies every hour, despite appearing largely reliable at first glance.
-
Google "AI overview" is their old #Ai model, it's a hybrid system RAGing off their search engine with an ancient AI that is likely preceding the already sub standard Gemini 1.5-7
In fact when #google first deployed it Gemini was but a twinkle in the devs eye, so it's some dead ended Bard version.
Why do they use it?
1) Because it uses little compute comparing to the later reasoner models. Scale it across 16billion searches a day, it's $1,600,000 / day on the cheapest query AI cost or a more realistic $160million per day. So it makes perfect sense financially to tune the engine so it's right 95% of the time so you can save $100mil a day.2) Frankly, because vast majority of queries can be satisfactorily served.
3) The quality of #googlesearch has been documented to decline so you spend more time on their page generating ad revenue.
Edit; numbers
-
#Google #AI Overviews cite #YouTube more than any #medical site for #health queries, study suggests | The Guardian
Researchers at SE Ranking, a search engine optimisation platform, found YouTube made up 4.43% of all #AIOverview citations. No #hospital network, government health portal, medical association or academic institution came close to that number, they said.
“YouTube is not a medical publisher,” the researchers wrote. “Anyone can upload content there ”
-
Ja, scho recht...
-
Weekly output: FCC frees Verizon from phone-unlocking rule, Donut Lab’s solid-state battery, Wikipedia turns 25, Google traffic trends
This week brought a lot of unpleasant headlines, but none made me feel uneasy like the news that Washington Post reporter Hannah Natanson was subjected to an unexpected visit by the FBI at her house Wednesday that concluded with investigators confiscating her work and personal laptops, her phone and her smartwatch–all seized, it now seems, on a fraudulent pretext. Since then, the man who spent his equivalent of pocket change to buy the Post in 2013 has said nothing about that in public, yet another way in which Jeff Bezos has shown himself an unworthy successor to Katharine Graham and her son Don Graham.
In addition to what you see below, I wrote a post for Patreon readers sharing bonus bits about what I learned at CES two weeks ago.
1/12/2026, FCC Unties Verizon From 60-Day Phone-Unlocking Rule, PCMag
Two days after the Federal Communications Commission granted Verizon’s wish, Verizon suffered an hours-long outage that people could work around relatively easily if their phones were unlocked to allow them to add a third-party eSIM.
1/15/2026: Wikipedia Is Now 25 Years Old [Citation Not Needed], PCMag
I had made a mental note to myself to write an essay about the free online encyclopedia hitting the quarter-century mark, but then the actual date snuck up on me. Fortunately, the lede and the headline basically wrote themselves.
1/17/2026: Donut Labs’ Solid-State Battery-Powered Motorcycle Turned Heads at CES, But Big Questions Remain, PCMag
Writing this left me feeling over my skis more than usual–I don’t cover battery technology in any great depth, this startup has disclosed almost nothing about the design of its solid-state batteries, and two outside analysts weren’t willing to assess their pitch. But reading up at length on this company and on others working on better EV batteries did make me realize one thing: There’s enough innovation happening here that we don’t need Donut’s invention to work.
1/17/2026: AI Is Still Hammering News Sites, Google Search and Social Referrals Plunge, PCMag
The brutal declines in Google search traffic reported in a study published Monday by the University of Oxford’s Reuters Institute for the Study of Journalism caught my eyes and those of my editors, but I also found some news value in this report’s stats about which online platforms and technologies newsroom leaders plan to emphasize or back away from.
#AI #AIOverview #BrendanCarr #ces #DonutLab #EVBattery #FCC #phoneUnlocking #SIMLock #smartphoneUnlock #solidStateBattery #verizon #Wikipedia
-
Google pushes Gmail into its Gemini era, but is this really what users want?
https://web.brid.gy/r/https://nerds.xyz/2026/01/google-gmail-gemini-ai-push/
-
Canadian fiddler Ashley MacIsaac had a concert cancelled after Google's AI labeled him a sex offender.
The Juno Award-winning musician learned about it when event organizers confronted him with the AI summary and pulled the gig.
-
Google AI oskarżyło znanego muzyka o najgorsze. Koncert odwołano, bo algorytm „pomylił osoby”
To miał być standardowy koncert Ashley’a MacIsaaca, legendy kanadyjskiej sceny muzycznej. Zamiast tego artysta otrzymał wilczy bilet. Powód? Organizatorzy wpisali jego nazwisko w Google, a sztuczna inteligencja w sekcji AI Overview poinformowała ich, że mają do czynienia z przestępcą seksualnym.
Pamiętacie, jak Google AI kazało ludziom smarować pizzę klejem, albo jeść kamienie? Śmialiśmy się. Teraz jednak żarty się skończyły. Błąd algorytmu (tzw. halucynacja) realnie wpłynął na czyjeś życie i karierę.
„Cyfrowy wyrok” bez sądu
Sprawę nagłośnił The Globe and Mail. Organizatorzy koncertu w społeczności Sipekne’katik First Nation w Nowej Szkocji odwołali występ MacIsaaca w ostatniej chwili. Powołali się na informacje z Google, według których muzyk miał na koncie wyroki za napaści seksualne i „internetowe nagabywanie” (internet luring).
Problem w tym, że to nie była prawda. Algorytm Google AI Overview „skleił” biografię słynnego skrzypka z życiorysem zupełnie innej osoby o tym samym nazwisku, która faktycznie miała kartotekę. Dla AI to tylko zbieżność tokenów. Dla człowieka – utrata reputacji i zarobku.
Z nawigatora w narratora
Eksperci biją na alarm. Clifton van der Linden z Uniwersytetu McMaster celnie zauważa: „Obserwujemy przejście wyszukiwarek z roli nawigatorów informacji w rolę narratorów”. Kiedyś Google odsyłało nas do źródeł (gdzie mogliśmy zweryfikować autora). Dziś AI Overviews podaje gotową „papkę”, którą większość użytkowników traktuje jak prawdę objawioną.
Google: „Uczymy się”
Rzecznik Google, Wendy Manton, w standardowym oświadczeniu stwierdziła, że „wyszukiwarka jest dynamiczna”, a firma „wykorzysta ten przykład do ulepszenia systemów”. Wyniki dla MacIsaaca zostały ręcznie poprawione. Organizatorzy koncertu przeprosili artystę, tłumacząc się błędem, i zaprosili go ponownie. Ale niesmak i strach pozostały. MacIsaac słusznie pyta: ilu organizatorów wcześniej widziało te bzdury i po prostu cicho rezygnowało z jego usług, nie mówiąc mu dlaczego?
To ostrzeżenie dla nas wszystkich. W 2026 roku, zanim uwierzysz w to, co „mówi Google” na górze strony – przewiń niżej i sprawdź prawdziwe źródła. Twoje lenistwo może kogoś skrzywdzić. Zwłaszcza, gdy na podstawie tego co wygeneruje AI podejmujesz decyzje, które mogą mieć wpływ na czyjeś życie.
#AIOverview #AshleyMacIsaac #fakeNews #Google #GoogleSearch #halucynacjeAI #sztucznaInteligencja #technologiaAI pomyliło klarnet z karabinem i wywołało lockdown szkoły. Twórcy systemu: „To nie był błąd”
-
For years I've been writing that AI is essentially another stage of control over semantics. The power of philosopher kings (Plato) who create new myths.
A process that has accelerated... building internet gateways that filter meaning.
New piece on the expert's paradox and the facade of trustworthiness.
https://dadalo.pl/en/tech/ai-slop-expert-paradox-cognitive-biases/
#AI #epistemology #Google #bias #AIOverview #disinformation #criticalthinking #tech #SearchEngines
-
Weekly output: Mozilla Firefox CEO, AI crawlers vs. publishers and creators, teenage AI chatbot use, Android Live Emergency Video, PCMag’s best tech bought in 2025, World App
Somehow I’m down to the last full workweek of the year–and yet my writing and gift shopping seem to have more than a week’s worth of work remaining.
12/8/2025: Mozilla is doing a delicate dance with AI, Fast Company
I spoke with Mozilla CEO Laura Chambers at a Web Summit event for the second time this year. One thing Firefox’s management no longer needs to worry about, unlike when I met with Chambers at Web Summit Qatar in February: the threat of Google being forced to stop paying browser developers to keep its search engine as the default.
12/9/2025: AI Platforms Are Paying (Some) Big Publishers, Leaving Smaller Ones Behind, PCMag
This post began with me taking notes from a Web Summit panel featuring Cloudflare CEO Matthew Prince talking about that Internet infrastructure company’s Pay Per Crawl initiative to push AI providers to pay Web publishers for access to their content, then I did some follow-up reporting that included setting up Cloudflare’s AI Crawl Control bot-blocking filter on this blog, and then I had to update the post the morning it was published after the European Commission opened an investigation into how Google runs its AI Overview search feature.
12/9/2025: 28% of Teens Use Chatbots Daily. You Can Probably Guess Which One They Like Best, PCMag
The latest survey by the Pew Research Center surfaced some interesting statistics about how much teenagers use AI chatbots and which ones they use the most.
12/10/2025: Need Help? Android Phones Can Now Share Live Video With 911 Dispatchers, PCMag
Google is shipping this feature a year after Apple did, but its emergency live video implementation works on far more devices than Apple’s.
12/11/2025: The Best Tech PCMag Editors Bought in 2025, PCMag
I wrote a short graf lauding the compact, quick-charging (and Wirecutter-endorsed) USB-C charger that I bought after losing the considerably bulkier model that came with my laptop.
12/13/2025: App That Verifies Your Existence Adds Encrypted Messaging, PCMag
Tools for Humanity announced an update to its World App that adds an end-to-end-encrypted chat feature and expands its cryptocurrency tools. I took advantage of this news peg to try out the app’s ability to verify a “World ID” by scanning the NFC tag on my U.S. passport; that did not go well at all for me.
12/15/2025: Updated to add the PCMag best-tech package that I forgot to check for on Sunday.
#AIChatbot #AIOverview #AISearch #ChatGPT #Firefox #GoogleZero #Mozilla #PayPerCrawl #PewResearchCenter #ToolsForHumanity #WebSummit #WorldApp #WorldID
-
Europe’s Antitrust Investigation Against Google Moves Ahead
AI Overviews has been zapping people's traffic. Now, Europe is initiating an antitrust investigation against Google in response.
https://www.freezenet.ca/europes-antitrust-investigation-against-google-moves-ahead/
#LawAndTechnology #News #Technology #AI #AIOverview #antitrust #Europe #Google
-
Great data & insights from @ahrefs 🔍
📊 New research shows that Google AI Overviews change every 2 days — yet their underlying meaning stays almost identical.
AI search is both highly dynamic (70% content change, 45% citation turnover) and semantically stable (0.95/1.0 similarity).
Full article → https://ahrefs.com/blog/ai-overview-change/
#SEO #AIOverview #Search #AI #GoogleAI #SGE #ContentStrategy #SearchMarketing #DigitalStrategy #OpenWeb #TechNews #MastodonTech -
“William Blake’s poem ‘Eternity’ was written in 1880, which was after his death in 1827.”
Great work, Google AI. The Zero-Click Impression is working out swimmingly. No notes.
#google #ai #artificialintelligence #googlegemini #chatgpt #williamblake #poetry #romanticism #literature #aioverview #poem #zeroclickimpression #search #snafu
-
More Businesses Step Forward to Say AI Overview is Killing Their Traffic
The consensus that AI Overview is gutting web traffic to independent website is growing. Australian businesses share their experiences.
https://www.freezenet.ca/more-businesses-step-forward-to-say-ai-overview-is-killing-their-traffic/
#Business #News #AIOverview #consulting #Google #journalism #SEO #traffic
-
Gli editori italiani contro #Google: “Il servizio Ai Overview riduce la visibilità dei giornali online”.
#Tecnologia #AiOverview #Ai #IntelligenzaArtificiale #Editoria #Giornalismo
-
it's hard not to think the "AI overviews" from #google #gemini are getting worse, not better...
(here it is telling me Bybit was hacked for $1.5 billion in August 2025 instead of February... this is only a few months after Gemini patiently explained to me that the US maintains an army base on the moon)
cc @davidgerard
-
More Research Concludes AI Overview is Killing Clicks
Google has long argued that AI Overview is not killing traffic to publishers. Additional research is showing otherwise.
https://www.freezenet.ca/more-research-concludes-ai-overview-is-killing-clicks/
#Business #News #AI #AIOverview #Google #journalism #SEO #traffic #website
-
Google Sued By Rolling Stone Over AI Overviews
A lawsuit against Google is claiming unjust enrichment and monopolistic abuse over AI Overview.
https://www.freezenet.ca/google-sued-by-rolling-stone-over-ai-overviews/
#Business #LawAndTechnology #News #AI #AIMode #AIOverview #Google #lawsuit #RollingStone #traffic
-
Google Suggests It’ll Be Making it “Easier” to Use AI Mode
A suggestion that AI mode would soon be the default from someone from Google was "clarified" to say that they're making it easier to use.
https://www.freezenet.ca/google-suggests-itll-be-making-it-easier-to-use-ai-mode/
-
An example of enshittification and how badly Google has broken its core reason for existence, search: Google is using an LLM to "hallucinate" postcodes.
I dunno about other countries, but in Australia, if you get someone's postcode correct and a detail as simple as the recipient's name or nickname, a street name without the number, or the name of a home, then our postal service has a miraculously high success-rate for delivering a letter or parcel to the correct destination. You can write the wrong suburb or state in the address, but if you use the right the postcode, Australia Post delivers. But, without the correct postcode, you're up the creek without a paddle. The wrong postcode can lead to delayed delivery or can result in your mail getting the "Return to sender" treatment.
Google Search results _could_ index the database from Australia Post or pay a moderate commercial licensing fee to access their API. But, instead, Google's got an LLM to make-up postcodes. The results are near enough to appear convincing at first glance, such as using '3' for first digit of postcodes in Victoria. Then, they present it in the top position of Search results and have the cheek to use the Australia Post logo. In my experience, there appears to be a correlation between numerical similarity of Google's fictional postcodes and geographical proximity of suburbs. Google's "AI Overview" presents misleading nonsense instead of using traditional indexing or a trivial lookup table.
(More details in the Alt-text of the screenshot.)
#Google #GoogleSearch #AI #AIOverview #LLMs #Enshittification #AustraliaPost #Australia #Postcode #Zipcode
-
Google Admits that Yes, The Open Web is In Rapid Decline
Google has long denied that its AI is killing the open web, but recently, they admitted that the open web is dying.
https://www.freezenet.ca/google-admits-that-yes-the-open-web-is-in-rapid-decline/
#LawAndTechnology #News #Technology #AI #AIMode #AIOverview #court #Google #traffic #web
-
Google dichiara che l’open web sta scomparendo. Tra AI Overview e pubblicità display in calo, il panorama digitale cambia radicalmente. Cosa significa per chi crede nella libertà online e nell’ecosistema open source? #OpenWeb #Google #Linux #AIOverview
https://www.linuxeasy.org/open-web-declino-google/?utm_source=mastodon&utm_medium=jetpack_social
-
Google AI Overviews: Wenn sich die KI auf KI als Quelle beruft https://www.computerbase.de/news/apps/google-ai-overviews-wenn-sich-die-ki-auf-ki-als-quelle-beruft.94239/ #Google #AIOverview
-
Google Moves Forward With AI Only Search Results
The next shoe is beginning to drop for independent websites as Google moved forward with AI only search results.
https://www.freezenet.ca/google-moves-forward-with-ai-only-search-results/