Firefly 补齐音乐与声音创作

Firefly brings audio into the creative studio

原文译览

Source brief

Adobe 宣布，Firefly 的 Generate Music、Generate Speech 与 Generate Sound Effects 已全面开放。创作者可以在同一个创作空间内处理图片、视频、设计、配乐、配音和音效，减少在不同工具之间来回切换。

Adobe says Generate Music, Generate Speech, and Generate Sound Effects are now broadly available in Firefly, placing image, video, design, music, voice, and effects in one creative workspace.

三种声音能力

Three audio capabilities

Generate Music 根据视频时长和情绪生成原创配乐；Generate Speech 可调节声音、节奏与情绪，并可选 ElevenLabs；Generate Sound Effects 则根据画面动作、时间点和能量生成定制音效。Adobe 将其自有音频模型的输出描述为可用于商业成品。

Generate Music fits an original track to a video's duration and mood. Generate Speech controls voice, pacing, and emotion with an ElevenLabs option. Generate Sound Effects follows action and timing. Adobe describes output from its own audio models as commercially safe.

编辑观察

Editorial note

对视频和品牌内容创作者来说，价值不只是多了三个生成按钮，而是声音第一次与画面生产进入同一套连续工作流。使用第三方模型时，仍应分别核对对应条款。

The practical shift is a continuous sound-and-picture workflow. Creators should still review the specific terms attached to any third-party model they choose.

Runway 把商品图做成 UGC 视频

Runway turns product images into UGC video

原文译览

Source brief

Runway 的 Product UGC 工作流以一张出镜人物参考图和一张商品图为必需输入，生成适合社交媒体的竖版 UGC 风格推广视频。商品卖点、语气、具体信息或完整对白脚本均可作为可选控制项。

Runway's Product UGC recipe takes a creator reference and a product image to generate a vertical, social-ready UGC-style promo. Product details, tone, messaging, or a full dialogue script can guide the result.

输出与限制

Output and limits

官方文档支持 4–15 秒输出，比例为 720:1280 或 1080:1920，默认生成音频。Runway 同时要求使用者拥有参考人物肖像的权利和同意；这一步对真实品牌合作不可省略。

The official recipe supports 4–15 second outputs at 720:1280 or 1080:1920 and generates audio by default. Runway explicitly requires rights and consent for any referenced person's likeness.

编辑观察

Editorial note

这类工作流把商品图、脚本和出镜人物直接连成短视频初稿，适合快速测试不同卖点；真正上线前仍要检查商品真实性、品牌规范、肖像授权和字幕。

This connects a product image, message, and presenter into a testable short-form draft. Product accuracy, brand rules, likeness permission, and captions still need human review.

MiniMax H3 开放原生音画视频

MiniMax H3 opens native audio-video generation

原文译览

Source brief

MiniMax H3 是一个全模态视频生成系统，可联合理解文本、图片、视频与音频。官方规格包括 4–15 秒、多种横竖比例、24 FPS、32 kHz 立体声；完整流程最高可生成 2K 视频。

MiniMax H3 jointly understands text, image, video, and audio context. Official specifications include 4–15 second outputs, multiple aspect ratios, 24 FPS, 32 kHz stereo audio, and up to 2K through the full workflow.

参考素材更多

Richer references

全参考模式最多可接收 9 张图片、3 段视频和 3 段音频，全部文件合计不超过 12 个。它支持人物与场景保持、动作和嘴型编辑、音色参考以及原音复用，面向多镜头和品牌一致性任务更有意义。

Reference mode accepts up to nine images, three videos, and three audio clips, capped at twelve files. It supports identity and scene preservation, motion and lip edits, voice reference, and audio reuse.

开放范围

What is open

H3-Base 已按 MiniMax H3 Community License 开放，可通过 SGLang、vLLM、diffusers 和 ComfyUI 部署；用于上下文编排与 2K 再生成的完整模块目前仍主要通过官方 API 提供。

H3-Base is available under the MiniMax H3 Community License with SGLang, vLLM, diffusers, and ComfyUI support. Context orchestration and full 2K regeneration still rely primarily on MiniMax APIs.

Shopify 开放 AI 商品发现与结账

Shopify opens agentic discovery and checkout

原文译览

Source brief

Shopify 宣布 UCP 与 Catalog API 可由开发者自助接入，无需再申请批准。开发者可以从商品搜索、购物车一直构建到结账；Catalog API 把商家商品整理为可供 AI 查询的结构化数据。

Shopify says UCP and Catalog API are now self-serve with no approval required, allowing developers to build from product search through carts and checkout on structured catalog data.

让 AI 看懂商品

Helping AI understand products

Catalog API 新增图片搜索与图文混合搜索，并提供尺码、颜色和预计送达时间等更丰富信息。对商家而言，商品资料不仅服务传统详情页，也决定 AI 能否正确发现、比较和呈现商品。

Catalog API adds image and multimodal search plus richer details such as size, color, and delivery estimates. Product data now affects how AI systems discover, compare, and present an item.

编辑观察

Editorial note

AI 电商的竞争会从“生成一张商品图”继续向“让商品能被 AI 准确理解并完成交易”延伸。结构化资料、库存、配送和结账能力会与创意素材同样重要。

AI commerce is moving beyond generated product images toward accurate discovery and transaction. Structured data, inventory, delivery, and checkout matter as much as creative assets.

OJO 组建 AI 设计团队

OJO puts an AI design team on one canvas

原文译览

Source brief

OJO 是一个 AI Design Agent Team Workspace。官方将它定位为从产品目标出发，把研究、产品判断、界面设计、交互原型和开发交付连接起来的工作区，而不是只根据一句提示生成单张页面。

OJO is an AI Design Agent Team Workspace connecting product goals, research, product decisions, interface design, interactive prototypes, and development handoff rather than generating one screen from one prompt.

从目标到原型

From goal to prototype

用户可以围绕一个产品想法组建由不同设计智能体组成的 Agent Team，并加入针对具体阶段的 Skills。系统可协助明确目标用户、使用场景、需求优先级和产品方向，再探索页面风格、组件系统与视觉方案，生成可点击的 UI 和交互原型。

Users can assemble an Agent Team, add Skills for a specific stage, clarify users and priorities, explore visual directions and component systems, and produce clickable UI and interactive prototypes.

共享上下文与交付

Shared context and handoff

产品推理、设计决策、工作原型和修改记录保留在同一张 Canvas 中。结果可导出为代码、HTML、PPT、PDF 和 PNG，也可以继续进入 Codex、Claude Code 或 Cursor 完成开发。

Product reasoning, design decisions, working prototypes, and revisions remain on the same Canvas. Deliverables can be exported as code, HTML, PPT, PDF, or PNG, or continued in Codex, Claude Code, and Cursor.

当前状态

Current status

OJO 官网目前标明产品处于 private beta，用户需要加入候补名单或使用邀请码。官方页面没有标注明确发布日期，因此本站将 2026.08.25 作为收录日期，而不是产品发布日期。

OJO currently describes the product as a private beta requiring a waitlist or invite code. Its official pages do not state a clear publication date, so 2026.08.25 is the AI World filing date, not a launch date.

编辑观察

Editorial note

OJO 值得关注的不是“又一个生成网页工具”，而是把需求澄清、设计探索、原型迭代和开发交接放进连续工作流。对创作者和小团队来说，真正的效率来自减少上下文丢失和工具切换。

OJO matters less as another page generator than as a continuous path from product framing and design exploration to prototype revision and implementation handoff.

Lovart 生成整套视觉资产

Lovart turns one brief into a visual system

原文译览

Source brief

Lovart 将自己定位为 AI design agent：用户提供需求、参考文件或 Brand Kit 后，系统负责选择模型、生成素材并在同一工作区继续编辑。官方列出的产出包括商品图、社交媒体素材、营销活动套件、品牌视觉和视频。

Lovart positions itself as an AI design agent. Given a brief, reference files, or a Brand Kit, it selects models, generates assets, and keeps editing inside one workspace. Official examples include product shots, social content, campaign kits, brand systems, and video.

一个画布完成生产

Production on one canvas

官方文档称，Lovart 会根据任务在 30 多种图片与视频模型之间进行选择，结果进入无限 Canvas。用户可以继续修改元素、组织不同版本并保持品牌信息；这比每生成一步就下载、换工具和重新上传更适合多素材项目。

Lovart says it routes work across more than 30 image and video models, then places the results on an infinite Canvas for editing, versioning, and brand consistency.

用于电商与营销

For commerce and marketing

对于电商和营销团队，重点不是单张图片，而是从同一个 Brief 延展出商品主图、场景图、社交帖子、广告素材和品牌视频。Brand Kit 可集中管理 Logo、配色、字体与设计规范，帮助同一批内容保持视觉一致。

For commerce and marketing teams, the useful unit is a coordinated set of product images, contextual scenes, social posts, ads, and brand video derived from one brief. Brand Kits centralize logos, colors, fonts, and design rules.

编辑观察

Editorial note

Lovart 更适合需要成套交付和持续修改的任务。是否真正提高效率，应看品牌一致性、修改成本和最终可用素材数量，而不是只看首轮生成效果。

Lovart is most relevant when a project needs coordinated deliverables and repeated revision. Its value should be judged by consistency, revision cost, and usable final assets rather than the first generation alone.

Stampli 用 Codex 将发布工时缩短 68%

Stampli cuts launch hours by 68% with Codex

原文译览

Source brief

Stampli 是一家采购到付款平台。为了发布 Deep Finance 产品，团队需要在固定期限内同步推进产品开发、定位、设计、传播、销售支持与运营，而设计资源和外部承包商当时已有其他任务。

Stampli is a procure-to-pay platform. Its Deep Finance launch required product development, positioning, design, communications, enablement, and operations to move in parallel under a fixed deadline, while design and contractor resources were committed elsewhere.

六周完成发布

A six-week launch

产品从原型演示到公开发布并交付首个版本约用了六周。团队使用 Codex，把持续变化的产品决策转化为可审核的内容，包括七篇博客、发布邮件、网络研讨会及演示文稿、社交与付费素材、新闻稿、产品网页和销售资料。案例称，Codex 还完成了主视觉动画约 90% 的制作工作，最后由承包商处理开场和输出格式。

The product moved from prototype demo to public launch and first shipment in about six weeks. Codex helped turn changing product decisions into review-ready blogs, emails, webinar materials, social and paid creative, PR, web, and sales assets. The case study says it also handled about 90% of the polished hero animation before final contractor work.

Stampli 估算，这套上市与内容生产流程原本需要约 243 个有效工时，使用 Codex 后约为 77 个，节省约 166 小时，相当于生产速度提升至 3.16 倍。所有面向客户的内容仍由人工审核并最终批准。

Stampli estimates that the defined go-to-market and content workflow fell from about 243 active role-hours to roughly 77, saving about 166 hours and making production 3.16 times faster. Human review and final approval remained in place for customer-facing work.

变成日常系统

Built for daily work

这套方法并不只用于发布。过去，营销团队需要分别采访产品经理、查看 Jira 和 GitHub、阅读会议记录，再把信息改写为帮助中心文章、演示文稿和单页资料。现在，基于 GPT 的系统会汇集产品系统与会议记录，并协助持续更新资料；连接内部事实来源的代理也会为网站和社交渠道生产内容。案例中的负责人称，小团队每周的内容产量从少量提升到数百项。

The workflow now supports daily work, not only launches. A GPT-powered system gathers product-system context and meeting notes, helps keep materials current, and supports agents that produce website and social content from company sources of truth.

让信息及时出现

Insights in the moment

ChatGPT Work 也被用来整理会议背景和辅助决策。一次管理层会议中，团队需要查询分散在 HubSpot 等系统里的指标；员工在会议中让 Codex 检索并分析相关数据。案例称，原本可能需要财务规划团队半天完成的报告和模型，这次通过少量输入快速得到可讨论的结果。

ChatGPT Work is also used to prepare meeting context and support decisions. In one executive meeting, Codex retrieved and analyzed metrics across HubSpot and other systems, turning a request that might have taken the FP&A team half a day into an answer available during the discussion.

把时间还给策略

More room for strategy

节省的时间让产品营销团队减少重建背景信息的工作，把更多精力投入产品与公司策略。产品、营销、客户成功、销售和培训团队共同使用 ChatGPT Work 与 Codex，把想法从原型推进到上市；Stampli 认为，这个过去可能需要数月甚至数个季度的过程缩短到了约六周。公司下一步计划把相同方式扩展到更多产品、工作流和上市项目。

The time saved shifted product marketing toward product and company strategy. Teams across product, marketing, customer success, sales, and enablement used ChatGPT Work and Codex to move from prototype to launch in roughly six weeks, and Stampli plans to extend the approach to more products and go-to-market programs.

编辑观察

Editorial note

真正值得复制的不是“用 AI 写得更快”，而是把产品资料、会议记录、决策和信息规范接入同一条可追溯的生产链。速度来自上下文持续更新、明确的事实来源和人工审核，而不是某个万能提示词。

The transferable lesson is not simply faster AI writing. The advantage comes from a traceable production system connecting product context, meeting notes, decisions, messaging rules, sources of truth, and human review.

Grok Build 开放从描述到发布的完整流程

Grok Build opens a describe-to-publish workflow

原文译览

Source brief

xAI 宣布 Grok Build 已面向所有套餐开放。用户可在网页、iOS 与 Android 中描述应用、游戏、网站或看板并生成可运行版本；支持链接发布、访问权限、自定义域名、GitHub 导出、密钥管理和连接器。

xAI says Grok Build is available on every plan across web, iOS, and Android. It can build and publish apps, games, websites, and dashboards, with access controls, custom domains, GitHub export, secrets, and connectors.

从描述到发布

Describe, build, publish

用户在 Grok 对话中描述需求，系统会实时生成可运行版本。发布后，每个项目获得独立的 grok.me 地址，可设置仅自己、持链接者或全网访问，也可以绑定自有域名。系统还会为分享链接生成封面，用户可重新生成、上传或在对话中修改。

Users describe a project in chat and receive a working version. Published projects get a grok.me address with private, link-only, or public access, plus custom-domain support and generated share artwork.

应用可以调用 Grok 的聊天、图片和语音能力，权限按项目控制并可撤销。其他新增能力包括允许他人 Remix、将完整项目导出到 GitHub、把第三方密钥存放在代码之外，以及通过连接器接入业务数据并生成可筛选看板。

Apps can call Grok chat, image, and voice models with per-app access. Additional features include remixing, full GitHub export, external secret storage, and connectors for live business dashboards.

编辑观察

Editorial note

门槛正在从“会不会写代码”转向“能否定义清楚目标并验收结果”。GitHub 导出尤其重要：生成不是终点，后续维护仍需要可移交的项目资产。

The bottleneck shifts from writing code to defining scope and reviewing output. GitHub export matters because generation is not the end; maintainable project assets still need a handoff path.

Shipt 把对话式需求变成可结账购物车

Shipt turns conversations into checkout-ready carts

原文译览

Source brief

Shipt 在 ChatGPT 与 Claude 中推出购物体验。美国用户可用自然语言规划餐食或活动、调整数量与替换商品，并把生成的购物车带到 Shipt 完成结账和当日配送。

Shipt launched shopping experiences in ChatGPT and Claude for U.S. users. Natural-language plans can become editable carts that transfer to Shipt for checkout and same-day delivery.

从计划到购物车

From plan to cart

用户可以描述一周餐食、聚会、返校清单等具体场景，让系统生成推荐并继续对话修改数量、替换商品或调整偏好。Shipt 会员连接账户后，还可使用历史购买和已保存偏好提高个性化程度。

Shoppers can describe meals, events, or school needs, then refine quantities, substitutions, and preferences in conversation. Connected member accounts can use past purchases and saved preferences.

在 ChatGPT 中，用户找到并连接 Shipt 后选择附近零售商，完成购物车后进入 Shipt 结账；在 Claude 中，则通过 Connectors 连接账户并用 Shipt 开始提示。官方信息显示，Shipt 服务覆盖约 80% 的美国人口，但不在阿拉斯加、怀俄明和西弗吉尼亚运营。

ChatGPT users connect Shipt, choose a nearby retailer, and hand the finished cart to Shipt checkout. Claude users connect through Connectors. Shipt says its service reaches about 80% of the U.S. population, excluding Alaska, Wyoming, and West Virginia.

编辑观察

Editorial note

当购买入口变成对话，商品目录、库存、偏好和历史购买记录会成为新的内容基础。电商团队需要把“可被 AI 准确理解”当成商品数据质量目标。

When shopping begins in conversation, catalog, inventory, preferences, and purchase history become the content layer. Commerce teams need product data that AI can interpret accurately.

Google Ads 与 Analytics 增加 AI 分析工具

Google Ads and Analytics add AI analysis tools

原文译览

Source brief

Google 宣布在 Ads 与 Analytics 中加入主页 AI 摘要、文本生成可视化报告、同类企业基准比较，以及可在产品内回答问题并采取行动的 Ask Advisor。

Google announced AI homepage summaries, prompt-built visual reports, peer benchmarks, and Ask Advisor, an in-product agent that can answer questions and support actions across Ads and Analytics.

发现变化

Spot changes

Google Analytics 首页的 AI Overviews 会总结用户上次登录后出现的重要变化，例如季节性销售高峰或流量波动；点击卡片后，可把上下文带入 Ask Advisor 继续分析。Google Ads 首页也会显示针对具体业务的洞察卡，并允许通过提示框生成自定义洞察。

AI Overviews in Analytics summarize important changes since the last visit and pass selected context into Ask Advisor. Google Ads adds business-specific insight cards and prompt-generated custom insights.

Google Ads 的新 Dashboards 可用文本提示生成可视化报告，并自动给出实时摘要；该功能随后也计划进入 Analytics。Analytics 的基准功能则使用匿名的同类企业平均值比较广告表现。官方脚注说明，这批能力目前以英文账户 Beta 形式提供。

Prompt-built Dashboards create visual reports and real-time explanations, with Analytics support planned. Benchmarking compares results with anonymized peers. Google notes that these capabilities are currently in beta for English-language accounts.

编辑观察

Editorial note

分析流程正在从翻找报表变成“提问—解释—行动”。效率提升的前提是团队仍统一指标定义，并核查 AI 对异常变化的解释。

Analysis is moving from dashboard hunting to question, explanation, and action. Teams still need shared metric definitions and verification of AI explanations.

Amazon Ads Agent 增加跨渠道触达建议

Amazon Ads Agent adds omnichannel reach recommendations

原文译览

Source brief

Amazon Ads 在 Ads Planner 中加入 Ads Agent 的跨渠道建议和实时触达预测，覆盖流媒体电视、在线视频、展示和音频。当前范围为美国自助式 Amazon DSP 广告主的认知型活动。

Amazon Ads added Ads Agent recommendations and real-time reach projections in Ads Planner across Streaming TV, Online Video, Display, and Audio. Availability is currently limited to U.S. self-service Amazon DSP awareness campaigns.

对话式媒介规划

Conversational planning

广告主可以直接询问哪些格式组合最适合认知目标，或是否应在流媒体电视和在线视频之外加入音频。Ads Agent 使用实时触达预测，估算第二或第三种格式相较单一格式能增加多少新受众。

Advertisers can ask which format mix best supports awareness or whether Audio should be added to Streaming TV and Online Video. Real-time forecasts estimate the incremental audience from a second or third format.

Amazon 引用其 2025–2026 年美国内部数据称，多格式活动获得的增量触达约为三倍，格式间受众重叠低于 5%；这些数字来自 Amazon 内部研究，不应当作所有广告主都能复制的结果。入口为 Amazon DSP → Ads Planner → Ads Agent，无需额外开通。

Amazon cites U.S. internal data showing three times the incremental reach and less than 5% overlap across formats; these are internal benchmarks, not guaranteed outcomes. The feature requires no additional setup in Ads Planner.

编辑观察

Editorial note

媒介规划正变成更快的对话式比较，但预测不等于结果。预算决策仍应回到受众重叠、频次、增量触达和实际投后数据。

Media planning is becoming a faster conversational comparison, but projections are not outcomes. Budget decisions still require overlap, frequency, incremental reach, and post-campaign evidence.

TikTok Symphony 将广告视频生成延长至 30 秒

TikTok Symphony extends AI ad videos to 30 seconds

原文译览

Source brief

TikTok 将 Dreamina Seedance 2.5 接入 Symphony，视频生成上限由 15 秒增至 30 秒，多模态参考由 9 个增至 50 个，并支持按时间点指导镜头。该能力正在部分市场向部分付费广告主推出，入口包括 Creative Studio、Ads Manager 自动化与 API。

TikTok integrated Dreamina Seedance 2.5 into Symphony, increasing generated video length from 15 to 30 seconds and multimodal references from 9 to 50, with timestamp-level direction. It is rolling out to select paid advertisers in select markets.

更完整的镜头控制

More production control

最多 50 个图片、视频和音频参考可用于限定产品、人物、节奏和场景，并通过时间点指令控制镜头顺序与流动。TikTok 同时宣称画面锐度、真实感、光照、动作、人物一致性，以及对产品语境与比例的理解有所提升。

Up to 50 image, video, and audio references can guide product detail, pacing, and scene flow with timestamp-level direction. TikTok also reports gains in realism, lighting, motion, character consistency, and product context.

安全与开放范围

Safety and access

Symphony 会对提示词、参考输入和生成内容进行多层审核，禁止上传真实人脸，并加入 AI 标签、不可见水印和 C2PA Content Credentials。当前仅向部分市场的部分付费广告主逐步开放，并非所有账户都可使用。

Symphony applies moderation to prompts, references, and outputs, blocks real-face uploads, and uses AI labels, invisible watermarking, and C2PA credentials. Access remains limited to selected paid advertisers in selected markets.

编辑观察

Editorial note

真正的变化不是单纯变长，而是参考素材和时间控制增强后，更接近可执行的广告分镜流程。由于仍是有限开放，使用前要核对账户资格，并保留品牌、肖像与版权审核。

The meaningful shift is not duration alone: more references and time controls make the workflow closer to executable storyboarding. Because access is limited, teams should verify eligibility and keep brand, likeness, and rights review.

Facebook Creator Assistant 把表现数据变成创作建议

Facebook Creator Assistant turns performance data into ideas

原文译览

Source brief

Meta 宣布 Creator Assistant 正在美国、加拿大和印度向 Facebook 创作者推出。助手可结合创作者的内容风格、受众、互动和表现数据解释内容为何有效，并根据趋势提供选题与优化建议。

Meta says Creator Assistant is rolling out to Facebook creators in the U.S., Canada, and India. It uses a creator’s content, audience, engagement, and performance context to explain results and suggest ideas.

解释表现

Explain performance

创作者可以在 Facebook Dashboard 中直接询问某条 Reels 为什么表现更好、受众如何变化，并通过追问继续分析。助手会结合格式、发布时间、受众消费行为和账号表现给出针对该创作者的回答。

Creators can ask why a Reel outperformed others or how their audience changed, then continue with follow-up questions. Responses use the creator's formats, timing, audience behavior, and account performance.

当需要选题时，助手会参考 Facebook 趋势、热门音频、文化事件和高表现内容形式提出角度，并根据创作者更重视增长、互动还是变现逐步调整建议。Meta 同时宣布 Reels AI 翻译将增加阿拉伯语、印尼语、法语、泰语和越南语，并称 Facebook 每周已有超过 5 亿用户观看 AI 翻译视频。

The assistant also draws on Facebook trends, audio, cultural moments, and top-performing formats, adapting to goals such as growth, engagement, or monetization. Meta also announced more Reels translation languages and says over 500 million Facebook users watch AI-translated video weekly.

编辑观察

Editorial note

平台数据开始直接参与选题，但建议天然受限于单一平台的目标和可见数据。适合用来发现假设，不适合替代跨平台判断和创作者自己的表达。

Platform data is moving directly into ideation, but recommendations remain bounded by one platform’s objectives and visibility. Use them to form hypotheses, not replace cross-platform judgment or creator voice.

Meta Business Agent 进入商品推荐与成交环节

Meta Business Agent moves into recommendations and sales

原文译览

Source brief

Meta 正把 Business Agent 扩展到全球不同规模的企业，可在 WhatsApp、Messenger 与 Instagram 回答业务问题、推荐目录商品、预约、筛选线索并衔接人工服务。Meta 还公布可连接 Shopify、Zendesk 和 Shopee 等系统的企业平台。

Meta is expanding Business Agent globally for businesses across WhatsApp, Messenger, and Instagram. It can answer questions, recommend catalog products, book appointments, qualify leads, and hand conversations to staff, with an enterprise platform connecting systems such as Shopify, Zendesk, and Shopee.

从客服到成交

From service to sales

Business Agent 可以使用商家的语气和客户的本地语言回复，在目录中推荐商品、预约、筛选潜在客户、决定何时转交员工并协助成交。Meta 称已有超过 100 万家企业在 WhatsApp 与 Messenger 使用这类代理，并正在扩展到 Instagram。

Business Agent can respond in a business's tone and the customer's language, recommend catalog products, book appointments, qualify leads, define human handoff, and support sales. Meta says more than one million businesses already use an agent on WhatsApp and Messenger.

它还会生成晨间简报，汇总夜间错过的对话并提供线索洞察。面向大型企业的 Meta Business Agent Platform 可连接 Shopify、Zendesk、Shopee 等数百个系统，并提供规则、护栏和效果衡量。当前部分运营辅助能力仍只向少量企业开放，后续订阅方案也尚在推出过程中。

It can also create morning briefings and thread insights. The enterprise platform connects systems such as Shopify, Zendesk, and Shopee with built-in controls, guardrails, and measurement. Some operational features remain limited, with subscription options still rolling out.

编辑观察

Editorial note

客服、营销和成交开始共用一条对话链路。上线前最重要的不是话术，而是目录准确性、人工接管规则、隐私权限和错误回复的纠正机制。

Support, marketing, and conversion are starting to share one conversational path. Before launch, catalog accuracy, human handoff rules, privacy permissions, and correction workflows matter more than clever copy.

YouTube 用 Gemini 加强创作者与受众洞察

YouTube adds Gemini-powered creator and audience insights

原文译览

Source brief

Google 宣布在 Ads Insights Finder 中加入更细的 YouTube 趋势数据，并推出 Content & Creator Insights API。

Google announced more granular YouTube trend data and a Content & Creator Insights API.

四项创作者洞察

Four insight updates

美国版 Ads Insights Finder 增加更细的 YouTube 趋势数据；部分 Brand Pulse 报告指标也会进入 Insights Finder，帮助品牌同时观察付费与自然内容表现。新的 Content & Creator Insights API 面向合作伙伴提供更丰富的创作者与受众信息，用于代理商媒介规划。

Ads Insights Finder adds more granular U.S. YouTube trend data and selected Brand Pulse metrics across paid and organic presence. A new partner API provides richer creator and audience information for agency planning.

Google 还预告，Gemini 将为 YouTube 的 Demand Gen 活动提供创意建议，例如根据表现提示更适合使用哪些视觉元素。官方页面中的趋势图与数据注明为示意内容，并非真实活动结果。

Google also says Gemini will soon provide Demand Gen creative tips, including suggestions about visual choices. The trend examples on the announcement page are explicitly marked as illustrative.

编辑观察

Editorial note

创作者营销正在从凭感觉选人，转向结合内容语境、受众和表现数据。

Creator marketing is moving toward content, audience, and performance evidence.

ChatGPT Ads 扩展到更多市场

ChatGPT Ads expands to more markets

原文译览

Source brief

OpenAI 称 ChatGPT Ads 已在英国、墨西哥、巴西、日本和韩国上线，并强调广告不会改变 ChatGPT 的回答。

OpenAI says ChatGPT Ads launched in the UK, Mexico, Brazil, Japan, and South Korea, while ads do not change answers.

广告如何出现

How ads appear

这项广告测试最初面向美国已登录的成年 Free 与 Go 用户，Plus、Pro、Business、Enterprise 和 Education 套餐不展示广告。广告会标记为 Sponsored 并与正常回答分开；系统根据当前对话主题、历史对话和过去的广告互动进行相关性匹配，但 OpenAI 表示广告不会改变回答内容。

The test began with logged-in adult Free and Go users in the U.S.; paid and organizational tiers were excluded. Sponsored ads are separated from answers and selected for relevance using conversation context and past ad interactions, while OpenAI says answers remain independent.

隐私与控制

Privacy and control

广告主不会获得聊天、历史记录、记忆或个人信息，只能看到浏览量和点击量等汇总数据。未成年人账户以及健康、心理健康、政治等敏感或受监管话题附近不展示广告。用户可以关闭广告、反馈原因、查看为何看到某条广告、删除广告数据并管理个性化设置。

Advertisers receive aggregate performance data rather than chats, history, memories, or personal details. Ads are excluded for minors and near sensitive topics, while users can dismiss ads, give feedback, inspect targeting reasons, delete ad data, and manage personalization.

编辑观察

Editorial note

AI 对话界面正在成为新的发现与决策入口，但规则和衡量仍处早期。

Conversational AI is becoming a discovery and decision surface, while rules and measurement remain early.

Google Ads 增加 AI 广告透明度说明

Google Ads expands AI transparency disclosures

原文译览

Source brief

Google 宣布在 Search、YouTube 和 Discover 增加“How this ad was made”信息，并自动标记使用 Google AI 工具制作的广告。

Google announced a “How this ad was made” panel and labels for ads made with Google AI tools.

披露如何工作

How disclosure works

用户在 Search、YouTube 或 Discover 的广告中点击三点菜单或信息图标，即可在 My Ad Center 查看广告是否由 AI 创建或修改。使用 Google 生成式广告工具制作的内容会自动加入说明；使用外部工具时，广告主可以手动声明。部分地区还可能依据当地要求直接在广告上显示标签。

Users can open My Ad Center from the three-dot or info menu on Search, YouTube, and Discover ads. Google-generated assets receive automatic disclosure, while advertisers can declare AI use from other tools; local rules may also require a visible label on the ad.

Google 还会在自有生成工具输出中嵌入 SynthID 等不可见信号，并继续执行广告主验证、合成或数字修改的选举广告披露，以及禁止误导性广告等既有政策。

Google also embeds signals such as SynthID in its generated outputs and continues advertiser verification, election-ad disclosure, and policies against misleading ads.

编辑观察

Editorial note

生成式创意规模化后，素材来源、修改记录和披露将成为营销流程的一部分。

At scale, provenance, edit records, and disclosure become part of marketing operations.

Asset Studio 将多模态创意与 A/B 测试放进同一流程

Asset Studio combines multimodal creative and A/B testing

原文译览

Source brief

Google 表示 Asset Studio 可基于 Brief、品牌规范、网站和目标生成素材，并计划整合视频和一键 A/B 测试。

Google says Asset Studio can generate from briefs and brand rules, with video and one-click A/B testing planned.

从简报到验证

From brief to test

Asset Studio 会读取营销简报、品牌规范、网站与活动目标，生成不同创意主题和素材类型，用户可继续用自然语言修改。Google 计划通过 Gemini Omni 在同一界面加入视频制作，并用 1-Click A/B Testing 按活动目标比较素材表现。

Asset Studio uses a brief, brand guidelines, website, and goals to generate multiple themes and asset types, with natural-language refinement. Gemini Omni video creation and one-click A/B testing are intended to keep production and validation in one workflow.

官方公告称，这批新能力计划在 2026 年夏季以英文形式向全球逐步推出。因此“已上线”和“即将上线”的部分需要区分，具体账户可用性仍应在 Google Ads 中确认。

Google says the new features are rolling out globally in English during summer 2026. Availability should therefore be checked at the account level rather than treated as universal.

编辑观察

Editorial note

创意生产和效果验证正在合并，团队更需要清楚控制测试变量。

Creative production and validation are converging, increasing the need for controlled variables.

Google 测试 AI Shopping Ads 与原生结账

Google tests AI Shopping ads and native checkout

原文译览

Source brief

Google 宣布测试由 Gemini 生成商品解释的 Shopping 广告、对话式广告和面向 UCP 商家的原生结账。

Google announced tests of Gemini product explainers, conversational ads, and native checkout for UCP merchants.

搜索中的新广告

New ads in Search

Google 正在 AI Mode 测试 Conversational Discovery ads：Gemini 根据用户的具体问题生成更匹配的广告创意；Highlighted Answers 则允许高相关、高质量的赞助结果进入 AI Mode 的推荐列表。两类广告都会包含独立的 AI 商品说明，并继续标记为 Sponsored。

In AI Mode, Conversational Discovery ads tailor creative to a specific question, while Highlighted Answers can place high-quality sponsored options inside recommendation lists. Both include an independent AI explainer and remain labeled Sponsored.

在普通 Search 中，AI-powered Shopping ads 会挑选相关商品并解释其适配原因；Business Agent for Leads 则把品牌问答代理放进广告，用对话替代静态线索表单。Direct Offers 还将增加促销组合、面向 Universal Commerce Protocol 商家的原生结账，以及 Booking、Expedia 等旅行优惠。官方把这些功能描述为测试、试点或即将推出，并非全部已普遍开放。

Search is also testing AI-powered Shopping explainers and an in-ad lead agent. Direct Offers adds promotion bundles, native checkout for Universal Commerce Protocol merchants, and travel offers. Google describes these as tests, pilots, or upcoming features rather than universal availability.

编辑观察

Editorial note

商品资料正在成为 AI 解释商品适配度与推进购买的事实基础。

Product data is becoming the factual layer for AI explanations and purchase flows.