2026
14 篇文章用時間點快照做回溯測試：Exa Snapshot 如何幫你把網頁資料變成可重現的評估環境
Exa 推出 Snapshot 功能，讓開發者用 snapshotAsOf 參數查詢過去某個時間點的網頁內容，解決模型評估中的資料洩漏與回測問題。
閱讀文章 ↗把賽前準備變成可查詢的流程：Google Search 三個備賽切入點
Google 在 2026 年 9 月 10 日說明 Search 的 AI 功能如何用於訓練計畫、跑步歌單與裝備比較。
閱讀文章 ↗從 Live Game Feed 到 Fantasy 串接：Google Search 如何把即時賽事變成可查詢的資料層
Google Search 新增即時賽事動態、跨聯盟比分輪播與 Yahoo/Sleeper 帳號串接，讓球迷與開發者都能用查詢取代手動追蹤。
閱讀文章 ↗AI Mode 旅遊規劃升級：追蹤票價、用點數訂房、直接完成飯店預訂
Google 在 AI Mode 加入三項旅遊功能：機票價格追蹤、顯示點數或哩程成本，以及在對話中直接預訂飯店。本文整理實際可用範圍與限制，供產品開發者參考。
閱讀文章 ↗Cloudflare AI Search：為你的 Agent 加上一個專屬搜尋引擎
Cloudflare AI Search 推出開發者體驗更新，讓 Agent 能直接搜尋你的資料，並提供可預測的定價預覽。本文整理重點功能與實際應用案例。
閱讀文章 ↗Google Images 25 週年：從文字搜尋到視覺生成
Google Images 迎來 25 週年，推出可瀏覽的首頁與 AI Overviews 影像生成功能。本文回顧視覺搜尋的演進，並探討對產品建構者的啟示。
閱讀文章 ↗Google I/O 2026：Pichai 宣告 agentic Gemini 時代，搜尋 25 年來最大改版
2026 年 5 月 19 至 20 日的 Google I/O 上，Sundar Pichai 宣告 agentic Gemini 時代：agent 走進搜尋、Gemini Spark 亮相、Gemini 3.5 Flash 發表，搜尋改版號稱逾 25 年來最大。
閱讀文章 ↗Exa Highlights：用 1/16 的 token 做到同等準確度，agent 搜尋的成本瓶頸有解了？
Exa 宣稱新版 Highlights 在部分 evals 上以約 94% 更少 token 產出更高品質結果：SimpleQA 上 500 字元追平前 8000 字元的準確度，token 少 16 倍。本文拆解逐請求、不快取、100 毫秒內完成的萃取機制，檢視自報數據的讀法與限制，並整理多輪 agent 搜尋的導入判斷。
閱讀文章 ↗Google 將 AI 學習工具直接放進搜尋：互動圖表、練習測驗與筆記本一次看懂
Google 在 2026 年 8 月推出五項 AI 搜尋學習功能，包括互動視覺、練習測驗、Lens 解題、Gemini Notebook 整合與自訂檔案。本文整理這些工具對學生與自學者的實際用途。
閱讀文章 ↗LinkedIn 擠進 AI 引用前五：專業查詢的第一來源
2026 年 3 月 10 日，兩份研究指出 LinkedIn 已成為 AI 聊天機器人的主要引用來源：Semrush 分析 32.5 萬個提示，將它列為第二名、僅次於 Reddit；Profound 則發現它三個月內從二十名外衝進前五，專業查詢更是第一名。
閱讀文章 ↗Perplexity 開源 pplx-embed：擴散預訓練打造網頁級檢索嵌入模型
Perplexity 於 2 月 26 日發布 pplx-embed-v1 與 pplx-embed-context-v1，各提供 0.6B 與 4B 開放權重，以擴散預訓練與雙向注意力主打網頁級檢索。本文解析架構、基準成績與成本工程。
閱讀文章 ↗Perplexity Model Council:三個前線模型同時作答,分歧也是答案
Perplexity 在 2 月 5 日推出 Model Council:同一個問題交給三個前線模型平行作答,再由另一個模型綜整共識與分歧。搭配升級版 Deep Research 與記憶改善,多模型交叉驗證正式走進搜尋產品。
閱讀文章 ↗ChatGPT 引用 Grokipedia:當 AI 開始引用 AI 生成的百科
《衛報》實測發現,GPT-5.2 回答超過十二個問題時引用 xAI 的 AI 生成百科 Grokipedia 九次,還重述已遭查證推翻的內容,Claude 也有類似案例。當聊天機器人引用 AI 寫的百科,來源把關成為所有 AI 搜尋產品的難題。
閱讀文章 ↗Google AI Overviews 醫療建議出錯:英國慈善團體提出警告
2026 年 1 月 2 日衛報調查發現,Google AI Overviews 在健康查詢上給出錯誤建議:胰臟癌飲食、肝指數正常範圍、抹片診斷角色都出錯。英國醫療慈善團體警告民眾暴露於風險,Google 回應稱絕大多數摘要準確。
閱讀文章 ↗
2025
4 篇文章Google AI Mode前進印度:英語先行、語音影像並進
2025年6月23日,Google在印度推出實驗性AI Mode搜尋:經Search Labs選用、初期僅支援英語、提供語音與Lens圖片查詢,由客製版Gemini 2.5驅動,早期測試者平均查詢長度達傳統搜尋的2到3倍。
閱讀文章 ↗彭博：蘋果高層曾內部討論收購 Perplexity
2025年6月20日彭博報導，蘋果高層曾就收購 AI 搜尋新創 Perplexity 進行內部討論；若成局將遠超 Beats 的30億美元紀錄。Perplexity 正以140億美元估值收尾5億美元募資，發言人則回應「不太可能」並否認已有併購討論。
閱讀文章 ↗Google AI Mode 開放語音對話：Search Live 走進搜尋
2025 年 6 月 18 日，Google 開始在 AI Mode 測試語音對話，美國 Labs 用戶可在 Google App 點 Live 圖示，用說的提問並追問，由客製化 Gemini 生成語音回答。本文整理功能細節、查詢擴散技術與後續規劃。
閱讀文章 ↗三星洽談入股Perplexity，Galaxy搜尋與Bixby將整合
2025年6月1日Bloomberg報導三星接近與Perplexity達成廣泛合作：參與新一輪融資、預載應用與助手、整合瀏覽器搜尋並可能支援Bixby；Perplexity正以140億美元估值洽談募集5億美元。
閱讀文章 ↗
2026
14 ARTICLESSearching the Past to Test What Agents Actually Solved
Exa Snapshot lets you query the web as of any date, so you can backtest agents and models without answer leakage.
READ POST ↗Search as a Race-Prep Tool: What the Three Workflows Actually Require
Google's Sept 10 post shows AI Mode handling training plans, playlists, and gear — but each depends on a setup step.
READ POST ↗Google Search's Live Game Feed and Fantasy Link Change How You Build for Real-Time Sports
Google Search adds a Live Game Feed, deeper stats, and fantasy account linking—a shift toward agentic, personalized search for builders.
READ POST ↗Google AI Mode Adds Flight Price Tracking, Points/Miles Rates, and Hotel Booking
Google's AI Mode in Search now supports flight price tracking, points/miles cost display, and hotel booking. Here's what changed and what it means for product builders.
READ POST ↗Cloudflare AI Search: Give Your Agents a Search Engine for Your Data
Cloudflare AI Search updates: index websites, skip sitemaps, public endpoints, MCP support, and predictable pricing with free embedding and reranking.
READ POST ↗Google Images at 25: From Text Links to AI Image Generation in Search
Google Images turns 25 with a new browseable homepage and AI image generation in AI Overviews. Here's what product builders need to know.
READ POST ↗Google I/O 2026: Pichai Declares the Agentic Gemini Era and Search's Biggest Overhaul in 25 Years
At Google I/O 2026 (May 19–20), Sundar Pichai declared the agentic Gemini era: agents in Search billed as the biggest overhaul in 25+ years, plus Gemini Spark and Gemini 3.5 Flash.
READ POST ↗When Search Engines Serve Thousands of AI Agents: Exa's DAG Pipeline for Observability
Exa uses a Directed Acyclic Graph (DAG) to orchestrate search pipelines for AI agents, enabling automatic parallelization, durable execution, and deep observability.
READ POST ↗Google Adds AI Study Tools to Search: Interactive Visuals, Practice Quizzes, and Notebooks
Google's new AI-powered study tools in Search include interactive visuals, practice quizzes, Lens help, notebooks, and custom file creation. Here's what builders should know.
READ POST ↗LinkedIn Is Now a Top-Cited Source in AI Chatbot Answers
Two studies put LinkedIn second only to Reddit across 325,000 AI prompts and first for professional queries — a clear sign discovery is shifting from search rankings to AI citations.
READ POST ↗Perplexity Open-Sources pplx-embed Retrieval Models
Perplexity released pplx-embed-v1 and pplx-embed-context-v1 on Feb 26 — open weights at 0.6B and 4B, diffusion-pretrained and bidirectional, built for web-scale retrieval and RAG pipelines.
READ POST ↗Perplexity's Model Council: Three Models Answer as One
Perplexity's Model Council runs one query through three frontier models in parallel, then synthesizes agreements and disagreements into one answer — ensemble verification shipped in a search product.
READ POST ↗ChatGPT Is Citing Grokipedia, and That's a Problem
Guardian tests found GPT-5.2 cited xAI's AI-written encyclopedia Grokipedia nine times, resurfacing debunked claims — a provenance problem for every AI search product.
READ POST ↗Google AI Overviews Health Answers Mislead, Charities Warn
A Guardian investigation found Google's AI Overviews giving false or misleading health advice, from cancer diets to mislabeled tests. UK charities warned of real harm; Google defended its accuracy.
READ POST ↗
2025
4 ARTICLESGoogle brings AI Mode to India in English first
On June 23, 2025, Google rolled out AI Mode in India via Search Labs: English only, with voice and Lens image queries, a custom Gemini 2.5, and 2-3x longer questions.
READ POST ↗Apple executives discussed buying Perplexity, Bloomberg says
Bloomberg reported June 20, 2025 that Apple executives held internal talks about buying AI search startup Perplexity, which was closing a $500 million round at $14 billion and denied deal talks.
READ POST ↗Google tests voice conversations in AI Mode via Search Live
On June 18, 2025, Google began testing voice conversations in AI Mode, letting US Labs users talk to Search and ask follow-ups. A custom Gemini powers it, fed by query fan-out.
READ POST ↗Samsung nears Perplexity deal covering Galaxy search, Bixby
Bloomberg reported June 1, 2025 that Samsung neared a wide Perplexity deal: investing in its round, preloading its app and assistant, browser search, and Bixby integration, at a $14 billion valuation.
READ POST ↗