惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Recent Announcements
Recent Announcements
爱范儿
爱范儿
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
宝玉的分享
宝玉的分享
T
Tailwind CSS Blog
博客园_首页
IT之家
IT之家
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
博客园 - 三生石上(FineUI控件)
有赞技术团队
有赞技术团队
大猫的无限游戏
大猫的无限游戏
雷峰网
雷峰网
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
博客园 - 司徒正美
WordPress大学
WordPress大学
Last Week in AI
Last Week in AI
人人都是产品经理
人人都是产品经理
Jina AI
Jina AI
月光博客
月光博客
小众软件
小众软件
S
SegmentFault 最新的问题
量子位
阮一峰的网络日志
阮一峰的网络日志
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知

MEDIANAMA

India in talks with US, Anthropic for Mythos access; no Indian firms in Project Glasswing yet Eternal Q4FY26: All Users Pay Higher Platform Fee, Only Some Get Discounts Amazon, Meta to challenge PhonePe-Google Pay dominance as UPI cap delayed since 2020 Meta failed to protect the safety of under-13s: European Commission If markets and regulators are ready for network slicing, we are ready: JIO Why defining ‘news’ won’t fix the free speech problems of draft IT Rules? #NAMA Eternal Q4FY26: Goyal Dismisses AI Disruption Risk as Zomato Quietly Builds Agentic Commerce Infrastructure Karnataka files appeal challenging the bike taxi ban lift in the Supreme Court How did WhatsApp turn 17 govt. flags into 9,400 digital arrest scam bans? Google Wallet integrates Aadhaar as digital ID, expands India’s mobile identity ecosystem Kerala HC issues notice on MediaOne’s Facebook page block in India MeitY warns VPN providers against enabling access to blocked betting platforms Shreya Singhal targeted private censorship. Today’s threat is the State #NAMA Amazon scales its quick delivery service ‘Amazon Now’ in 100 cities Can MeitY issue binding rules via advisories? Experts raise alarm over draft IT Rules #NAMA How 2019 election code of ethics became India’s three-hour content takedown mandate #NAMA Australia proposes new levy on big tech to fund news, opens draft law for consultation ‘judge, jury, executioner’: experts warn of Inter-Departmental Committee (IDC) overreach under New draft IT Rules Lowdown: TRAI flags low deployment under PM-WANI in public Wi-Fi consultation paper Why the NBFC licence matters for MobiKwik China blocks Meta-Manus deal, asserts origin-country jurisdiction: what this means for India ‘No transparency’: experts warn of expanding powers to block online speech in India #NAMA X launches standalone iOS messaging app XChat with encryption in India How India’s content takedown framework was built and where It has gone wrong #NAMA Claude Mythos puts India on alert: CERT-In, telcos, banks assess unprecedented cyber risks Explained: why did the RBI cancel Paytm’s banking licence? Meta now instantly blocks content in India Govt. asks ZEE5 to halt ‘Lawrence of Punjab’ web series release Online Gaming Rules notified, to be in effect from May 1, what are the major changes? RBI mandates additional factor authentication for e-mandates
Google adds forums, direct links, subscription news to AI...
Rohit Singh · 2026-05-07 · via MEDIANAMA
  • Access the original blog post from Google here

Google is expanding its AI-powered Search features by adding more direct links, user discussion snippets, and subscription-based news content into AI-generated answers, even as concerns over the accuracy of those responses continue to grow.

In a blog post published on May 6, Google said its AI Mode and AI Overviews will now show quotes and links from online discussions, social media posts, blogs, forums, and other “firsthand sources”. The company is also introducing labels for articles from users’ paid news subscriptions and adding more inline links inside AI responses to help users “explore the web” further.

Google said the changes are aimed at helping users find “authentic voices” and “original content” more easily. The company also plans to show previews of websites when users hover over links in AI responses on desktop.

Google Pushes AI Search Further: The update comes as Google faces continued criticism over the reliability of AI Overviews, which were rolled out widely in 2024 and now appear at the top of many Search results.

A recent The New York Times investigation found that more than 4,300 Google search queries tested by AI startup Oumi had AI-generated answers that were accurate about 91% of the time. But the report said that error rate could still translate into “hundreds of thousands of inaccuracies every minute” because Google processes more than 5 trillion searches annually.

The report also found that Google’s AI Overviews often cited sources that did not fully support the claims made in the answers. According to the analysis, Google frequently pulled information from Facebook and Reddit, especially in responses that contained inaccuracies.

Separate studies have also raised concerns about the impact of AI Overviews on publishers and websites. A February 2026 study by marketing platform Ahrefs found that AI Overviews reduced click-through rates for top-ranking search results by 58%, as users increasingly got answers directly on Google instead of visiting external websites.

Earlier Controversies Around AI Overviews: Google’s AI search tools have repeatedly faced criticism for generating misleading or false answers. Soon after AI Overviews launched in May 2024, users shared examples of inaccurate responses, including suggestions to eat “at least one rock per day” and advice to use glue on pizza to stop cheese from sliding off. In another case, the system incorrectly claimed former US President Barack Obama was Muslim.

At the time, Google said many viral screenshots were fake or edited, while some genuine failures happened because the system was responding to “uncommon queries” with limited reliable information online. The company described these situations as “information gaps”, where the AI could pull answers from satirical posts, jokes, or unreliable user-generated content.

Google later said it had updated its systems to reduce the appearance of satire, humour, and misleading forum content in AI answers.

Errors, Hallucinations, and Source Concerns: Despite those changes, concerns around hallucinations and unreliable sourcing remain. The New York Times report documented several cases where AI Overviews gave incorrect or contradictory information despite linking to seemingly reliable sources. In one example, Google incorrectly stated there was “no single entity known as the ‘Classical Music Hall of Fame’” while linking to a website that listed Yo-Yo Ma as an inductee.

Google’s AI search tools have repeatedly faced criticism for generating misleading or false answers. The company still warns users below AI responses that “A.I. can make mistakes, so double-check responses.”

Google defended its system and criticised Oumi’s methodology. “This study has serious holes,” Google spokesperson Ned Adriance said. “It doesn’t reflect what people are actually searching on Google.”

At the same time, Google appears to be leaning further into AI-generated Search rather than scaling it back. The company said it is improving how links are surfaced inside AI answers and using techniques such as “query fan-out” to pull information from a wider range of websites and discussions across the web.

Read more: