Pinterest, Inc. (PINS) 2026 Q2 法說會逐字稿

Hello, everyone. Thank you for joining us, and welcome to Pinterest's second quarter 2026 earnings conference call. (Operator Instructions)

大家好。感謝各位加入我們，歡迎參加 Pinterest 2026 年第二季財報電話會議。(接線員指示)

I will now hand the conference over to Andrew Somberg, Vice President of Investor Relations and Treasury. Andrew, please go ahead.

接下來我將把會議交給投資人關係與財務管理副總裁 Andrew Somberg。Andrew，請開始。

Good afternoon, and thank you for joining us. Welcome to Pinterest's earnings call for the second quarter ended June 30, 2026. Joining me on today's call are Bill Ready, Pinterest's CEO; and Julia Donnelly, our CFO.

各位午安，感謝加入我們。歡迎參加 Pinterest 截至 2026 年 6 月 30 日止第二季的財報電話會議。今天與我一同出席的有 Pinterest 執行長 Bill Ready，以及我們的財務長 Julia Donnelly。

The statements we make on this call reflect management's view as of today and will include forward-looking statements. Such statements involve a number of assumptions, risks and uncertainties, and actual results may differ materially. We disclaim any obligation to update these statements. For information about assumptions, risks, uncertainties and other factors that could affect our results, please refer to our earnings press releases and the periodic reports we file with the SEC and are available on our Investor Relations website at investor.pinterest.com.

我們在本次電話會議中所做的陳述，反映管理團隊截至今日的觀點，並將包含前瞻性陳述。此類陳述涉及多項假設、風險與不確定性，實際結果可能有重大差異。我們不承擔更新這些陳述的任何義務。有關可能影響我們業績之假設、風險、不確定性及其他因素的資訊，請參閱我們的財報新聞稿，以及我們向美國證券交易委員會（SEC）提交的定期報告；相關資料亦可於我們的投資人關係網站 investor.pinterest.com 取得。

During this call, we will present both GAAP and non-GAAP financial measures. A reconciliation of non-GAAP to GAAP measures is included in today's earnings press release and presentation, which are distributed and available to the public through our Investor Relations website. Lastly, all growth rates discussed today are on a year-over-year basis unless otherwise specified.

在本次電話會議中，我們將同時呈現 GAAP 與非 GAAP 財務衡量指標。非 GAAP 與 GAAP 指標之調節表已包含於今日的財報新聞稿與簡報中，並已透過我們的投資人關係網站對外發布、供大眾查閱。最後，除非另有說明，今天討論的所有成長率皆以年對年為基準。

And now I'll turn the call over to Bill.

接下來我把電話交給 Bill。

Thanks, Andrew. Good afternoon, and thank you for joining our second quarter 2026 earnings call. Our strong Q2 results reflect the progress we're making against our strategic priorities. We ended the quarter with 640 million monthly active users and delivered revenue growth of 18% globally and in UCAN, a 5-point acceleration in our largest market.

謝謝你，Andrew。各位午安，感謝參加我們 2026 年第二季財報電話會議。我們第二季的強勁表現，反映出我們在策略優先事項上的推進成果。本季結束時，我們的月活躍用戶達 6.40 億，全球與 UCAN 地區營收皆成長 18%，在我們最大的市場中成長率加速 5 個百分點。

We continue to build momentum across our three strategic priorities. First, continuing to build a differentiated visual search, discovery and shopping experience for users. Second, keeping AI at the core of everything we do, from powering the user experience to our ads platform and our internal operations. And third, accelerating monetization through improved go-to-market and measurement capabilities.

我們在三大策略優先事項上持續累積動能。第一，持續為用戶打造具差異化的視覺搜尋、探索與購物體驗。第二，讓 AI 成為我們一切工作的核心，從驅動用戶體驗到廣告平台與內部營運。第三，透過改善上市策略（go-to-market）與衡量能力，加速變現。

At the center of our strategy is a simple idea. Pinterest helps people discover what they want and then go do it in the real world. That idea comes through clearly in our new brand campaign built around the line, the best thing you can find online is a reason to go offline. It reflects what makes Pinterest different. People come here with intent. They find inspiration that is personalized for them, refine their style and taste, and they can take action directly from our platform. The human curation that is so unique to Pinterest is the highly differentiated signal that trains our AI to make such exceptionally relevant recommendations to users.

我們策略的核心是一個簡單的理念。Pinterest 幫助人們發現自己想要的，然後在現實世界中去實現。這個理念在我們全新的品牌行銷活動中清楚呈現，主軸標語是：你能在網路上找到最棒的事物，是一個走到線下的理由。這反映了 Pinterest 與眾不同之處。人們帶著意圖來到這裡。他們找到為自己量身打造的靈感，精煉自己的風格與品味，並且可以直接在我們的平台上採取行動。Pinterest 獨特的人為策展，是高度差異化的訊號，用來訓練我們的 AI，為用戶提供極其相關的推薦。

Today, I will focus my remarks on two topics that cut across those priorities, how AI is powering our user momentum and how we're turning that strong engagement into more durable monetization. With that, let me start with the user side of the story.

今天我將把重點放在兩個橫跨上述優先事項的主題：AI 如何推動我們的用戶動能，以及我們如何把強勁的互動轉化為更具持久性的變現能力。接下來，先從用戶面向開始。

Q2 marked our 12th consecutive quarter of record users and our 11th straight quarter of double-digit user growth. In our core UCAN market, users grew 4%. We're also winning significantly with the next generation of users as Gen Z continues to be our largest and fastest-growing cohort, representing over half of our user base.

第二季是我們連續第 12 季創下用戶新高，也是連續第 11 季維持雙位數用戶成長。在我們核心的 UCAN 市場，用戶成長 4%。我們也在下一代用戶上取得顯著勝利，Z 世代仍是我們最大且成長最快的族群，占用戶基礎的一半以上。

AI has been at the heart of this momentum. We've transformed Pinterest into an AI-powered shopping assistant. Effectively, every Pin a user sees is personalized and served by AI. We were early to move on AI and its enabling technologies, having completed our full transition to GPU serving more than two years ago, which materially improved relevance and personalization using larger models and more data. Our scale and the unique human curation that occurs on our platform have fueled our proprietary Taste Graph.

AI 一直是這股動能的核心。我們已將 Pinterest 轉型為 AI 驅動的購物助理。實際上，用戶看到的每一個 Pin 都是由 AI 進行個人化並提供。我們很早就投入 AI 及其關鍵技術，兩年多前已完成全面轉換至 GPU 服務，透過更大型的模型與更多資料，實質提升了相關性與個人化程度。我們的規模，以及平台上獨特的人為策展，推動了我們自有的 Taste Graph（品味圖譜）。

With more than 80 billion monthly searches on Pinterest, the vast majority visual and over half commercial, and over 16 billion boards created on our platform, we have a uniquely valuable signal, human taste and curation at scale. This enables us to train our AI with a deep understanding of taste, context and intent that gets more personalized every time a user comes back. This powerful feedback loop is what helps users discover what they want before they have the exact words, brand or product in mind. That's why so many commercial journeys begin on Pinterest.

Pinterest 每月搜尋量超過 800 億次，其中絕大多數為視覺搜尋，且超過一半具有商業意圖；此外，我們的平台上已建立超過 160 億個看板（boards），因此我們擁有極具價值且獨特的訊號：以規模化的人類品味與策展。這使我們能以對品味、情境與意圖的深度理解來訓練 AI，且每次用戶回訪時都能變得更個人化。這個強大的回饋循環，能幫助用戶在尚未想到確切用詞、品牌或產品之前，就先發現自己想要的。這也是為什麼許多商業旅程都從 Pinterest 開始。

Today, over 96% of text-based searches on our platform are unbranded. For instance, cool running shoes, or new fall wardrobe ideas. This is a great moment for marketers to meet prospective customers when they have clear commercial intent, but haven't yet decided what to buy. As users move from inspiration to a more specific decision, their needs evolve from discovery to research with questions like, do these shoes run true to size? Or can you compare these two brands?

目前，我們平台上超過 96% 的文字搜尋都是非品牌搜尋。例如：很酷的跑鞋，或秋季新衣櫥穿搭靈感。這對行銷人員而言是絕佳時機：在潛在客戶具有明確商業意圖、但尚未決定要買什麼時與其接觸。當用戶從靈感階段走向更具體的決策時，他們的需求會從探索轉為研究，並提出像是：這雙鞋的尺寸是否準確？或是：你能比較這兩個品牌嗎？

To handle this part of the commercial journey, we're bringing AI further into the foreground of the user experience with our AI conversational layer, Pinterest Assistant. This intelligence layer, which we made available to the vast majority of US users as of the end of July, helps our users answer those later-stage research questions, enabling them to move from an idea to a finished plan or purchase entirely on Pinterest.

為了處理商業旅程中的這一段，我們透過 AI 對話層「Pinterest Assistant」讓 AI 更加前置於用戶體驗之中。截至 7 月底，我們已將這個智慧層提供給美國絕大多數用戶，協助用戶回答後期研究問題，使他們能在 Pinterest 上完成從想法到完整計畫或購買的全流程。

Pinterest Assistant is now woven throughout the user experience. Users can engage with it where they already explore and plan with contextual entry points and prompt suggestions across the experience, and we'll be able to generate product comparisons, step-by-step instructions and visual forward guidance.

Pinterest Assistant 現在已融入整體用戶體驗。用戶可以在原本就用來探索與規劃的地方與其互動，透過全體驗中的情境式入口與提示建議；我們也將能生成產品比較、逐步操作指引，以及視覺化的前瞻引導。

For example, someone planning a living room refresh can ask what styles work in their space, get help finding the right rug and explore ways to pull it all together within their budget, all without leaving Pinterest. What makes Pinterest intelligence differentiated is that it is built on our Taste Graph, years of visual curation signals from user searches, saves and boards, giving a deep understanding of products and style that is highly relevant to how people shop and to each individual's taste and interest. Notably, it's also visual first, consistent with the experience users expect from Pinterest.

例如，正在規劃客廳翻新的人，可以詢問哪些風格適合自己的空間，獲得協助找到合適的地毯，並在預算內探索如何把整體搭配完成，全程不必離開 Pinterest。Pinterest 智慧功能的差異化在於，它建立在我們的 Taste Graph 之上，累積多年來用戶搜尋、收藏（saves）與看板所產生的視覺策展訊號，對產品與風格具有深度理解，且高度貼近人們的購物方式與每個人的品味與興趣。值得注意的是，它也以視覺為優先，符合用戶對 Pinterest 的體驗期待。

We are also beginning to introduce memory and conversational history, creating an even more personalized experience. Importantly, we are building this capability in a highly differentiated, effective and cost-efficient way. Our approach to model deployment includes our own compact fit-for-purpose models built for Pinterest-specific use cases and suitable open-source models post-trained in our own environment within our secure cloud infrastructure.

我們也開始導入記憶與對話歷史，打造更個人化的體驗。重要的是，我們正以高度差異化、有效且具成本效率的方式建構這項能力。我們的模型部署方法包括：為 Pinterest 特定使用情境打造的自有精簡、適用目的（fit-for-purpose）模型，以及在我們安全的雲端基礎設施內、於自有環境中進行後訓練（post-train）的合適開源模型。

When we leverage open-source models, such as with Pinterest Assistant, we are seeing superior performance for our use cases when compared to closed third-party models because we are able to post-train open models on our highly unique data. With open models, we are achieving cost per transaction at less than 8% of the cost of comparable closed proprietary models. This gives us substantial headroom to deepen and extend these capabilities over time in a way that is differentiated, highly effective and cost efficient.

當我們運用開源模型（例如 Pinterest Assistant 所採用者）時，因為能以我們高度獨特的資料對開源模型進行後訓練，相較於封閉式第三方模型，我們在自身使用情境上看到更優異的表現。使用開源模型時，我們達成的每筆交易成本不到同等封閉式專有模型成本的 8%。這讓我們在未來能以差異化、高效且具成本效率的方式，持續深化並擴展這些能力，擁有相當大的發揮空間。

Stepping back, this is the next iteration of work we've been doing for years, taking the discovery and intent that already exists on Pinterest and making it even more actionable for users, helping them move seamlessly from dreaming to deciding, doing and buying. With the launch of our AI conversational layer, we are taking the next step to enhance that actionability even further and making it even more seamless for our users to move through their entire commercial journey on Pinterest. This is what sets up the monetization story I'll turn to next.

退一步來看，這是我們多年來持續推進工作的下一次迭代：把 Pinterest 上既有的探索與意圖變得更可執行，協助使用者從「想像」順暢地走到「決定、行動與購買」。隨著我們推出 AI 對話層，我們正邁出下一步，進一步提升可執行性，讓使用者在 Pinterest 上完成整段商業旅程的流轉更加無縫。這也為我接下來要談的變現故事奠定基礎。

On the monetization side, AI is helping us improve advertiser performance across the funnel from better targeting and bidding to more automated creative and stronger measurement. As we shared last quarter, advertisers using Pinterest Performance+ campaigns see meaningfully better ROAS and grow spend faster than those who haven't adopted it yet. That uplift comes from automating more of the campaign setup, bidding, budgeting, targeting and creative optimization that advertisers historically had to manage manually while still giving them clear controls.

在變現方面，AI 正協助我們在漏斗各階段提升廣告主成效，涵蓋更好的定向與出價、更自動化的創意，以及更強的衡量能力。如同我們在上季分享的，使用 Pinterest Performance+ 活動的廣告主，ROAS 顯著更佳，且相較於尚未採用者，投放支出成長更快。這樣的提升來自於將更多活動設定、出價、預算、定向與創意最佳化自動化；這些過去多由廣告主手動管理，同時我們仍提供清楚的控制選項。

As part of our efforts to serve a broader group of advertisers with Pinterest Performance+ campaigns, this quarter, we launched Smart Assembly to help more advertisers benefit from creative optimization tools. This is a new Pinterest Performance+ creative capability for advertisers promoting brands, services or content who do not have existing shopping product catalogs. Advertisers can upload multiple images and Pinterest automatically builds and serves the best performing ad for each impression.

作為我們擴大 Pinterest Performance+ 活動服務更廣泛廣告主群體的一部分，本季我們推出 Smart Assembly，讓更多廣告主受益於創意最佳化工具。這是 Pinterest Performance+ 的一項全新創意能力，適用於推廣品牌、服務或內容、但沒有既有購物商品目錄的廣告主。廣告主可上傳多張圖片，Pinterest 會自動為每一次曝光建立並投放表現最佳的廣告。

In early alpha testing, Smart Assembly delivered a 6% improvement in click-through rate on average, demonstrating that creative testing and diversification can meaningfully improve performance. Over time, our goal is for nearly every lower funnel campaign on Pinterest to start in an AI-powered best practice setup with added controls for more complex buyers who need them. Our near-term road map, including a simpler campaign creation flow, more sophisticated bidding and more automated creative, is designed to drive the next leg of Pinterest Performance+ adoption, including among advertisers that require both performance and increased control.

在早期 alpha 測試中，Smart Assembly 平均帶來 6% 的點擊率提升，顯示創意測試與多樣化能實質改善成效。長期而言，我們的目標是讓 Pinterest 上幾乎每一個下漏斗活動，都能以 AI 驅動的最佳實務設定起步，並為需要的更複雜買家提供額外控制。我們的近期路線圖（包含更簡化的活動建立流程、更精密的出價，以及更自動化的創意）旨在推動 Pinterest Performance+ 採用的下一階段成長，亦包括同時要求成效與更高控制度的廣告主。

We're also investing in Business Assistant, our conversational AI collaborator for advertisers, which is currently in beta. Business Assistant combines an advertiser's business context with Pinterest platform insights to surface actionable recommendations, including relevant trends, top-performing Pins and optimization opportunities. The goal is to help advertisers understand what is resonating on Pinterest, decide where to put the next dollar and make Pinterest easier to use and scale, especially for advertisers that do not have large, dedicated teams.

我們也在投資 Business Assistant（目前為 beta），這是我們為廣告主打造的對話式 AI 協作夥伴。Business Assistant 結合廣告主的業務情境與 Pinterest 平台洞察，提供可執行的建議，包括相關趨勢、表現最佳的 Pins，以及最佳化機會。目標是協助廣告主理解 Pinterest 上哪些內容正在引起共鳴，決定下一筆預算該投向何處，並讓 Pinterest 更容易使用與擴大規模，特別是對於沒有大型專職團隊的廣告主。

At the same time, we are upgrading our bidding and measurement systems so we can do two things better: help advertisers find the highest value impressions on Pinterest and prove that value through the metrics and measurement systems advertisers already use.

同時，我們正在升級出價與衡量系統，以便把兩件事做得更好：協助廣告主在 Pinterest 上找到價值最高的曝光，並透過廣告主既有使用的指標與衡量系統證明這些價值。

With a small initial group of some of our largest and most sophisticated advertisers, we are continuing to pilot integration between their in-house measurement systems and our AI bidding systems, allowing us to optimize bidding toward their unique set of desired outcomes. That initial group is seeing strong performance, partially contributing to our strong Q2 results.

我們正與一小群規模最大、也最成熟的廣告主持續試點，將其內部衡量系統與我們的 AI 出價系統整合，使我們能依其獨特的目標結果來最佳化出價。這個初始群體已看到強勁成效，並部分促成我們 Q2 的亮眼表現。

We are now expanding testing to a small number of additional advertisers over the course of the third quarter. The learnings are also informing broader bidding enhancements across the platform that can scale to many more advertisers. We are always working on making our core ad delivery engine better through adapting new and better signals and driving enhancements in our AI modeling work.

我們目前正於第三季期間，將測試擴大到少數其他廣告主。相關學習也正在推動平台更廣泛的出價強化，並可擴展至更多廣告主。我們始終致力於讓核心廣告投放引擎更好，包含導入更新、更優的訊號，並推進 AI 建模工作的強化。

This quarter, we have made our shopping ad delivery systems better at deciding which products to show, when to show them and how to allocate budgets against the highest value advertiser opportunities. That includes improving product selection for advertisers with large catalogs and increasing the variety of relevant products users see. This package of changes are designed to improve lower funnel performance across both large enterprises and smaller advertisers. The early results are encouraging, helping drive stronger, more consistent ROAS.

本季，我們讓購物廣告投放系統在判斷「要展示哪些商品、何時展示，以及如何將預算配置到價值最高的廣告主機會」方面做得更好。這包括為擁有大型商品目錄的廣告主改善商品選擇，並提升使用者看到的相關商品多樣性。這一系列變更旨在同時提升大型企業與中小型廣告主的下漏斗成效。早期結果令人鼓舞，有助於帶動更強且更一致的 ROAS。

Finally, we are continuing to extend Pinterest's unique consumer intent signal and audience to connected TV through tvScientific, and early advertiser reception continues to be strong. In 2027, we expect to fully integrate tvScientific capabilities directly into Pinterest Performance+, turning Pinterest into a full funnel search, social and CTV platform performance solution and opening access to larger and incremental budget pools. We're pairing this progress across our ad platform with a more disciplined performance-oriented sales and go-to-market motion, helping us better monetize the strong engagement and commercial intent on Pinterest.

最後，我們持續透過 tvScientific，將 Pinterest 獨特的消費者意圖訊號與受眾延伸到連網電視（CTV），且廣告主的早期反饋依然強勁。我們預期在 2027 年把 tvScientific 的能力完整整合進 Pinterest Performance+，使 Pinterest 成為涵蓋全漏斗的搜尋、社群與 CTV 平台成效解決方案，並開啟更大且增量的預算池。我們也將廣告平台的進展，與更具紀律、以成效為導向的銷售與上市（go-to-market）推進方式相結合，協助我們更好地將 Pinterest 上強勁的互動與商業意圖變現。

Early progress on sales and go-to-market transformation was reflected in our strong Q2 results, particularly in UCAN. Our focus on clear seller accountability, better packaging of commercial moments and a tighter connection between product performance and advertiser conversations are supporting strong UCAN demand. We're also making progress on our mid-market and managed SMB go-to-market efforts, including restructuring account coverage teams and realigning incentives to better serve this cohort over time. We're now applying our UCAN playbook internationally, where we see significant opportunity to close the gap between engagement and revenue. We're earlier in this work, but our progress in UCAN gives us confidence in the opportunity.

銷售與 go-to-market 轉型的早期進展，已反映在我們強勁的 Q2 表現中，特別是在 UCAN。我們聚焦於更清楚的銷售責任歸屬、更好的商業時刻（commercial moments）包裝，以及產品成效與廣告主對話之間更緊密的連結，正支撐 UCAN 的強勁需求。我們也在中型市場與受管理的 SMB go-to-market 工作上取得進展，包括重整客戶覆蓋團隊並重新調整激勵機制，以便長期更好地服務這一族群。我們正把 UCAN 的作戰手冊應用到國際市場，在那裡我們看到縮小互動與營收差距的重大機會。這項工作仍在早期階段，但 UCAN 的進展讓我們對機會更有信心。

With new international leadership, we're sharpening our global go-to-market approach and bringing more global discipline to how we drive performance selling internationally. We're also testing the expansion of third-party demand into Europe. Taken together, the product-led improvements across our ad platform and a more disciplined go-to-market engine are strengthening monetization and helping revenue better reflect the value of the engagement we're seeing on Pinterest.

在新的國際領導團隊帶領下，我們正強化全球 go-to-market 方法，並在國際市場推動成效型銷售時引入更多全球一致的紀律。我們也在測試將第三方需求擴展至歐洲。綜合而言，廣告平台的產品驅動改善與更具紀律的 go-to-market 引擎，正在強化變現能力，並讓營收更能反映我們在 Pinterest 上看到的互動價值。

In closing, Q2 shows that we are making progress on the priorities that matter most. We are building an even more differentiated visual search, discovery and shopping experience. AI is improving both the user experience and advertiser performance and stronger go-to-market and measurement capabilities are helping revenue better reflect the engagement we see on Pinterest. We are still early on many of these initiatives, but the initial results indicate we are on the right path. As we build for the long term, we remain focused on making Pinterest a positive platform centered on time well spent.

總結來說，Q2 顯示我們在最重要的優先事項上正取得進展。我們正在打造更具差異化的視覺搜尋、探索與購物體驗。AI 正同時改善使用者體驗與廣告主成效，而更強的 go-to-market 與衡量能力，正協助營收更好地反映我們在 Pinterest 上看到的互動。我們在許多計畫上仍處於早期階段，但初步結果顯示我們走在正確的道路上。在為長期建設的同時，我們仍專注於讓 Pinterest 成為一個以「時間花得值得」為核心的正向平台。

As the global conversation around online safety and youth well-being continues to grow, we believe that foundation matters more than ever. We will keep making deliberate choices that put user trust and well-being at the center of the experience, and we remain confident that building a positive platform and building a strong business reinforce each other.

隨著全球對線上安全與青少年福祉的討論持續升溫，我們相信這樣的基礎比以往任何時候都更重要。我們將持續做出審慎的選擇，把使用者信任與福祉置於體驗核心；我們也仍然有信心，打造正向平台與建立強健業務能相互強化。

With that, I'll turn it over to Julia to walk through the Q2 financials and our outlook in more detail.

接下來，我把時間交給 Julia，請她更詳細說明 Q2 財務表現與我們的展望。

Thanks, Bill, and good afternoon, everyone. Today, I'll be discussing our second quarter 2026 financial results and provide an update on our third quarter 2026 outlook. All financial metrics, except for revenue, will be discussed in non-GAAP terms unless otherwise specified, and all comparisons will be discussed on a year-over-year basis unless otherwise noted. Q2 was a strong quarter. We've transformed Pinterest into a scaled AI-powered shopping destination for 106 million users in the US and Canada and 640 million users globally.

謝謝你，Bill，各位下午好。今天我將討論我們 2026 年第二季的財務結果，並更新我們對 2026 年第三季的展望。除營收外，所有財務指標除非另有說明，將以非 GAAP 口徑討論；除非另有註明，所有比較將以年對年為基礎。Q2 是表現強勁的一季。我們已將 Pinterest 轉型為具規模、由 AI 驅動的購物目的地，在美國與加拿大服務 1.06 億名使用者，全球則有 6.40 億名使用者。

Importantly, we have a direct relationship with our users. 100% of our reported users are logged in and 85% come to Pinterest directly through our mobile app, meaning we are not heavily reliant on third parties for traffic. Revenue exceeded $1 billion for the fourth consecutive quarter, growing 18% year-over-year. This growth was led by strength in UCAN, our core market, where revenue growth accelerated to 18% year-over-year.

重要的是，我們與使用者之間有直接關係。我們所報告的使用者 100% 都是登入狀態，且有 85% 透過我們的行動 App 直接造訪 Pinterest，這表示我們不高度依賴第三方帶來流量。營收連續第四季超過 10 億美元，年增 18%。這項成長主要由我們的核心市場 UCAN 的強勁表現帶動，該市場的營收成長加速至年增 18%。

This performance reflects a number of our initiatives coming together. First, our ongoing ad platform enhancements to improve return on ad spend and to advance our bidding and measurement capabilities are driving advertiser performance and spend. In addition, we are seeing the early results of our sales and go-to-market transformation take hold in UCAN, where increased rigor and discipline in how we go to market are beginning to show up in our results.

這項表現反映了我們多項計畫的共同推進。首先，我們持續強化廣告平台，以提升廣告支出回報（ROAS），並推進我們的出價與衡量能力，正帶動廣告主成效與投放支出。此外，我們也看到銷售與市場進入（go-to-market）轉型在美國與加拿大（UCAN）開始發揮早期成效；我們在市場進入方式上更嚴謹、更有紀律，正逐步反映在成果上。

As we discussed entering 2026, we have more work to do, so our revenue consistently reflects the strength of our user activity. While this will take some time to fully take hold, particularly internationally, our results in UCAN give us confidence that we are moving in the right direction.

如同我們在進入 2026 年時所討論的，我們仍有更多工作要做，讓營收能更一致地反映用戶活動的強勁程度。雖然這需要一些時間才能完全落實，尤其是在國際市場，但我們在 UCAN 的成果讓我們有信心正朝正確方向前進。

Now I'll move to the details of our second quarter results. We ended the quarter with 640 million global monthly active users or MAUs, growing 11% and reaching another record high. We continue to demonstrate user growth across all of our geographic regions. In Q2, our US and Canada region had 106 million MAUs, growing 4%. Our Europe region had 157 million MAUs growing 8%. And in the Rest of World markets, we had 377 million MAUs growing 15%.

接下來我將說明第二季業績細節。本季結束時，我們的全球月活躍用戶（MAU）達 6.40 億，年增 11%，再創新高。我們持續在所有地理區域展現用戶成長。第二季，美國與加拿大地區 MAU 為 1.06 億，成長 4%。歐洲地區 MAU 為 1.57 億，成長 8%。而在世界其他地區市場，MAU 為 3.77 億，成長 15%。

Shifting to revenue. In Q2, our global revenue was $1.180 billion, up 18% or up 17% on a constant currency basis, with strength led by our conversion and consideration objectives. Across verticals, we continue to see strength in retail as well as smaller but faster-growing emerging verticals on our platform, including financial services, travel and health.

接著談營收。第二季全球營收為 11.80 億美元，年增 18%；以固定匯率計年增 17%，成長動能主要由轉換（conversion）與考慮（consideration）目標帶動。就產業別而言，我們持續看到零售的強勁表現，以及平台上規模較小但成長更快的新興垂直領域，包括金融服務、旅遊與健康。

Turning to our geographical breakouts for Q2. Revenue in the US and Canada was $880 million, growing 18%. Growth came from retail and emerging verticals, including financial services, travel and health. In Europe, revenue was $213 million, growing 12% on a reported basis or 7% on a constant currency basis. Growth in Europe was driven by retail. Revenue from Rest of World was $87 million, growing 38% on a reported basis or 32% on a constant currency basis.

接著看第二季各地區拆分。美國與加拿大營收為 8.80 億美元，成長 18%。成長來自零售與新興垂直領域，包括金融服務、旅遊與健康。歐洲營收為 2.13 億美元，按報告基礎成長 12%，以固定匯率計成長 7%。歐洲的成長由零售帶動。世界其他地區營收為 8,700 萬美元，按報告基礎成長 38%，以固定匯率計成長 32%。

In Q2, overall ad impressions grew 16%. The deceleration from prior quarters was partly driven by lapping the ramp in ad impressions from previously undermonetized international markets. Ad pricing in Q2 increased 1% year-over-year. This was driven in part by stronger demand in our UCAN region as well as the higher relative mix of UCAN ad impressions, which carry higher average pricing overall.

第二季整體廣告曝光量年增 16%。相較前幾季的放緩，部分原因是去年同期基期包含先前變現不足的國際市場廣告曝光量快速拉升所致。第二季廣告定價年增 1%。這部分由 UCAN 地區需求更強，以及 UCAN 廣告曝光占比相對提高所帶動；UCAN 的平均定價整體較高。

Moving to expenses. In Q2, cost of revenue was $245 million, up 25% year-over-year and up 6% versus Q1, driven by the full quarter impact from tvScientific and our investment in additional GPU capacity. Our non-GAAP operating expense was $629 million, up 13%. The increase was primarily driven by sales and marketing due to our brand campaign and sales headcount investments as well as R&D to support our AI and product initiatives.

接著談費用。第二季營收成本為 2.45 億美元，年增 25%，較第一季增加 6%，主要受 tvScientific 全季影響，以及我們投資新增 GPU 產能所帶動。非 GAAP 營運費用為 6.29 億美元，成長 13%。增加主要來自銷售與行銷（品牌活動與銷售人力投資），以及研發以支持我們的 AI 與產品計畫。

In Q2, we delivered $311 million in adjusted EBITDA with an adjusted EBITDA margin of 26%, up 130 basis points versus Q2 last year. The higher-than-expected adjusted EBITDA was driven by flow-through from higher revenue. We also delivered Q2 free cash flow of $270 million. On a trailing 12-month basis, we've generated nearly $1.3 billion of free cash flow, representing 94% free cash flow conversion.

第二季我們達成調整後 EBITDA 3.11 億美元，調整後 EBITDA 利潤率為 26%，較去年第二季提升 130 個基點。高於預期的調整後 EBITDA 主要來自較高營收的貢獻（flow-through）。我們第二季自由現金流為 2.70 億美元。以過去 12 個月計，我們產生了接近 13 億美元的自由現金流，自由現金流轉換率為 94%。

In Q2, we allocated $58 million towards share repurchases. Separately, we entered into a capped call transaction for a total consideration of $99 million, which protects against dilution from our previously issued convertible notes up to a price of $30.59 per share. As a reminder, year-to-date, we've repurchased over $2 billion of stock, retiring nearly 111 million shares. We ended the quarter with cash, cash equivalents and marketable securities of $1.3 billion.

第二季我們投入 5,800 萬美元進行庫藏股回購。另外，我們簽訂了一筆上限買權（capped call）交易，總對價 9,900 萬美元，可在每股 30.59 美元的價格上限內，防範先前發行之可轉換公司債所造成的稀釋。提醒一下，截至今年迄今，我們已回購超過 20 億美元股票，註銷近 1.11 億股。本季結束時，我們持有的現金、約當現金與有價證券為 13 億美元。

Now I'll discuss our guidance for the third quarter. We expect Q3 revenue to be in the range of $1.190 billion to $1.210 billion, representing 13% to 15% growth year-over-year.

接下來我將說明第三季展望。我們預期第三季營收將介於 11.90 億至 12.10 億美元之間，代表年增 13% 至 15%。

As we move from Q2 to Q3, there are a few sequential factors to keep in mind. First, based on current spot rates, we expect foreign exchange to be a modest headwind in Q3 after providing a 1-point tailwind in Q2. Second, the shift of Prime Day from Q3 last year into Q2 this year resulted in an approximately 0.5-point benefit to Q2 and will represent a roughly 0.5-point headwind to Q3 as multiple brands and retailers increase their advertising spend around that moment. Lastly, in Q2, we saw a nearly 1-point benefit from World Cup-related spend that will not repeat in Q3.

從第二季進入第三季，有幾個需要留意的季對季因素。第一，依目前即期匯率，我們預期外匯在第三季將帶來小幅逆風；而第二季則提供了 1 個百分點的順風。第二，去年 Prime Day 落在第三季、今年移至第二季，約為第二季帶來 0.5 個百分點的助益；而在第三季將形成約 0.5 個百分點的逆風，因多個品牌與零售商會在該時點前後提高廣告支出。最後，第二季我們看到與世界盃相關的投放支出帶來接近 1 個百分點的助益，第三季不會重現。

Moving down the P&L. We expect Q3 adjusted EBITDA to be in the range of $335 million to $355 million. We anticipate Q3 non-GAAP cost of revenue expense to be roughly flat versus Q2 2026 due to the accelerated recognition of certain contractual benefits in Q3 associated with our recently executed multiyear infrastructure agreement. In Q3, within non-GAAP operating expense, our primary area of year-over-year investment will be sales and marketing and R&D to support our AI and product initiatives.

再往下看損益表。我們預期第三季調整後 EBITDA 將介於 3.35 億至 3.55 億美元。由於我們近期簽署的多年期基礎設施協議中，部分合約利益將在第三季加速認列，我們預期第三季非 GAAP 營收成本費用相較 2026 年第二季大致持平。第三季在非 GAAP 營運費用中，我們年對年投資的主要領域將是銷售與行銷，以及研發，以支持我們的 AI 與產品計畫。

As we saw in the first half of 2026, we continue to expect modest headwinds from cost of revenue as a percentage of revenue in the second half as a result of the investments in areas such as additional GPU capacity as well as the impact from the inclusion of tvScientific. Given our first half revenue outperformance, we now expect full year 2026 adjusted EBITDA margins of approximately 30% versus our prior expectation of approximately 29%.

如同我們在 2026 年上半年所見，由於在新增 GPU 產能等領域的投資，以及納入 tvScientific 的影響，我們仍預期下半年營收成本占營收比重將面臨溫和逆風。鑑於上半年營收表現優於預期，我們現在預期 2026 全年調整後 EBITDA 利潤率約為 30%，高於先前預期的約 29%。

In closing, I'm proud of our teams for yet another strong quarter of results as we execute against our strategic priorities. I'm encouraged by our performance in the first half of the year as we continue to deliver for our users and advertisers.

最後，我為團隊感到自豪，我們在執行策略優先事項的同時，再次交出強勁的一季成果。我們對上半年表現感到鼓舞，並將持續為用戶與廣告主創造價值。

With that, I'll hand it over to Bill for some final words.

接下來我把時間交給 Bill 做最後的補充。

Thanks, Julia. I want to thank our teams at Pinterest, our advertising partners and all the people that come to Pinterest to find inspiration and take action. And with that, we can open up the call for questions.

謝謝，Julia。我要感謝 Pinterest 的團隊、我們的廣告合作夥伴，以及所有來到 Pinterest 尋找靈感並採取行動的人。接著，我們可以開放提問。

(Operator Instructions) Mark Shmulik, Bernstein.

（接線員指示）Bernstein 的 Mark Shmulik。

Julie, you shared a little bit of color on that kind of revenue guide puts and takes, but hoping to get a little bit more color on perhaps some of the regional differences you might be seeing. I noticed a little bit of a slowdown in Europe. Is that just kind of tougher compares?

Julie，你對這份營收指引的利多與利空因素分享了一些背景，但希望能再多了解一些你們可能看到的區域差異。我注意到歐洲似乎有些放緩。那只是因為比較基期更具挑戰嗎？

Sure. Thanks, Mark. So yes, overall, we're really pleased with a very strong Q2. And on a global basis for Q3, the high end of our Q3 guidance range is consistent with what we just delivered in Q2 when you adjust for the 3-point sequential impact I called out in my prepared remarks from foreign currency, Prime Day and World Cup spend.

當然。謝謝你，Mark。是的，整體而言，我們對第二季非常強勁的表現感到很滿意。而就第三季的全球層面來看，在你把我在事先準備的講稿中提到的三個季對季影響（外匯、Prime Day 與世界盃投放支出）調整後，我們第三季指引區間的上緣與我們第二季剛交出的表現是一致的。

At the core of our Q2 outperformance is our ongoing strength in users and engagement. We continue to win shopping demand from consumers globally and in UCAN, and that multiyear trend gives us the most conviction in the trajectory of our business. We're now making it easier for advertisers to tap into that consumer behavior through our AI-driven ad platform improvements and the early dividends from our go-to-market transformation. And that's really what's translating into the strong revenue performance that you saw in Q2, particularly in UCAN.

我們第二季表現優於預期的核心，在於我們在使用者與互動參與度方面持續展現的強勁動能。我們持續在全球以及 UCAN 市場贏得消費者的購物需求，而這個延續多年的趨勢，讓我們對業務發展軌跡最具信心。我們也正透過 AI 驅動的廣告平台改進，以及市場推進（go-to-market）轉型帶來的初步成果，讓廣告主更容易切入這些消費者行為。而這正是你在第二季看到的強勁營收表現（尤其在 UCAN）背後的關鍵。

So let's start with UCAN since that was the biggest revenue driver in Q2. We drove a 5-point acceleration sequentially to 18% year-over-year in UCAN, and this reflects a few different things coming together. So first, as a result of our ad platform improvements, we began to see pockets of large retailer spend accelerate with certain retailers leaning in as a result of our ongoing ROAS enhancements and AI-driven bidding optimizations as well as Asia-based cross-border retailer spend into UCAN.

那我們先從 UCAN 開始，因為它是第二季最大的營收驅動因素。我們在 UCAN 的年增率達到 18%，較前一季加速 5 個百分點，這反映了幾項因素同時發酵。首先，受惠於我們的廣告平台改進，我們開始看到部分大型零售商的投放加速；一些零售商因我們持續提升 ROAS（廣告投資報酬率）以及 AI 驅動的出價最佳化而加大投入，同時也包括以亞洲為基地的跨境零售商在 UCAN 的投放。

Second, in UCAN in Q2, the strong year-over-year growth outside of our largest retailers continued across mid-market, managed SMB and emerging verticals as many of these advertisers are benefiting from our ad platform improvements, including Pinterest Performance+ campaign specifically.

第二，在 UCAN 的第二季，除我們最大型零售商之外的年增強勁成長，持續擴及中型市場、受管理的中小企業（SMB）以及新興垂直領域；其中許多廣告主正受惠於我們的廣告平台改進，尤其是 Pinterest Performance+ 活動。

Lastly, UCAN in Q2 also benefited from Prime Day, shifting from Q3 to Q2, World Cup spend, and a full quarter contribution from tvScientific. So overall, as we look to Q3, we expect strong growth in UCAN to continue.

最後，UCAN 在第二季也受惠於 Prime Day 從第三季移到第二季、世界盃相關投放，以及 tvScientific 的完整季度貢獻。因此整體而言，展望第三季，我們預期 UCAN 的強勁成長將會延續。

For Europe and Rest of World, we noted on the last earnings call that we expected growth to moderate in Q2 as a result of the deliberate leadership and structural changes we're making to our go-to-market organization and more difficult last year comparisons in certain markets. So in Q2, we were lapping the ramp of resellers in Rest of World last year and lapping a significant influx of cross-border spend into Europe last year following changes in the US tariff environment. That moderation was generally consistent with our expectations.

至於歐洲與世界其他地區（Rest of World），我們在上一通財報電話會議中提到，因我們正對市場推進（go-to-market）組織進行有意識的領導與結構調整，以及部分市場去年同期比較基期更具挑戰，我們預期第二季成長將放緩。因此在第二季，我們面臨去年世界其他地區經銷商（resellers）擴張所形成的高基期，以及去年因美國關稅環境變化而帶動跨境投放大量流入歐洲所形成的高基期。這樣的放緩大致符合我們的預期。

In addition, we faced incremental pressure mid-quarter from Asia-based cross-border retailers impacted by regulatory actions, particularly in Europe, and that pressure is continuing in the third quarter. We have more work to do to fully realize the benefits of our go-to-market transformation in Europe and Rest of World, and we expect some level of disruption to continue in Q3.

此外，我們在季中也面臨來自以亞洲為基地的跨境零售商的額外壓力，這些零售商受到監管行動影響，尤其在歐洲；而這股壓力在第三季仍將持續。我們仍需做更多工作，才能充分實現市場推進（go-to-market）轉型在歐洲與世界其他地區的效益，我們預期第三季仍會持續出現一定程度的擾動。

As we look to the third quarter, it's also worth noting that Q3 represents the most difficult comparison of the year for our Europe region in particular. So stepping back, the underlying trends in our business remain healthy. We're pleased with the strong Q2 overall, led by growth in our core UCAN market, in particular, which continues to be underpinned by our durable user and engagement strength that we continue to see.

展望第三季，也值得注意的是，第三季對我們的歐洲區域而言，將是全年最具挑戰的比較基期。退一步看，我們業務的基本趨勢仍然健康。我們對第二季整體的強勁表現感到滿意，尤其是由核心 UCAN 市場的成長所帶動；而這仍持續建立在我們所看到的、穩健且持久的使用者與互動參與度優勢之上。

Colin Sebastian, Baird.

Colin Sebastian，Baird。

Bill, you talked about the evolution of the Pinterest Assistant and updates there as usage takes hold. And I guess as the competitive landscape continues to evolve, what are you seeing in terms of early engagement signals there? And then how are you innovating on the broader Pinterest user experience side?

Bill，你談到 Pinterest Assistant 的演進以及在使用逐步普及後的更新。我想在競爭格局持續演變之際，你們在那裡看到哪些早期的互動參與訊號？另外，在更廣泛的 Pinterest 使用者體驗方面，你們如何持續創新？

Thanks for the question, Colin. Over the last three years, we've effectively turned Pinterest into an AI-driven shopping assistant. So even today, every pin you see on Pinterest is personalized and served by our AI. And that's been really at the core of the significant acceleration in our user growth over that three-ish-year time period. And what we're doing now is we're bringing more of those AI capabilities in the foreground of the experience for users to interact with directly. So I'll give you a little bit more detail about how we're doing that.

謝謝你的提問，Colin。過去三年，我們已有效地把 Pinterest 打造成一個由 AI 驅動的購物助理。所以即使在今天，你在 Pinterest 上看到的每一個 Pin，都是由我們的 AI 進行個人化並提供給你。而這正是我們在這大約三年的期間內，使用者成長顯著加速的核心原因。我們現在正在做的是，把更多 AI 能力帶到使用者體驗的前景位置，讓使用者可以直接互動。我會再多提供一些我們如何做到這點的細節。

We're leveraging AI to make Pinterest more personalized, more actionable and more valuable users. And that advantage continues to compound, particularly the human curation on the platform. So we've been applying AI at scale for years. We completed our migration to GPU serving over two years ago.

我們正運用 AI 讓 Pinterest 更個人化、更可採取行動，並為使用者帶來更高價值。而這個優勢仍在持續累積，尤其是平台上的人工策展（human curation）。我們多年來一直在大規模應用 AI。我們在兩年多前完成了遷移至 GPU 服務（GPU serving）。

And as I mentioned, every pin you see on Pinterest is effectively served by AI. But it's really that human curation that 640 million-plus users on our platform that are creating hundreds of billions of unique connections across our Taste Graph that really gives us proprietary signal that we can use to train our AI and personalize the experience with every interaction. And that feedback loop has accelerated over the past three years. As consumers are responding, that's what's led to the 12 consecutive quarters of record users, 11 straight quarters of double-digit user growth.

如我所提到的，你在 Pinterest 上看到的每一個 Pin，基本上都是由 AI 提供。但真正關鍵的是平台上超過 6.4 億名使用者的人工策展，他們在我們的 Taste Graph 上建立了數千億個獨特連結，這為我們提供了專有訊號，可用來訓練 AI，並在每一次互動中個人化體驗。而這個回饋循環在過去三年加速運轉。隨著消費者的回應，這也帶來了連續 12 個季度創新高的使用者數，以及連續 11 個季度的雙位數使用者成長。

And on the competitive landscape, we've long believed search will continue to fragment with a small number of horizontal general-purpose platforms, but more vertical-specific experiences that solve distinct consumer needs. And that trend of the search market fragmenting has actually been happening for more than two decades. Think of how many product searches start on large retailers or travel searches at online travel companies versus a general-purpose search engine.

至於競爭格局，我們長期以來相信，搜尋將持續碎片化：會有少數橫向、通用型的平台，但也會出現更多垂直領域的特定體驗，以解決不同的消費者需求。而搜尋市場碎片化的趨勢，其實已經發生了二十多年。想想有多少商品搜尋是從大型零售商開始，或旅遊搜尋是在線上旅遊公司進行，而不是透過通用型搜尋引擎。

So as AI becomes more widely available and open-source models continue to improve, we believe there can be many winners, particularly in vertical-specific search. The user growth in our platform alongside the growth in AI over these last few years is very clear evidence of that. You're also seeing that the general-purpose AI applications, they're discovering what those of us that have built much larger platforms have known for a long time that it's very hard to be all things to all people. And you've seen them pull back from many of their vertical-specific experiences.

因此，隨著 AI 更廣泛普及、開源模型持續進步，我們相信會有許多贏家，尤其是在垂直領域的特定搜尋。我們平台的使用者成長，以及過去幾年 AI 的成長，正清楚證明了這一點。你也會看到通用型 AI 應用正在發現：那些打造過更大型平台的人早就知道的一件事——要同時滿足所有人的所有需求非常困難。而你也看到他們從許多垂直領域的特定體驗中撤退。

So at Pinterest, we're very focused on visual search, particularly around shopping, and we're using AI to make the experience more actionable across the shopping journey. People already come to Pinterest to discover products, ideas and aesthetics they love. And research shows that shoppers find AI chatbots most valuable when researching and comparing products and narrowing their choices.

因此在 Pinterest，我們非常聚焦於視覺搜尋，特別是圍繞購物場景，並運用 AI 讓整個購物旅程的體驗更可採取行動。人們本來就會來 Pinterest 發掘他們喜愛的產品、點子與美學風格。研究也顯示，購物者在研究與比較產品、以及縮小選擇範圍時，最能感受到 AI 聊天機器人的價值。

The Pinterest Assistant, our AI conversational layer that we've now made available for the vast majority of US users, it brings those LLM capabilities into our visual, personalized and proactive experience. So it helps users move from inspiration through consideration and decision-making and allows them to complete more of their shopping journey on Pinterest.

Pinterest Assistant 是我們的 AI 對話層，目前已提供給絕大多數美國使用者使用；它把這些 LLM 能力帶入我們的視覺化、個人化且主動式的體驗中。因此它能協助使用者從靈感啟發，走向評估與決策，並讓他們能在 Pinterest 上完成更多購物旅程。

By combining our proprietary signals with cost-efficient open-source models, that really allows us to run those securely in our own environment. So we're able to keep improving the experience, scale those innovations with strong ROI. And the result is a better product today and compounding advantage as more people use Pinterest.

透過結合我們的專有訊號與具成本效率的開源模型，讓我們能在自有環境中安全地運行這些模型。因此我們能持續改善體驗，並以良好的 ROI（投資報酬率）擴大這些創新。結果就是：今天的產品更好，而且隨著更多人使用 Pinterest，我們的優勢會持續累積。

And lastly, I'll just say, it has been really, really exciting to see just how much the open-source ecosystem is expanding. We've been very early on that. It's been woven into our experience. It's core to what we're doing, part of how we're doing that very effectively. But it's great to see that momentum building. And this is one good example of how we're using that, but I think you're going to see more of that across the ecosystem as well.

最後我想說，看到開源生態系擴張得如此之快，真的非常、非常令人振奮。我們很早就投入其中。它已被編織進我們的體驗之中。它是我們正在做的事情的核心，也是我們能非常有效落地的方式之一。但看到這股動能持續增強，真的很棒。這是一個我們如何運用它的好例子，但我想你也會在整個生態系中看到更多類似的應用。

Brian Nowak, Morgan Stanley.

Brian Nowak，Morgan Stanley。

Maybe I want to ask about the UCAN. It seems like you've made a lot of progress on the improvement in attribution, the entire go-to-market on the UCAN. Bill, let me ask you, as you sort of look ahead the next 12 months, what are the next sort of sources of innovation or unlocks that you're looking for to even drive faster and more durable growth out of the UCAN business?

我想問一下 UCAN。看起來你們在歸因（attribution）改善，以及 UCAN 的整體市場推進（go-to-market）方面都取得了很大進展。Bill，我想請教你，當你展望未來 12 個月時，你認為接下來有哪些創新來源或關鍵解鎖點，可以推動 UCAN 業務更快且更具持久性的成長？

Yes. Thanks for the question, Brian. First thing I'd say is just we've talked about this previously. We're still a long way from having fully monetized all the commercial intent that already exists on our platform. So while user growth and engagement continues to be one of the highlights of our platform, especially as you look across sort of the social media landscape where you're seeing platforms they're generally having a harder time growing, 11 straight quarters of double-digit user growth, that user growth and the fact that more than half of them are here to shop continues to be the strongest part of that.

是的。謝謝你的提問，Brian。我首先想說的是，我們之前已經談過這件事。我們距離把平台上既有的所有商業意圖完全變現，仍然還有很長一段路要走。因此，雖然用戶成長與互動持續是我們平台的一大亮點，尤其是放眼整體社群媒體版圖，你會看到各平台普遍更難成長；我們已連續 11 個季度維持雙位數的用戶成長，而這樣的用戶成長，以及其中超過一半是為了購物而來，仍然是這一切中最強的部分。

But as we've shared before, we shared on the last call, we 5x the number of clicks to advertisers over that last three-year period, but we certainly didn't 5x the revenue. So the things that we've talked about in terms of aligning our AI bidding systems with measurement systems, which I talked about in my remarks, getting more integrated into a broader set of measurement systems so that the advertisers can measure those things.

但如同我們先前分享過的、也在上次電話會議中提到的，我們在過去三年把導向廣告主的點擊數提高了 5 倍，但我們的營收當然沒有增加 5 倍。所以，我們談到的重點包括：讓我們的 AI 出價系統與衡量（measurement）系統對齊——我在致詞中也提到——並且更深度整合到更廣泛的一組衡量系統中，讓廣告主能夠衡量這些成效。

And then our AI-driven ad platform, where we just keep making it easier and easier and easier for advertisers to tap into that very high commercial intent on the platform. But I think the -- we see that flywheel continuing to spin both in the user engagement and the fact that users and the engagement there is coming in the way that we want it with high intent, with search, with actionability and that advertisers are leaning in more and more as we give them better ability to go access that commercial intent.

接著是我們以 AI 驅動的廣告平台，我們持續讓廣告主更容易、也更容易、再更容易地切入平台上非常高的商業意圖。但我認為——我們看到這個飛輪持續轉動：一方面是用戶互動，另一方面是用戶與互動正以我們希望的方式發生——高意圖、搜尋、可採取行動——而且當我們讓廣告主更有能力去觸及這些商業意圖時，廣告主也愈來愈積極投入。

Ross Sandler, Barclays.

Ross Sandler，Barclays。

Bill, you mentioned the integrations with first-party measurement and the like AI-optimized bidding as a growth driver at some of these larger accounts. Could you just elaborate on where we are on that progression and what you're seeing thus far?

Bill，你提到與第一方衡量（first-party measurement）的整合，以及類似 AI 最佳化出價，作為一些大型客戶的成長驅動因素。你能否再多談談我們目前在這個進程中的位置，以及到目前為止你看到的情況？

Definitely. Thanks for the question, Ross. So we're still early in the work, but the initial results from the pilot are very encouraging and contributed to our outperformance in Q2. The small number of advertisers that were participating in the pilot, they're seeing meaningful performance improvements, which is giving them greater confidence to increase their spend on Pinterest. And so we're expanding this work to a limited number of additional advertisers in Q3. And by integrating with advertisers' measurement sources of truth, we're able to align our AI bidding systems to optimize for the hyper-specific outcomes each advertiser values most, whether that's customer lifetime value, incremental ROAS or another metric they use internally to evaluate performance.

當然。謝謝你的提問，Ross。我們在這項工作上仍處於早期階段，但試點（pilot）的初步結果非常令人鼓舞，並且對我們第二季的優於預期表現有所貢獻。參與試點的少數廣告主看到了顯著的成效提升，這讓他們更有信心增加在 Pinterest 的投放支出。因此，我們在第三季會把這項工作擴展到數量有限的更多廣告主。透過與廣告主的「衡量真實來源」（measurement sources of truth）整合，我們能讓 AI 出價系統對齊並最佳化每位廣告主最重視的超特定成果，無論是客戶終身價值（customer lifetime value）、增量 ROAS（incremental ROAS），或是他們內部用來評估成效的其他指標。

So the more we're able to line up precisely with what they're looking for, the more we can make sure that AI is delivering their exact outcome. And then so they see more value captured from Pinterest, and they're seeing that across the funnel, upper, mid and lower. And while we started this work with some of our most sophisticated advertisers because they have the data and the measurement infrastructure and the technical resources required to deploy that, our most customized solutions will remain focused on that relatively smaller number with that level of sophistication. But as we move beyond that initial group of advertisers, it's reasonable to expect the full adoption curve will take place across the rest of our advertisers, but it will take some time to play out.

因此，我們愈能精準對齊他們想要的目標，就愈能確保 AI 交付他們所要的確切成果。接著，他們就能從 Pinterest 擷取更多價值，而且他們在整個漏斗——上層、中層與下層——都看到了這點。雖然我們是先從最成熟的廣告主開始這項工作，因為他們具備所需的資料、衡量基礎設施與技術資源來部署，但我們最客製化的解決方案仍會聚焦在那一小群具備這種成熟度的廣告主。不過，當我們走出最初那一群廣告主之後，合理預期完整的採用曲線會在其餘廣告主之間展開，但這需要一些時間才能完全發酵。

You're seeing a few things here like the industry shifting beyond last click attribution to recognize value earlier in the funnel where more than just the last click matters, great marketers have always known that. The ability to take action on that, it's the most sophisticated to do these things first, but we're seeing that take hold through a broader swath of other advertisers through third-party measurement platforms. And so as those capabilities become easier to deploy through third-party measurement systems, we're making sure that we tie in closely to those things.

你會看到幾件事，例如產業正從「最後一次點擊歸因」（last click attribution）轉向，去認可漏斗更前段的價值——不只是最後一次點擊才重要，優秀的行銷人一直都知道這點。而把這件事落地執行的能力，通常是最成熟的業者先做，但我們也看到透過第三方衡量平台，這種做法正在更廣泛的一群其他廣告主中逐步成形。因此，當這些能力透過第三方衡量系統變得更容易部署時，我們也確保能與這些能力緊密串接。

So again, those -- that adoption cycle takes time. Advertisers always have sort of cycles they go through in terms of how they adjust their measurement and they're cautious and very methodical about that. But we absolutely see that happening across the industry, not just with us, but with other platforms. And we think that general trend bodes quite well for us.

所以再說一次——這樣的採用週期需要時間。廣告主在調整衡量方式時，總是會有一套他們的週期；他們也會很謹慎、非常有方法地推進。但我們確實看到這件事正在整個產業發生，不只是我們，其他平台也是如此。而我們認為這個大趨勢對我們相當有利。

So as I mentioned already, the paid clicks to advertisers on Pinterest are up well over 5 times over the last three years. So all of these things help us to capture more of that value that we're already creating. And the feedback that we're getting is encouraging. Simply stated, when advertisers can see the full picture and we can tune our bidding to what matters most to them, they have more confidence and they increase their investment in Pinterest.

如我先前提到，過去三年 Pinterest 上導向廣告主的付費點擊（paid clicks）已增加到遠超過 5 倍。因此，所有這些事情都能幫助我們擷取更多我們已經在創造的價值。而我們收到的回饋也令人鼓舞。簡單來說，當廣告主能看到全貌，而我們也能把出價調校到他們最在意的事情上時，他們就更有信心，並且會增加在 Pinterest 的投資。

John Blackledge, TD Cowen.

John Blackledge，TD Cowen。

As you expand your AI product offerings, how are you managing token-spend usage and AI-compute costs? And what KPIs are you tracking to make sure those choices translate into real productivity gains for your teams?

當你們擴大 AI 產品供給時，你們如何管理 token 支出用量與 AI 運算成本？你們在追蹤哪些 KPI，以確保這些選擇能為團隊帶來真正的生產力提升？

Thanks, John. I can take that one. So in 2026, we've expanded our AI capabilities across our product, both for users and advertisers and also for internal productivity-related use cases. We also just raised our margin outlook for the year. So I think that demonstrates our ability to invest in these capabilities while continuing to improve profitability by managing AI and compute costs efficiently. We've discussed our incremental investment in GPU capacity throughout 2026, which is reflected in our cost of revenue line item. We are already seeing returns through ROAS gains for advertisers and improvements to our user-facing experiences, as Bill just noted.

謝謝你，John。這題我來回答。在 2026 年，我們已把 AI 能力擴展到產品的各個面向，涵蓋用戶與廣告主，也涵蓋內部與生產力相關的使用情境。我們也剛剛上調了今年的利潤率展望。因此，我認為這展現了我們在投資這些能力的同時，也能透過有效管理 AI 與運算成本，持續提升獲利能力。我們在 2026 年持續討論了 GPU 容量的增量投資，這已反映在我們的營收成本（cost of revenue）項目中。我們已經看到回報，包含廣告主的 ROAS 提升，以及用戶端體驗的改善，正如 Bill 剛才提到的。

In general, GPU investments have a relatively short payback period because increased compute capacity can be deployed quickly to enable rapid improvements to our AI models, which translate quickly into business outcomes.

整體而言，GPU 投資的回收期相對較短，因為新增的運算容量可以很快部署，用以快速改善我們的 AI 模型，而這些改善也會很快轉化為業務成果。

As we said, we use a blended model approach that includes our own compact fit-for-purpose proprietary models, open-source models and in very limited cases, select closed third-party models. We evaluate models across the ecosystem constantly and leverage our centralized model routing layer to optimize production traffic for the right balance of quality, cost, latency and reliability. And this allows us to reserve more capable higher-cost models for complex work while using lighter, lower-cost options for routine tasks.

如我們所說，我們採用混合模型策略，包括我們自有、精簡且為特定目的量身打造的專有模型、開源模型，以及在非常有限的情況下，選用少數封閉式的第三方模型。我們會持續評估整個生態系中的模型，並利用我們集中式的模型路由層（centralized model routing layer），在品質、成本、延遲與可靠性之間取得最佳平衡，以最佳化線上（production）流量。這也讓我們能把能力更強、成本更高的模型保留給複雜工作，同時用更輕量、成本更低的選項處理例行任務。

In the second half, we're expanding this model routing infrastructure to even more internal use cases, including tools to increase productivity across engineering, sales and other functions. These costs appear in our OpEx line item. And here, we're tracking adoption, developers throughput, cycle time, incident rates and service uptime to ensure the greater speed does not come at the expense of reliability. For example, in July, weekly pull requests per engineer increased 45% versus last year, while incident rates and service uptime remained relatively consistent. So that gives us confidence that we're improving velocity without compromising reliability.

在下半年，我們會把這套模型路由基礎設施擴展到更多內部使用情境，包括用來提升工程、銷售與其他職能生產力的工具。這些成本會反映在我們的營業費用（OpEx）項目中。在這裡，我們追蹤採用率、開發者產出（developers throughput）、週期時間（cycle time）、事故率（incident rates）與服務上線時間（service uptime），以確保速度提升不會以犧牲可靠性為代價。例如在 7 月，每位工程師每週的 pull request 數量較去年增加 45%，同時事故率與服務上線時間仍大致維持一致。因此，這讓我們有信心：我們正在提升速度（velocity），同時不會損害可靠性。

As AI adoption continues to grow, we expect both AI-related compute and token spend to grow over time. We see that these investments are ROI positive, and that investment growth is planned and included in our outlook that we've provided. So we're scaling AI deliberately, expanding user capabilities, driving advertiser outcomes and improving employee productivity, all while maintaining a disciplined approach to profitability as well.

隨著 AI 採用持續成長，我們預期與 AI 相關的運算與 token 支出也會隨時間增加。我們認為這些投資具有正向 ROI，而投資成長也已規劃並納入我們所提供的展望之中。因此，我們正以審慎的方式擴大 AI 規模：擴展用戶能力、推動廣告主成果、提升員工生產力，同時也維持對獲利能力的紀律化管理。

Justin Patterson, KeyBanc.

Justin Patterson，KeyBanc。

You've talked about broadening your revenue base beyond your largest retailers. Can you update us on that progress, particularly with the SMBs and discuss how Performance+ adoption and the product road map supporting this effort?

你們談到要把營收基礎擴大到不只依賴最大型的零售商。你能否更新一下這方面的進展，特別是針對中小企業（SMBs），並談談 Performance+ 的採用情況以及支援這項工作的產品路線圖？

Certainly. Thanks for the question. So we're really pleased with our continued progress in broadening the revenue base. As Julia noted, in Q2 and UCAN, we saw strong growth outside our largest retailers across mid-market, managed SMB and emerging verticals. And we've got ongoing ad platform improvements, notably Pinterest Performance+, that's been a key driver of this strength across those.

當然。謝謝你的提問。我們對於持續拓展營收基礎所取得的進展感到非常滿意。如 Julia 所提到，在第二季以及 UCAN，我們在最大型零售商之外的客戶中，看見了強勁成長，涵蓋中型市場、受管理的 SMB，以及新興垂直領域。同時我們也持續改善廣告平台，尤其是 Pinterest Performance+，這一直是推動上述各領域強勁表現的關鍵動能。

So Performance+ gives advertisers an automated best practice setup across bidding, budgeting, targeting and creative. And so that's especially valuable for smaller advertisers that don't have large teams. So we're also adding the controls that more sophisticated advertisers need so they can take advantage of these things as well.

Performance+ 為廣告主在出價、預算、受眾定向與創意等方面，提供自動化的最佳實務設定。因此，這對於沒有大型團隊的小型廣告主特別有價值。我們也在加入更成熟的廣告主所需要的控制項，讓他們也能善用這些能力。

As we shared last quarter, approximately 30% of our lower funnel revenue was running through Performance+ campaigns as of Q1. Adopters also grew their lower funnel spend more than twice as fast as non-adopters. So we're really seeing that be effective for our advertisers. And then I'd say more recently, the updates to our ad delivery models drove a 28% improvement in ROAS during testing for SMB advertisers using Performance+ with ROAS bidding. So these improvements continue to help smaller advertisers achieve better outcomes with less manual work.

如同我們在上季分享的，截至第一季，約有 30% 的下漏斗營收是透過 Performance+ 活動產生。採用者的下漏斗支出成長速度也超過未採用者的兩倍。因此我們確實看到這對廣告主很有效。另外更近期的是，我們對廣告投放模型的更新，在測試中讓使用 Performance+ 且採用 ROAS 出價的 SMB 廣告主，其 ROAS 提升了 28%。這些改善持續幫助小型廣告主以更少的人工操作，達成更好的成效。

And as we look ahead, we continue to make our best-performing automation the default for lower funnel campaigns, doing more and more of that while also improving bidding, creative and measurement and adding controls for more complex buyers. So simpler campaign flow targeted for early next year will put best practice setup in place from the very start.

展望未來，我們會持續把表現最佳的自動化功能，作為下漏斗活動的預設選項，並在此基礎上持續改善出價、創意與衡量，同時為更複雜的買方加入控制項。預計明年初推出的更簡化活動流程，將從一開始就把最佳實務設定到位。

So we want to continue to move towards that kind of experience where we're giving really, really simple setup with best practices baked in. And so as we've consistently said, this is a multiyear product and customer adoption cycle. And part of that is managing customers through not only their adoption, but how they think about leveraging these tools. But we continue to release new functionality every quarter, and we see good adoption happening across our user base.

我們希望持續朝向那種體驗前進：提供非常、非常簡單的設定流程，並把最佳實務內建其中。因此如我們一再強調，這是一個跨多年的產品與客戶採用週期。其中一部分工作，是引導客戶不僅完成採用，也理解如何思考並運用這些工具。但我們仍持續每季推出新功能，也看到在整體用戶基礎中有良好的採用進展。

And other platforms that have deployed AI-driven automation suites saw adoption play out over several years. In fact, even some of the earliest to have done this are still working through their adoption cycle. And we expect a similar dynamic for ours as well, where we've created a lot of capability, but there's a lot more in front of us than behind us.

其他已部署 AI 驅動自動化套件的平台，其採用通常會在數年內逐步展開。事實上，即便是最早開始做的其中一些平台，現在仍在推進其採用週期。我們也預期自身會有類似的動態：我們已建立許多能力，但前方要做的仍遠多於已完成的。

And then lastly, just on the go-to-market aspect of this, we're also evolving how we go to market from mid-market and SMB advertisers. We talked about how we're restructuring account coverage, realigning incentives, so we can provide direct support where it can unlock growth and scaled support where that is a better fit. And so broadening our revenue base, including internationally, it's a multiyear journey, and we're still relatively early, but we're building the product foundation, coverage model and go-to-market discipline required to serve a much broader range of advertisers. And again, we believe there's significantly more opportunity ahead of us.

最後，從 go-to-market 的面向來看，我們也在演進面向中型市場與 SMB 廣告主的市場推進方式。我們談到正在重整客戶覆蓋方式、重新對齊激勵機制，以便在能解鎖成長之處提供直接支援，而在更適合的情境下提供可規模化的支援。因此，拓展我們的營收基礎（包含國際市場）是一段跨多年的旅程；我們仍處於相對早期階段，但正在建立產品基礎、覆蓋模型與 go-to-market 紀律，以服務更廣泛的廣告主族群。同樣地，我們相信前方仍有顯著更多的機會。

Nitin Bansal, Bank of America.

Nitin Bansal，美國銀行。

Julia, can you provide some color on the increase in SBC expense this quarter? And as we -- as you continue to invest in AI talent, how should we think about the trajectory of SBC for the second half and next year?

Julia，你能否補充說明本季 SBC 費用增加的原因？另外，隨著你們持續投資 AI 人才，我們應該如何看待下半年以及明年的 SBC 走勢？

Sure. So we don't guide stock-based compensation expense specifically, but it's certainly an area we're focused on. Q2 had some unique factors and as a result, should be the highest quarter for stock-based comp expense in 2026. The increase in Q2 stock-based compensation expense was driven by the annual equity grant cycle for our employees and the lower stock price at the time of grant, which meant we needed to issue more RSUs to remain competitive on compensation. Equity is an important part of how we attract and retain our strong talent while also aligning employees with long-term shareholder value.

當然。我們不會針對以股票為基礎的薪酬（SBC）費用提供具體指引，但這確實是我們關注的領域。第二季有一些特殊因素，因此將會是 2026 年 SBC 費用最高的一季。第二季 SBC 費用的增加，主要是因為員工年度股權授予週期，以及授予時股價較低，意味著為了在薪酬上保持競爭力，我們需要發放更多 RSU。股權是我們吸引並留任優秀人才的重要方式，同時也讓員工與長期股東價值保持一致。

We also saw some timing impact in Q2 as a greater portion of our annual grant expense was recognized this quarter. While stock-based comp will remain elevated in 2026, starting in Q3, we do expect both stock-based compensation expense dollars and the year-over-year growth in stock-based compensation to step down versus Q2. As in recent years, we also expect to be profitable on a full year GAAP net income basis. I'll also note that we've leaned in opportunistically to manage dilution, particularly this year at a moment of dislocation in our stock price. We've repurchased over $2 billion of stock at an average price of $18 per share. As a result, net dilution for Q2 was down 12% year-over-year, leading to growth in Q2 non-GAAP EPS of 30%.

我們也在第二季看到一些時點因素影響，因為年度授予費用有較大比例在本季認列。雖然 2026 年的 SBC 仍將維持在較高水準，但自第三季起，我們預期無論是 SBC 費用金額，或是 SBC 的年增率，都會相較第二季下降。如同近年，我們也預期以全年 GAAP 淨利口徑來看仍將維持獲利。我也要補充，我們今年在股價出現錯位的時點，機會性地加大力度管理稀釋。我們已以每股平均 18 美元的價格回購超過 20 億美元的股票。因此，第二季的淨稀釋年減 12%，帶動第二季 non-GAAP EPS 成長 30%。

So we have a growing, highly profitable and cash-generative business, and we'll continue to be thoughtful in balancing employee retention and incentive alignment with opportunistic capital return and ongoing dilution management going forward, just as we've done in the past.

因此，我們擁有一個持續成長、高度獲利且能產生現金流的業務；未來我們也會一如既往，審慎在員工留任與激勵一致性、機會性資本回饋，以及持續的稀釋管理之間取得平衡。

Jason Helfstein, Oppenheimer.

Jason Helfstein，Oppenheimer。

So first question for Bill. You touched on the capabilities of open-source AI models. Maybe how does Pins leverage that today? And how does that fit into your broader AI strategy?

第一個問題給 Bill。你提到了開源 AI 模型的能力。Pins 目前是如何運用這些能力的？這又如何融入你們更廣泛的 AI 策略？

And then second, just, Julie, a quick one. Is MAU really the best KPI for UCAN? Or is there another metric that you could share with us over time to help us better understand what's going on in the business around engagement?

第二個問題，Julie，簡短問一下。MAU 真的是 UCAN 最好的 KPI 嗎？或者是否有其他指標，你們未來可以逐步與我們分享，以幫助我們更好理解業務在互動參與度方面的狀況？

Thanks for the question, Jason. So on open source, it's open-source AI, it's an area that we're really, really excited about. I've been talking with you all about this since right after the DeepSeek moment, what, 1.5 years or so ago. And as we've discussed before, we take a model-agnostic approach. We use the right model for the task, whether that's one of our in-house compact fit-for-purpose models, an open-source model that's post-trained on our own data or in some very limited cases, a selected closed third-party model.

謝謝你的提問，Jason。關於開源，也就是開源 AI，這是我們非常、非常興奮的領域。自從 DeepSeek 那個時刻之後（大概 1.5 年前左右），我就一直在和各位談這件事。如同我們先前討論過的，我們採取模型不可知（model-agnostic）的做法。我們會針對任務選用最合適的模型：可能是我們自研、精簡且為特定用途打造的模型；也可能是以我們自有資料進行後訓練（post-trained）的開源模型；在極少數情況下，則會選用特定的封閉式第三方模型。

But we're one of the largest practitioners of applied AI at scale at over 640 million users. And we've been at the forefront of leveraging open source. It's become an increasingly important part of our strategy because it gives us three important advantages: cost, customization and control.

我們是規模化應用 AI 的最大實踐者之一，服務超過 6.4 億用戶。我們也一直站在運用開源的前沿。開源已成為我們策略中愈來愈重要的一部分，因為它帶來三個關鍵優勢：成本、客製化與控制力。

So starting with cost. For our use cases, with open models, we're achieving cost per transaction at less than 8% of the cost of comparable closed proprietary models. And that's a substantial advantage. And it becomes even more compelling as open-source models, including those from US-based labs continue to improve. Our early adoption of open source has enabled us to scale AI efficiently to serve our 640 million users while also expanding margins and generating significant free cash flow. But it's the combination of those -- our in-house compact models and leveraging open source really effectively.

先從成本談起。就我們的使用情境而言，採用開放模型後，我們的單筆交易成本不到同等封閉式專有模型成本的 8%。這是非常顯著的優勢。而且隨著開源模型（包含美國實驗室推出的模型）持續進步，這個優勢會更加明顯。我們早期採用開源，使我們能以高效率擴大 AI 規模來服務 6.4 億用戶，同時擴張利潤率並產生可觀的自由現金流。但關鍵在於兩者的結合——我們自研的精簡模型，以及對開源的有效運用。

And I would also just say, with the open source not only is it much lower cost, the fact that we can post-train on our own data makes it more effective. So as you're looking at what's happening in the open-source ecosystem, those models continue to get better and better. It's not just how they do on the benchmarks. It's how are they going to work in your own environment. And when you can post-train on your own data, that makes them much more effective.

我也想補充一點，開源不僅成本低得多，而且我們能用自有資料進行後訓練（post-train），使其更有效。因此，當你觀察開源生態系正在發生的事情時，這些模型會持續變得越來越好。重點不只是它們在基準測試上的表現。而是它們將如何在你自己的環境中運作。當你能用自有資料做後訓練時，就會讓它們更有效。

So then you're looking at something that is more effective than the way you get from the closed models and at less than 1/10 of the cost, that really makes it so that at this point, any CEO that's not taking advantage of open-source models is almost certainly wasting a lot of their shareholders' money. And especially that's true now that you have hyperscalers making it really easy to take advantage of that open source in their secure environment so that you can leverage the open source in your own cloud environment so that it's safe, it's secure. You don't have to worry that it's phoning home. And as others have called out, you don't have to worry that you're giving your company's most important data over to somebody that's going to potentially use it against you.

因此，你看到的是一種比封閉式模型所能提供的方式更有效、而且成本不到十分之一的方案；在這個時間點，任何沒有善用開源模型的執行長，幾乎可以肯定是在浪費股東的大量資金。尤其是現在，超大規模雲端服務商（hyperscalers）讓你能在其安全環境中非常容易地運用開源，讓你可以在自家雲端環境中槓桿開源能力，確保安全、確保防護。你不必擔心它會「回傳資料」（phoning home）。而且如同其他人所指出的，你也不必擔心把公司最重要的資料交給某個可能會用來對付你的人。

So again, we're very, very excited about the progress in open source. That pairs with our own work on our own compact models that serve our use cases. But we're quite encouraged by what we're seeing in the advancement of the ecosystem there. And I think there's going to be a lot more of that to come.

所以再次強調，我們對開源的進展感到非常、非常興奮。這也與我們自身針對使用情境所打造的精簡模型（compact models）相互搭配。但我們對於生態系的進步所呈現的樣貌感到相當鼓舞。我認為接下來還會有更多這樣的發展。

And it's, I think, a great democratizing agent. Most of Silicon Valley, most of the web, most of mobile was built on top of open-source software. US tech needs a thriving open-source ecosystem. And I think with the hyperscalers showing up, with US labs showing up, we're getting exactly that. And again, we've been one of the largest practitioners of it, but we think it's a really great thing for the industry.

而且我認為，這是一個很好的「民主化」推動力量。矽谷的大多數成果、網路的大多數服務、行動端的大多數應用，都是建立在開源軟體之上。美國科技需要一個蓬勃的開源生態系。我認為隨著超大規模雲端服務商加入、隨著美國實驗室加入，我們正在得到我們所需要的那種生態。再次強調，我們一直是其中最大的實踐者之一，我們也認為這對產業而言是非常好的事情。

And then maybe just to take the second part of the question on UCAN and MAUs and engagement. I'd say we always internally look at sort of various baskets of metrics to measure engagement across the platform. We continue to see -- we're adding new users, and we're continuing to maintain strong engagement with those users that we're adding.

接著也許我來回答問題的第二部分，關於 UCAN、MAU 與互動度。我會說，我們內部一向會看各種不同的指標組合（baskets of metrics），用來衡量整個平台的互動情況。我們持續看到——我們正在新增用戶，同時也持續維持我們新增用戶的強勁互動度。

So for example, in UCAN, for instance, searches and boards created continued to grow faster than users in Q2. But I'd say just sort of across the platform as we're looking at strength overall, we're seeing across a number of those basket of metrics, strength that gives us confidence that we're really resonating with users in UCAN in particular, with Gen Z.

例如在 UCAN，以 Q2 來看，搜尋次數與建立的看板（boards）成長速度持續快於用戶數成長。但我想說的是，從整體平台的強度來看，我們在多項指標組合上都看到強勁表現，這讓我們有信心：我們確實與 UCAN 的用戶——特別是 Z 世代——產生了共鳴。

Eric Sheridan, Goldman Sachs.

Eric Sheridan，高盛。

Julia, maybe building on some of the commentary you've given so far with respect to EBITDA margins and given the performance of EBITDA in the quarter, maybe just put a finer point on two questions. One, how are you thinking about what the highest return areas are for incremental growth investments when you look out over the next couple of years? And when you think about flexing the EBITDA margin beyond 2026, how do you think about the scope for both headcount reductions, productivity gains and/or increased growth investments in the ad platform as different variables that could play into how margin evolves going forward?

Julia，或許延續你目前對 EBITDA 利潤率的評論，以及本季 EBITDA 的表現，我想把兩個問題問得更精準一些。第一，當你展望未來幾年時，你如何思考增量成長投資中回報最高的領域？第二，當你思考 2026 年之後進一步調整 EBITDA 利潤率時，你如何看待人力縮減、效率提升，以及／或在廣告平台加大成長投資等不同變數，對未來利潤率演進可能造成的影響範圍？

Sure. Thanks, Eric. I'd say, in general, we are internally just becoming ever more AI native company over time, and we're continuing to move further in that direction. AI is embedded now in how we build every single product, operate the business and allocate resources internally. So the restructuring actions we took in Q1 were really about aligning the organization with our highest priority opportunities and creating capacity to invest in the areas where we see the greatest potential returns, including AI. And so you're seeing that reflected in how we're rebalancing our cost structure a bit in 2026, where we're directing more investment capacity towards GPUs and internal AI tools while being thoughtful and disciplined about where we add headcount.

當然。謝謝你，Eric。我會說，整體而言，我們內部正隨著時間推移，愈來愈成為一家以 AI 為原生（AI native）的公司，並且持續朝這個方向更進一步。AI 現在已經嵌入我們打造每一項產品、營運業務，以及內部資源配置的方式之中。因此，我們在 Q1 採取的重組行動，主要是為了讓組織與我們最高優先順序的機會對齊，並創造產能以投資於我們認為具備最高潛在回報的領域，包括 AI。因此你會看到，這反映在我們於 2026 年對成本結構做了一些再平衡：把更多投資能力導向 GPU 與內部 AI 工具，同時在新增人力方面保持審慎與紀律。

Reported headcount was up 3% year-over-year in Q2, which includes employees who joined Pinterest through the tvScientific acquisition. So if you exclude that acquisition, headcount was down year-over-year in Q2 and is down 6% since the end of 2025. So to be clear, we'll continue to invest in specialized talent where we see a clear and compelling ROI, but an ongoing focus for us is finding the right mix of talent, infrastructure, and AI-enabled productivity tooling to produce durable growth and expand margins over time.

Q2 公布的人力規模年增 3%，其中包含透過 tvScientific 併購加入 Pinterest 的員工。因此若排除該併購，Q2 的人力規模其實是年減，且自 2025 年底以來已下降 6%。所以說清楚一點，我們會持續在看到明確且具說服力的投資報酬（ROI）之處投資專業人才；但我們長期的重點之一，是找到人才、基礎設施與 AI 賦能的生產力工具之間的正確組合，以帶來可持續的成長，並隨時間推移擴大利潤率。

So for 2026, as I noted in my prepared remarks, we are increasing our full year 2026 projection for EBITDA margin to 30% given the strong performance, particularly in revenue that we saw in the first half. That's up from 29% that we had noted previously.

至於 2026 年，如同我在事先準備的發言中提到的，鑑於強勁表現——特別是上半年我們看到的營收表現——我們將 2026 全年 EBITDA 利潤率預估上調至 30%。這高於我們先前提到的 29%。

And as we look forward to 2027 and beyond, I think it's still too early to talk through specifics, but I will reiterate sort of my prior commentary on the appropriate sort of near- to medium-term margins for the business that's sort of consistent with the long-term range we gave at Investor Day a few years ago of 30% to 34%. We still feel like those are the right ones to be tracking, but no specific commentary yet on '27.

而展望 2027 年及之後，我認為現在談具體細節仍為時過早，但我會重申我先前對於公司在近中期適當利潤率水準的看法：這與我們在幾年前投資人日所給出的長期區間 30% 到 34% 一致。我們仍然認為這些是應該追蹤的正確區間，但目前對於 2027 年尚無具體評論。

Michael Morris, Guggenheim Securities.

Michael Morris，Guggenheim Securities。

I wanted to follow up on the Pinterest Assistant comments and hoping you can expand a bit more on the path to monetization of the product. I understand how it further strengthens the user experience. It's very clear. But are all the pieces in place from advertisers to be on that monetization journey? Or do you need more there?

我想追問一下關於 Pinterest Assistant 的評論，也希望你能再多談一些這項產品的變現路徑。我理解它如何進一步強化使用者體驗。這非常清楚。但就廣告主而言，所有推動變現旅程所需的要素都已到位了嗎？還是你們還需要補足更多？

And if I could ask one more on the -- just a follow-up on the comments on the deceleration in non-US international growth that's driven by decisions that have been made by the new leadership team. Can you expand a little bit on what's changing in that go-to-market and when we will see the sort of fruit of that labor?

如果我可以再問一個——關於你提到在非美國的國際成長放緩，這是由新領導團隊所做決策所驅動的。你能否再多說一點，這個市場進入策略（go-to-market）正在改變什麼，以及我們何時會看到這些努力的成果？

Yes. So on the monetization for our assistant, one of the things I talked about on the call is that we've actually woven it throughout our existing app. And so that means it has monetization built in right from the start. So we don't think of this as a build it now and then later monetize type of thing with our assistant capabilities that are woven throughout the app. It's woven throughout the app right alongside of the monetization that is already baked in there. And I would just say, as you look at our journey over the last few years as we made Pinterest an AI-driven shopping assistant, you've seen us do that while having a pretty consistent margin expansion through that time, which has been twofold.

是的。關於我們助理的變現，我在電話會議中提到的一點是：我們其實已經把它編織進我們現有的 App 各處。因此，這代表它從一開始就內建了變現能力。所以我們不會把這視為「先做出來、之後再變現」的事情，因為我們的助理能力是貫穿整個 App 的。它與 App 中既有、已經內建的變現機制並行地貫穿其中。我也想說，回顧過去幾年我們把 Pinterest 打造成 AI 驅動的購物助理的歷程，你已經看到我們在做到這件事的同時，也在那段期間維持相當一致的利潤率擴張，而這是雙重因素所致。

It's been one about us using that AI for highly commercial moments where monetization, again, is sort of baked in right from the start as well as all the cost efficiency that we've talked about with compact models and using open source or open-weight models -- so those are the things that we've been doing to make sure that we've got not only the monetization, but the margin profile that we want from leveraging AI. And so that's exactly what we've been doing for the last several years in the background.

這一直是關於我們在高度商業化的時刻使用 AI，在這些時刻，變現同樣從一開始就已經內建其中，同時也包括我們談到的所有成本效率——例如使用精簡模型，以及採用開源或開放權重模型——因此，這些就是我們一直在做的事情，以確保我們在運用 AI 時，不僅能實現變現，也能達到我們想要的利潤率結構。而這正是我們過去幾年在幕後一直在做的事情。

Now to bring to the foreground, the same type of approach that is sort of woven right in right with the monetization there. And of course, there's always things that we'll do to keep adding to that. We've talked about like our Business Assistant for advertisers, things like that, where we're bringing more assistant capabilities to advertisers, all those things that will continue to enhance that. But we think about monetization from the beginning of our product design.

現在要把同樣類型的方法帶到台前，也就是把這種做法與變現緊密地編織在一起。當然，我們也總是會持續做一些事情來不斷加碼。我們談過例如面向廣告主的 Business Assistant 之類的產品，把更多助理能力帶給廣告主，這些都會持續強化這一點。但我們從產品設計一開始就把變現納入考量。

And then on the second point, in terms of the go-to-market changes the monetization there. So these -- we expect -- we talked about these last quarter. It tends to take -- it's going to be a multi-quarter journey on these things. But the thing I'd point you to is we had work to do in UCAN as well. And when you look at the reacceleration of UCAN, the things that we did to reaccelerate in UCAN particularly around our go-to-market and sales approach, you saw those reflected in Q2. And that's what we're deploying, those similar strategies we're deploying internationally. So while it can take a little bit of time to take hold, you're already seeing that happen for us in UCAN, and that's what gives us really good confidence on that for Europe. And while there's still a little bit more of that for us to do and it can be a multi-quarter journey, we feel quite confident in the approach.

接著談第二點，關於 go-to-market（上市/商業化推進）變更對變現的影響。所以這些——我們預期——我們在上個季度也談過。這通常需要時間——這會是一段跨多個季度的旅程。但我想提醒你注意的是，我們在 UCAN 也有工作要做。當你看 UCAN 的再加速時，你會看到我們為了在 UCAN 重新加速所做的事情，特別是在 go-to-market 與銷售方法上的調整，已在第二季反映出來。而這正是我們正在部署的做法，我們正把類似策略在國際市場推行。因此，雖然需要一點時間才能發酵，但你已經看到它在 UCAN 對我們開始奏效，這也讓我們對歐洲的推進很有信心。而且雖然我們仍有一些工作要做、也可能是一段跨多個季度的旅程，但我們對這個方法相當有信心。

I don't know, Julia, is there anything you want to add to that?

我不知道，Julia，你還想補充什麼嗎？

Yes. I think that's spot on. I think just to add a little more color, our new Head of International Sales is in place now, focused on improving kind of market prioritization and coverage, deepening senior relationships with advertisers and agencies in the region and bringing out just greater consistency overall to measurement and cross-functional support across regions. We're also testing the expansion of third-party demand into Europe as we work to broaden demand. So just a little bit more color there. And again, as Bill said, I think seeing nice results here from the UCAN playbook and looking to extend those to international over time.

是的。我認為這說得很到位。我想再補充一些細節：我們新的國際銷售主管現在已到位，專注於改善市場優先順序與覆蓋策略，深化與該地區廣告主與代理商的高層關係，並在各地區的衡量（measurement）與跨職能支援方面帶來更高的一致性。我們也在測試把第三方需求擴展到歐洲，因為我們正努力拓寬需求來源。以上是一些補充說明。而且如 Bill 所說，我們看到 UCAN 劇本帶來不錯的成果，並希望隨時間推進把這些做法延伸到國際市場。

Shweta Khajuria, Wolfe Research.

Shweta Khajuria，Wolfe Research。

I have one on engagement. What is -- specifically what is driving engagement strength? It's continued to remain strong, particularly among Gen Z. And I guess, what gives you confidence in these trends being durable?

我有一題關於互動參與（engagement）。具體來說，是什麼在推動互動參與的強勁表現？它持續保持強勁，尤其是在 Z 世代之中。我想問的是，你們對這些趨勢具有持久性（durable）的信心來自哪裡？

Yes. Thanks for the question. On the durability of these trends is 12 straight quarters of record high users. And through all 12 of those quarters, it's been the same things that we've been talking about in terms of really making Pinterest a highly shoppable platform with great personalization and with AI powering great recommendations. But not the AI alone, it's the AI powered on human taste and curation, that unique signal that we get.

是的。謝謝你的問題。關於這些趨勢的持久性，我們已連續 12 個季度創下使用者新高。而在這 12 個季度中，我們一直在談的核心因素都相同：真正把 Pinterest 打造成一個高度可購物（shoppable）的平台，具備很好的個人化，並由 AI 驅動提供優秀的推薦。但不只是 AI 本身，而是以人類品味與策展（curation）為基礎所驅動的 AI，也就是我們獲得的那個獨特訊號。

Stated very simply, the AI by itself doesn't have style or taste. People have style and taste. The 640 million people come to our platform really create a flywheel effect that we get really unique signal that we can train our AI or post-train open models on that then let us give better and better results to users.

簡單說，AI 本身沒有風格或品味。人有風格與品味。有 6.4 億人來到我們的平台，形成一個飛輪效應，讓我們獲得非常獨特的訊號；我們可以用這些訊號來訓練我們的 AI，或在開放模型上做後訓練（post-train），進而讓我們能為使用者提供越來越好的結果。

And as I mentioned in my prepared remarks, we see that flywheel continuing to accelerate. And when we think about how differentiated that is, I shared this on prior calls, that when we look at our latest multimodal models and the relevancy of the shopping recommendations we make, we talked about how we saw those outperforming closed models by 30 full percentage points on the relevancy of shopping recommendations. Not anything that it could do on the relevancy of shopping recommendations. So those are things that give us confidence in the durability.

如我在事先準備的發言中提到的，我們看到這個飛輪持續加速。當我們思考這種差異化有多明顯時，我在先前的電話會議也分享過：當我們看最新的多模態模型，以及我們所做的購物推薦的相關性（relevancy）時，我們談到我們看到它在購物推薦相關性上，比封閉模型（closed models）高出整整 30 個百分點。不是說它在購物推薦相關性上能做到任何事。所以，這些就是讓我們對其持久性有信心的原因。

But I'd also just point to what's been happening over the last three years, where we're several years into AI chatbots. Gen Z is more than half our platform, our largest, fastest-growing demographic. They have all almost certainly used chatbots, and they see something different from Pinterest. And with Pinterest, it's not only the visual-first nature. It's how we are bringing that human taste and curation and which is completely unique to our platform and that we don't see happening any place else in the Western world. So there's a lot more for us to do to make that better and better. But we see that with Gen Z. They absolutely are thinking of it that way. And you see that not just in our user growth and engagement.

但我也想指出過去三年發生的事情：我們已經進入 AI 聊天機器人的時代好幾年了。Z 世代占我們平台的一半以上，是我們最大、成長最快的人口族群。他們幾乎肯定都用過聊天機器人，而他們在 Pinterest 上看到的是不同的東西。在 Pinterest，不僅是以視覺為先（visual-first）的特性。更在於我們如何把人類品味與策展帶進來，這是我們平台完全獨有的，我們也沒有在西方世界的其他地方看到同樣的情況。因此，我們還有很多工作要做，讓它變得越來越好。但我們在 Z 世代身上看到了這點：他們確實就是以這種方式在看待它。而你看到的不只是使用者成長與互動參與。

You see it in third-party research as well. We've previously cited, Adobe did a study, independent of us, that 39% of Gen Z now thinks that Pinterest as a first place to go search. And the reason for that is more than 70% of Gen Z sees Pinterest as more personalized. So I think those are places where you can look and see that flywheel really taking effect.

你也能在第三方研究中看到。我們之前引用過 Adobe 的一項研究（與我們無關的獨立研究），指出現在有 39% 的 Z 世代把 Pinterest 視為首先會去搜尋的地方。而其原因是，超過 70% 的 Z 世代認為 Pinterest 更加個人化。所以我認為，這些都是你可以觀察並看到飛輪效應真正開始發揮作用的地方。

And as models commoditize, which they clearly are with open source, we're able to take those capabilities, train on our total unique signal and give really great recommendations to users. We've got great distribution to those users and have become really a beloved platform for those users, which is why, as Julia noted in some of her earlier comments, 85% plus of our users come to our mobile app directly. We're not dependent on others for that traffic. Pinterest is a destination that's beloved by its users. And I think that has proven out to be quite durable as well.

而隨著模型商品化（commoditize）——開源顯然正在推動這一點——我們能夠取得這些能力，用我們完全獨特的訊號來訓練，並為使用者提供非常出色的推薦。我們對這些使用者有很好的觸達與分發（distribution），也已成為他們非常喜愛的平台；這也是為什麼，如 Julia 在先前一些評論中提到的，我們 85% 以上的使用者會直接進入我們的行動 App。我們不依賴他人來帶來這些流量。Pinterest 是一個深受使用者喜愛的目的地。我認為這也已被證明相當具有持久性。

We have reached the end of the Q&A session. I will now turn the call back to Bill Ready for closing remarks.

問答環節到此結束。我現在把電話會議交回給 Bill Ready 作結語。

Thanks again to all of you for joining the call and for your questions. We look forward to keeping this dialogue going, and we hope you all enjoy the rest of your day.

再次感謝各位參加本次電話會議並提出問題。我們期待持續進行這樣的對話，也希望各位今天剩下的時間過得愉快。

This concludes today's call. Thank you for attending. You may now disconnect.

今天的電話會議到此結束。感謝各位出席。您現在可以掛線。