Siteimprove Deploys AI Agents to Automate Web Accessibility Compliance

Siteimprove 部署 AI 智能体实现网页无障碍合规自动化 企业数字内容必须在源代码层面达到合规要求，以消除发布后代价高昂的修复成本，并在 AI 问答引擎中保持搜索可见度。 在本次会议专题中，Siteimprove 高管探讨了企业面临的内容激增与监管要求（如 ADA 和 EAA），以及从传统搜索向 AI 问答引擎的转变。他们介绍了 Siteimprove 的智能体内容智能平台（Agentic Content Intelligence Platform），该平台通过模型上下文协议（MCP）将无障碍合规、SEO 及品牌治理“左移”至 Figma 和 Lovable 等创作工具中。通过现场演示“识别、行动与验证”（Identify, Act, and Prove）这一运营飞轮，他们展示了智能体协同与对话式分析如何赋能品牌自动修复代码、优化 AI 搜索呈现并验证投资回报率。 2026 年的企业领导者正从孤立的 AI 实验转向构建与核心业务战略相契合的自主、目标驱动型 AI 系统。 (0:08) 数字内容的快速增长在《美国残疾人法》（ADA）和《欧洲无障碍法案》（EAA）等法律框架下面临重大合规挑战。 (0:58) 搜索行为正从 Google 等传统搜索引擎根本性地转向 ChatGPT、Grok、Claude 和 Perplexity 等 AI 问答引擎。 (0:58) 企业对 AI 编程工具、低代码平台和大语言模型（LLM）的依赖，往往会大规模生成不合规且偏离品牌规范的数字内容。 (3:04) 为避免发布后付出高昂代价进行修复，企业需要在设计和开发期间检测并自动修复内容、无障碍合规及 SEO 问题，以此实现“左移”。 (3:04)

Siteimprove executives argue that the explosion of generative AI content and the transition from traditional search to conversational answer engines require enterprises to shift compliance and optimization directly into the creation workflow. Regulatory Compliance and the Answer Engine Shift (0:58) According to CEO Nayaki Nayyar, organizations face twin pressures in the current digital landscape: strict accessibility regulations and changing search dynamics. Legally binding frameworks, such as the Americans with Disabilities Act (ADA) in North America and the European Accessibility Act (EAA) in Europe, require external digital properties above specific thresholds to be accessible. Simultaneously, user search traffic is moving away from traditional Google queries toward AI answer engines including ChatGPT, Grok, Claude, and Perplexity. Nayyar notes that this shift exposes brands to dual risks: losing digital visibility or suffering brand dilution when AI systems misrepresent products and corporate messaging. The Operational Urgency to 'Shift Left' (3:04) Enterprises scaling content production with AI coding and low-code design tools frequently produce assets that violate brand guidelines and accessibility standards. Siteimprove's Customer Advisory Board—including representatives from McDonald's, Harvard, Yale, and major financial institutions—identified post-publication remediation as an unsustainable drain on budget, labor, and compliance resources. Nayyar argues that traditional reactive workflows (publish, scan, and manually fix) must be replaced by a 'shift left' model. Under this paradigm, automated checks intercept and remediate accessibility, SEO, and brand issues directly within developer and design environments like Figma, Lovable, Claude, and VS Code prior to production deployment. Agentic Architecture and Agent-to-Agent MCP Workflows (12:31) Siteimprove organizes its agentic architecture across four pillars: Accessibility, Analytics, Search (SEO and Answer Engine Optimization), and Content Strategy, all anchored by a central system of record. Chief Product Officer Kalvin Brite outlines an operational flywheel based on three actions: identify compliance and search gaps, act via automated or human remediations, and prove performance impact against established baselines. To automate remediation during the development phase, Siteimprove deployed a Model Context Protocol (MCP) server. In a live demonstration integrating with the AI web builder Lovable, Siteimprove's accessibility agent evaluated an application under construction, flagged WCAG and EAA violations, and collaborated directly with Lovable's AI developer agent. The agents resolved the accessibility defects in the source code without requiring manual developer intervention. AEO Tracking, Conversational Proof, and Structured Data (17:24) To monitor visibility across conversational models like ChatGPT, Gemini, and Perplexity, Siteimprove tracks metrics such as total brand mentions, topic authority against competitors, and direct web citations. Rather than relying on traditional analytics dashboards, a conversational interface allows enterprise users to query raw platform performance data, traffic trends, and compliance metrics to verify return on investment. Brite emphasizes that accessibility compliance and AI search visibility rely on the same technical foundations. Semantic code structures and proper tags mandated by accessibility regulations provide the exact machine-readable framework that LLMs, web crawlers, and AI search engines need to parse, comprehend, and cite enterprise web content accurately. Kalvin Brite concludes that enterprise digital content must be 'born compliant and born performant' to avoid long-term operational liabilities. Organizations must integrate automated prevention into authoring tools, maintain continuous validation across dynamic runtime environments, and systematically prove ROI through unified systems of record.

Siteimprove 首席执行官 Nayaki Nayyar 与首席产品官 Kalvin Brite 提出，面对海量 AI 内容爆发、严格法规约束以及搜索行为向大模型转移的现状，企业必须将传统 SaaS 治理转向 Agentic AI 架构，实现数字内容在创作源头即具备合规性与高性能。 内容激增与搜索范式转变的双重合规风险 (0:58) Nayaki Nayyar 指出，AI 生成内容的爆发式增长给大型企业带来了严峻的合规与运营挑战。北美地区的《美国残疾人法案》（ADA）以及欧洲即将实施的《欧洲无障碍法案》（EAA）等严格法律框架，要求外部数字内容必须具备完全的无障碍访问性，这对于达到特定规模的企业而言已成为不可规避的法定红线。与此同时，用户的搜索习惯正从以 Google 为主的传统引擎快速转向 ChatGPT、Grok、Claude 及 Perplexity 等 AI 驱动的问答工具。品牌正面临双重威胁：若无法被主流大模型收录和引用，将直接失去数字流量入口；而一旦模型对品牌的产品、服务或核心价值产生偏差或错误表述，将直接损害品牌声誉与商业利益。 企业规模治理困境与“左移”架构转向 (3:04) Siteimprove 拥有超过 20 年行业经验，服务包括壳牌（Shell）、约翰迪尔（John Deere）、卡特彼勒（Caterpillar）、麦当劳、耶鲁大学及哈佛大学在内的逾 5,500 家全球企业客户。Nayaki 透露，在一次汇聚顶尖学术与金融机构代表的芝加哥客户顾问委员会上，客户普遍反映内部各团队正广泛使用低代码平台、v0/v1 编程工具与大模型批量生成内容，导致大量违规且脱离品牌规范的页面直接上线。传统的“发布后扫描并手动修复”流程耗费巨大人力、时间和预算，并让企业长期暴露于法律诉讼风险中。为此，Siteimprove 提出“左移”（Shift Left）策略，主张在 Figma、Lovable、Claude 或 VS Code 等设计与开发工具中，通过智能体在创作的源头阶段实时拦截并自动修复无障碍、品牌规范及 SEO 问题，确保内容自诞生即合规。 智能体飞轮：识别、执行与验证的一体化中枢 (9:37) Kalvin Brite 详细介绍了平台以“记录系统”（System of Record）为核心驱动的业务飞轮，该飞轮由三个核心步骤构成：识别（Identify）、执行（Act）与验证（Prove）。在识别维度，高级 AEO（答案引擎优化）分析面板能够超越静态 SEO 报表，实时监控 ChatGPT、Gemini 和 Perplexity 等主流 AI 模型中的品牌声量份额，系统化追踪品牌提及量（Mentions）以及具体网页内容被模型直接引用的频次（Citations），并针对竞品动态列出高商业回报的优化建议。记录系统统一存储所有发现的缺陷、修改记录以及影响历史，为后续智能体提供精确的上下文支撑。 MCP 协议驱动的跨智能体协作与代码级自愈 (12:31) 在执行层面，Siteimprove 实现了人机协同创作与代码级自动修复的双重落地。面向创作者的内容智能体（Content Agent）通过浏览器插件直接嵌入 Google Docs 和 Microsoft Word，结合历史数据协助撰写文本，弥合搜索缺口并预测上线后的表现评分。而在工程端，Siteimprove 部署了基于模型上下文协议（Model Context Protocol, MCP）的专用服务器，使无障碍与性能分析智能体能够直接与 Lovable 等外部 AI 网页构建工具无缝对接。现场演示展示了两个 AI 之间的跨智能体协作：Siteimprove 智能体识别出新生成应用中的 WCAG 和 EAA 无障碍违规项后，Lovable 的开发者智能体直接在源码底层即时修正代码，全过程无需人工工程师介入。 对话式分析验证 ROI 与底层语义统一逻辑 (16:00) 在飞轮的最终验证阶段，平台内置了对话式数据分析智能体，团队成员可直接使用自然语言询问流量趋势、合规评分变化或区域细分表现。该系统基于受信任的企业记录系统，生成带有数据源引用、动态图表及原始数据表格的结构化回答，以证明合规与内容优化对业务增长的实际 ROI。Kalvin 总结指出，无障碍规范所要求的语义化代码和高结构化数据，恰好与大模型抓取、解析并精确引用品牌内容的技术需求完全重合；因此，让内容在生成之初即兼具合规性与高性能，不仅消除了合规隐患，也是构建 AI 时代品牌声量壁垒的底层基石。 Siteimprove 认为，未来的数字体验治理必须坚持“自诞生即合规且高效”的原则，将无障碍建设视作构建 AI 搜索权威的基础设施；然而，随着 AI 问答平台算法的高频迭代，企业如何在大模型生态中持续稳定地维持声量份额与引用准确率，仍是需要长效验证的实践课题。

## Background 该视频由 Ai4 频道于 2026 年 8 月 19 日在 YouTube 平台发布。内容由 Siteimprove 首席执行官 Nayaki Nayyar 与首席产品官 Kalvin Brite 共同主讲，官方简介指出 2026 年 AI 实验时代已宣告结束，演讲围绕企业如何构建自主目标驱动的智能体系统、搭建信任架构与消除战略执行鸿沟展开深度探讨。 ## TL;DR 面对内容爆炸、严苛法规（如 ADA、EAA）以及搜索流量向 ChatGPT 和 Perplexity 等答案引擎迁移的双重冲击，Siteimprove 宣布推出“智能体内容智能平台”。平台秉持“向左移动”（Shift Left）战略，将治理前置到开发源头，通过无障碍、对话分析、搜索/AEO 及内容策略四大智能体构建运营防线。依托“识别、行动、验证”三步飞轮与 MCP（模型上下文协议）实现的 Agent-to-Agent 自主协作，企业能够在代码与内容生成初期自动修复缺陷，践行“内容必须生而合规、生而高效”的全新范式。 ## Atmosphere & style 整场演讲洋溢着干练、前瞻且极具务实精神的企业级商业氛围。两位高管配合默契，从麦当劳、哈佛等顶级客户的真实治理焦虑切入，无缝衔接到底层架构拆解与生动的现场演示。叙事节奏明快紧凑、逻辑严密，既有对宏观 AI 范式转移的深刻洞察，又有落地的技术解法，传递出对下一代智能体治理格局的笃定与自信。 ## Key takeaways - [00:08](https://www.youtube.com/watch?v=2rGbkmlRoxs&t=8s) 企业正面临由 AI 驱动的内容大爆发，伴随北美 ADA 与欧洲 EAA 等法规的严格出台，数字内容的无障碍与合规压力急剧上升。 - [00:08](https://www.youtube.com/watch?v=2rGbkmlRoxs&t=8s) 搜索流量正快速从传统 Google 检索转向 ChatGPT、Grok、Perplexity 等 AI 答案引擎，促使企业核心诉求从传统 SEO 转向答案引擎优化（AEO）。 - [04:42](https://www.youtube.com/watch?v=2rGbkmlRoxs&t=282s) 在 AI 辅助与“氛围编码”（Vibe Coding）普及下，员工生成海量脱离规范的内容，企业亟需将治理与修复能力深度嵌入创作工具链中。 - [04:42](https://www.youtube.com/watch?v=2rGbkmlRoxs&t=282s) Siteimprove 坚决推进“向左移动”（Shift Left）战略，主张在 Figma、VS Code、Lovable 等创作源头实时拦截与自动修复问题，终结事后修补的高昂成本。 - [04:42](https://www.youtube.com/watch?v=2rGbkmlRoxs&t=282s) 智能体内容智能平台依托四大核心支柱：无障碍智能体、对话式分析智能体、搜索与 AEO 智能体，以及内容策略智能体。 - [09:37](https://www.youtube.com/watch?v=2rGbkmlRoxs&t=577s) 平台构建了由“识别问题（Identify）、采取行动（Act）、成效验证（Prove）”组成的智能体飞轮闭环，并以统一记录系统（System of Record）为中枢。 - [09:37](https://www.youtube.com/watch?v=2rGbkmlRoxs&t=577s) 高级 AEO 洞察仪表盘能够跨主流 LLM 实时追踪品牌的声量份额（Share of Voice）、提及次数（Mentions）与事实引用次数（Citations）。 - [12:25](https://www.youtube.com/watch?v=2rGbkmlRoxs&t=745s) 嵌入式 AEO 内容智能体支持常驻 Google Docs 等编写环境，基于 AI 搜索差距数据实时协同创作并进行答案引擎表现评分。 - [12:25](https://www.youtube.com/watch?v=2rGbkmlRoxs&t=745s) 通过标准化模型上下文协议（MCP），Siteimprove 实现了与 Lovable 等外部 AI 构建工具的 Agent-to-Agent 自主协作，在代码生成阶段自动注入合规能力。 - [16:00](https://www.youtube.com/watch?v=2rGbkmlRoxs&t=960s) 对话式分析模块允许用户通过自然语言直接洞察业务指标，所有回答均严格锚定于第一方真实数据并支持原始溯源，杜绝数据幻觉。 - [16:00](https://www.youtube.com/watch?v=2rGbkmlRoxs&t=960s) 未来数字内容的核心治理哲学是“生而合规、生而高效”（Born compliant, born performant），否则海量内容将直接沦为企业的负债。 - [16:00](https://www.youtube.com/watch?v=2rGbkmlRoxs&t=960s) 人类无障碍阅读与 AI 大模型解析在底层架构上完全同构，良好的无障碍结构化数据是让 AI 引擎精准引用品牌内容的基石。 ## Detailed walkthrough ### [00:08](https://www.youtube.com/watch?v=2rGbkmlRoxs&t=8s) Siteimprove 的智能体转型与内容爆发时代的双重合规挑战 Siteimprove 首席执行官 Nayaki Nayyar 与首席产品官 Kalvin Brite 共同登台，开门见山地阐述了公司如何从传统的核心 SaaS 业务全面迈入“智能体时代”（Agentic Era）。Nayaki 指出，当今企业正身处一个由 AI 驱动、内容呈爆炸式无限增长的世界——网页、社交媒体、移动端应用以及企业文档的海量涌现，给大型企业带来了前所未有的治理危机。其中第一大严峻挑战是法律与合规压力：在北美有《美国残疾人法案》（ADA），在欧洲则有《欧洲可访问性法案》（EAA），这些法规硬性要求达到一定规模的企业必须确保其所有面向公众的数字化外部内容完全具备无障碍访问性（Accessibility）与合规性。 与此同时，企业面临的第二大结构性冲击是搜索形态的范式转移。以往用户寻找答案习惯性地依赖 Google 等传统搜索引擎，但在当今时代，巨量搜索流量已迅速转向 ChatGPT、Grok、Perplexity 等 AI 驱动的生成式答案引擎。对于拥有 20 多年历史、服务超过 5,500 家客户（涵盖壳牌 Shell、约翰迪尔 John Deere、卡特彼勒 Caterpillar 等全球 2000 强与财富 500 强企业）的 Siteimprove 而言，所有大型品牌都在焦虑同一个问题：品牌不仅要在 AI 搜索中抢占可见度，更要确保其产品与服务在生成式回答中被“精准无误地表述”，避免因模型幻觉或错误引用而造成严重的商业声誉损失。 - **合规法规驱动**：北美 ADA 与欧洲 EAA 强制要求企业面向公众的数字资产必须满足无障碍与合规标准。 - **搜索流量迁移**：用户注意力与流量大规模从传统 Google 检索转向 ChatGPT、Grok、Perplexity 等 AI 驱动的答案引擎。 - **市场地位与客户规模**：Siteimprove 获 Gartner 与 Forrester Wave 权威认证，服务全球超 5,500 家跨行业标杆客户（金融、医疗、公共事业、重工业制造等）。 --- ### [04:42](https://www.youtube.com/watch?v=2rGbkmlRoxs&t=282s) 从事后修补到“向左移动”：智能体内容智能平台的四大支柱 随着氛围编码（Vibe Coding）与 AI 辅助构建工具的普及，“人人都是开发者”的时代已经到来，但这也在企业内部引发了合规与品牌失控的连锁反应。Nayaki 分享了他们在芝加哥召开的客户咨询委员会（CAB）上的见闻：麦当劳（McDonald's）、耶鲁大学（Yale）、哈佛大学（Harvard）以及多家大型金融机构的高管们围坐在一起，异口同声地表达了对各类 AI 编码工具泛滥的担忧。员工利用这些工具生成了海量脱离品牌规范、缺乏无障碍支持且未优化的内容与应用。面对这一痛点，客户们强烈呼吁：“如果你能帮我们解决一个问题，请务必找到与这些 AI 驱动工具深度集成的方法，在流程早期就检测并修复这些问题。” 这正是 Siteimprove 坚决推行的“向左移动”（Shift Left）战略。过去，企业往往在网页已经发布、成千上万个页面上线后再去扫描和修复问题，这种“事后弥补”消耗了极为庞大的人力与时间成本，导致团队疲于应付监管截止日期。Siteimprove 主张将检测与修复机制直接前置到资产创造的源头：无论开发者使用的是 Lovable、Claude 还是 VS Code，抑或设计师在 Figma 中构建视觉稿，智能体都能实时发现缺陷并自动完成修复。为此，Siteimprove 推出了“智能体内容智能平台”（Agentic Content Intelligence Platform），Nayaki 幽默地将其形容为“像芝麻街一样简单明了”（Sesame Street simple），并通过四大支柱提供完整赋能： ```mermaid flowchart TD accTitle: 智能体内容智能平台四大支柱 Platform["Siteimprove 智能体内容智能平台"] P1["1. 无障碍智能体<br/>(Accessibility Agents)<br/>早期检测与自动合规修复"] P2["2. 对话式分析智能体<br/>(Analytics Agents)<br/>自然语言生成报表与洞察"] P3["3. 搜索与 AEO 智能体<br/>(Search & AEO Agents)<br/>优化 AI 引擎能见度与品牌呈现"] P4["4. 内容策略智能体<br/>(Content Strategy Agents)<br/>更新内容大纲与规范修正"] Platform --> P1 Platform --> P2 Platform --> P3 Platform --> P4 ``` - **无障碍智能体（Accessibility）**：在创作初期自动捕获无障碍违规项并执行自动修复。 - **对话式分析智能体（Analytics）**：告别繁琐的手动报表搭建，通过自然语言提问即可瞬时生成多维仪表盘与深度洞见。 - **搜索与 AEO 智能体（Search & AEO）**：结合传统 SEO 与前沿答案引擎优化（AEO），确保品牌在各主流 LLM 答案中维持高可见度与准确表达。 - **内容策略智能体（Content Strategy）**：当发现品牌或产品描述失真时，自动纠正并更新内容大纲（Content Briefs）与策略文档。 --- ### [09:37](https://www.youtube.com/watch?v=2rGbkmlRoxs&t=577s) 智能体飞轮闭环：识别、行动与成效验证的三步框架 首席产品官 Kalvin Brite 接过演讲，系统展示了支撑这一平台运转的底层产品飞轮。Kalvin 解释道，AI 智能体的嵌入将复杂的运营流程高度提炼为围绕飞轮循环展开的三个核心问题：“识别问题”（Identify）、“采取行动”（Act）以及“成效验证”（Prove）。飞轮运转的中枢是一个统一的记录系统（System of Record），平台将所有的诊断发现、变更历史及最终业务影响全部汇聚在此，使智能体能够在统一的上下文数据中持续进化，代表品牌做出更精准的决策。 ```mermaid flowchart LR accTitle: 智能体工作流飞轮 subgraph Flywheel["智能体运营飞轮"] direction TB A["1. 识别 (Identify)<br/>洞察 AEO 表现、声量份额与违规缺陷"] B["2. 行动 (Act)<br/>人机协同创作 + MCP 智能体间协作预防"] C["3. 验证 (Prove)<br/>对话式分析追踪业务回报与排名提升"] A --> B B --> C C --> A end SoR[("统一记录系统<br/>System of Record<br/>沉积历史数据、上下文与业务影响")] Flywheel <--> SoR ``` 在“识别”阶段，Kalvin 现场演示了平台打造的“高级 AEO 洞察仪表盘”（Advanced AEO Insights）。这绝非传统的 SEO 报表，而是实时追踪品牌在 ChatGPT、Gemini、Perplexity 等主流大模型中的声量份额（Share of Voice）与主题权威度（Topic Authority），并与竞争对手进行横向动态对标。该系统重点提炼出两项核心指标：其一是“提及次数”（Mentions，即品牌在 AI 生成对话中出现的频次），其二是“引用次数”（Citations，即品牌官方内容被 LLM 引述为事实来源的频次）。不仅如此，系统还会深度测算每个优化机会背后所关联的潜在商业营收与影响力，帮助营销人员快速锁定最高回报率的改进方向。 - **工作流三步闭环**：识别现状漏洞（Identify） $\rightarrow$ 人工或智能体自主执行纠偏（Act） $\rightarrow$ 量化业务与可见度提升（Prove）。 - **AEO 核心衡量维度**：涵盖跨主流模型声量份额（Share of Voice）、提及频次（Mentions）、引用频次（Citations）以及主题权威度趋势。 - **商业机会量化**：自动将内容调整与潜在收入增长和流量影响进行绑定分析，明确优先级。 --- ### [12:25](https://www.youtube.com/watch?v=2rGbkmlRoxs&t=745s) 赋能与预防：AEO 内容智能体与基于 MCP 的 Agent-to-Agent 协作 在明确改进机会后，系统进入“采取行动”阶段。Kalvin 指出，行动包含“高效赋能内容生产”与“源头杜绝缺陷产生”两方面。针对内容创作，Siteimprove 推出了作为浏览器插件运行的内容智能体（Content Agent），可直接嵌入 Google Docs、Microsoft Word 等日常文档编辑流中。该智能体读取平台沉淀的 AEO 差距数据，协同写作者补全内容空白，并在编写过程中实时对文本进行模拟评分，预测其在各大答案引擎中的表现。 在源头预防方面，Siteimprove 实现了过去二十年技术架构的最大飞跃——通过发布自主的模型上下文协议（Model Context Protocol, MCP）服务器，将其平台智能体无缝接入外部 AI 构建工具。Kalvin 现场演示了与全栈 AI 开发工具 Lovable 的集成场景：当开发者在 Lovable 中生成网站时，Lovable 智能体与 Siteimprove 智能体通过 MCP 实现“Agent-to-Agent”（智能体间）的自主通信。Siteimprove 智能体主动扫描生成的代码和 UI，评估其距离 EAA 或 WCAG（Web 内容无障碍指南）合规目标的差距，随后直接指导 Lovable 智能体在代码层完成无障碍优化，使得网页在诞生之初便天然合规，彻底摆脱了人工审核与上线后返工的泥潭。 - **嵌入式内容编写智能体**：以浏览器插件形态常驻编辑流，基于 AI 搜索缺口指导文案创作并实时打分。 - **MCP 开放架构**：部署标准化 MCP Server，将无障碍、SEO 及治理能力暴露给外部 AI 智能体与构建平台。 - **Agent-to-Agent 协同实操**：在 Lovable 环境下，智能体之间自主握手，实现代码生成阶段的实时扫描与自动合规重构。 --- ### [16:00](https://www.youtube.com/watch?v=2rGbkmlRoxs&t=960s) 对话式分析验证与“生而合规、生而高效”的未来范式 在闭环的最后一环“成效验证”中，Kalvin 演示了内置于平台的对话式分析模块（Conversational Analytics）。由于所有操作、流量表现及品牌影响数据都统一沉淀在平台记录系统中，用户无需编写复杂的 SQL 或配置透视表，只需使用自然语言提问（如查询特定地区或特定周期的性能变化），系统便能像 Claude 或 ChatGPT 一样瞬间返回清晰的清单或可视化图表。更为关键的是，这些回答严格锚定在企业真实的第一方数据源上，并提供可溯源的原始数据引用，彻底消除了数据幻觉。 Kalvin 在演讲尾声提出了 Siteimprove 对未来数字内容的终极主张：“内容必须生而合规，生而高效，否则它将沦为企业的负债”（Content must be born compliant and must be born performant. Otherwise, it's going to be a liability）。他总结道，“无障碍”与“高性能”绝非二选一的权衡，而是相辅相成的统一体——因为为人类无障碍阅读打下的结构化数据基础，恰恰也是让 LLM、爬虫和 AI 答案引擎能够准确解析、理解并引用品牌内容的底层基石。演讲最后，Kalvin 热情邀请与会者前往 318 号展位进一步交流探讨。 - **基于真实数据的对话式分析**：支持通过自然语言自由探索任意颗粒度的数据指标，结果完全基于第一方真实记录。 - **核心治理哲学**：“生而合规、生而高效”，构建从“创作时预防”（Prevention at creation）到“生产环境验证”（Verify in production）的全生命周期防线。 - **技术共性统一**：无障碍合规基础设施与 AI 大模型读取解析基础设施在底层架构上完全同构。 ## Notable quotes > [04:42](https://www.youtube.com/watch?v=2rGbkmlRoxs&t=282s) 如果你能帮我们解决一个问题，请务必找到与这些 AI 驱动工具深度集成的方法，在流程早期就检测并修复这些问题。 — Nayaki Nayyar > [04:42](https://www.youtube.com/watch?v=2rGbkmlRoxs&t=282s) 像芝麻街一样简单明了 — Nayaki Nayyar > [09:37](https://www.youtube.com/watch?v=2rGbkmlRoxs&t=577s) 识别问题（Identify）、采取行动（Act）以及成效验证（Prove）。 — Kalvin Brite > [16:00](https://www.youtube.com/watch?v=2rGbkmlRoxs&t=960s) 内容必须生而合规，生而高效，否则它将沦为企业的负债 — Kalvin Brite > [16:00](https://www.youtube.com/watch?v=2rGbkmlRoxs&t=960s) 创作时预防（Prevention at creation）到生产环境验证（Verify in production） — Kalvin Brite ## Mentioned Siteimprove, Nayaki Nayyar, Kalvin Brite, ADA, EAA, WCAG, Google, ChatGPT, Grok, Perplexity, Gemini, Shell, John Deere, Caterpillar, McDonald's, Yale, Harvard, Lovable, Claude, VS Code, Figma, MCP, Google Docs, Microsoft Word, Gartner, Forrester Wave ## Video description Nayaki Nayyar - CEO, Siteimprove Kalvin Brite - Chief Product Officer, Siteimprove The era of AI experimentation is over. In 2026, enterprise competitive advantage belongs to leaders who move beyond isolated deployments to architect autonomous, goal-driven AI systems that directly advance core business outcomes. Yet the path from pilot to enterprise-wide scale is rarely a technology problem, it is a strategy problem. In this session, Nayaki Nayyar, CEO of Siteimprove and Kalvin Brite, Chief Product Officer of Siteimprove, present a proven framework for closing the strategy-execution gap: the critical distance between AI ambition and measurable business impact. Drawing on Siteimprove's own agentic transformation, including the deployment of specialized agents across its agentic content intelligence platform, Nayaki offers executives a clear-eyed view of what it takes to scale agentic AI across the pillars that matter most: trust architecture, product to platform, and revenue growth. This is a conversation about leadership, governance, and the strategic decisions that separate organizations that capitalize on AI from those that are disrupted by it. Key Takeaways for Attendees: From Silos to Ecosystems: How to move beyond fragmented AI tools toward a unified agentic layer that supports cross-functional goals. Architecting for Integrity: Why data accuracy, accessibility, and brand governance are the essential "guardrails" for autonomous agents acting on behalf of your brand. Quantifying Agentic ROI: Moving past vanity metrics to measure how AI agents actually move the needle on market authority and digital performance. [Source](https://www.youtube.com/watch?v=2rGbkmlRoxs)

[4:42] Nayaki Nayyar: 如果你能帮我们解决一个问题，请务必找到与这些 AI 驱动工具深度集成的方法，在流程早期就检测并修复这些问题。 [4:42] Nayaki Nayyar: 像芝麻街一样简单明了 [9:37] Kalvin Brite: 识别问题（Identify）、采取行动（Act）以及成效验证（Prove）。 [16:00] Kalvin Brite: 内容必须生而合规，生而高效，否则它将沦为企业的负债 [16:00] Kalvin Brite: 创作时预防（Prevention at creation）到生产环境验证（Verify in production）

[0:08] Thank you all for joining us. Um this is Nyaki  Ner. I'm the CEO of site improve. Uh I also have [0:14] Kelvin Bright who is our chief product officer  joining us today. Uh there are three main things [0:21] that we will cover in our session today. We'll  start with the journey that site improve has gone [0:27] through. how we transformed ourselves from the  core SAS business into the entire agentic era. [0:35] We'll go over our vision and strategy, but most  importantly, we'll walk you through some of the [0:40] agentic use cases that we have built and Kelvin  will actually do the demos of those agentic [0:45] use cases. And towards the end, we'll also share  with you some of the customer examples of how our [0:50] customers are going through this journey with us.  Before I jump into all those details, let's take [0:58] a step back and look at what's really happening  around us in the world of content. You know, we [1:07] live in an AIdriven world with infinite content.  And when I say content, we mean web content, [1:16] social content, mobile content, documents. the  sheer volume of content that is being generated [1:25] around us, especially with all these AI tools. And  one of the biggest problems that every customer of [1:34] ours, every large enterprise, every organization  is facing with is to keep this content compliant. [1:44] The reason they have to keep it compliant  because both in North America and in Europe, [1:51] North America has ADA or what we call American  Disability Act and Europe has EAA, European [1:59] Accessibility Act that requires every organization  of a certain size and above to be compliant with [2:06] all their external facing content to make sure  all their external facing content is accessible [2:12] and compliant. So that is one problem that every  single large enterprise customer or company is [2:19] trying to solve. The second problem in this world  of AI as search is transforming as traditional [2:29] search you know when we used to search in uh years  before we would always go to Google and find our [2:35] answers but in today's day and age very few people  are going to Google. the traffic has shifted from [2:43] traditional searches into all these what we call  AIdriven searches. So the organizations, every [2:50] organization, every brand is trying to figure  out how do they show up, how are they visible, [2:56] how their content is visible on all these AIdriven  searches. We as site improve have been in the [3:04] business for over 20 plus years. We are a leader  in Forester wave and a recognized leader uh by [3:13] Gartner. We have over 5,500 customers across every  single industry. Global 2000 companies, Fortune [3:22] 500 companies across financial institutions,  healthcare, public sector, government customers, [3:30] you name it. All the big large brands. And you can  only imagine customers like Shells of the world, [3:37] um, John Deers of the world, Caterpillars of  the world. Just imagine the sheer volume of [3:43] content these companies have. just the sheer  volume of content and as the traffic as the [3:51] content search is shifting away from traditional  searches from Googles of the world into chat [3:57] GPT gro AI perplexity doesn't matter which AI  tool every organization every brand is trying [4:05] to make sure how they become visible not just  visibility but are their product and services [4:11] being represented accurately are their product  and services that they offer is are these AIdriven [4:18] tools actually representing those services  accurately and they really really risk not [4:24] just being not visible but their services  being incorrectly represented. So that's [4:29] one big challenge that every organization that  is struggling with that we are helping solve. [4:34] Kelvin will do a demo of how we are solving that  problem for all of our customers. The second [4:42] this world of AI when everyone is a developer as  everyone is using WE coding tools, everyone can [4:50] build apps, everyone can build content, everyone  can design content. How do you ensure that content [4:59] is compliant and is performing? You know, we  were at our customer advisory board last month in [5:06] Chicago. We had companies like McDonald's of the  world, yields of the world, Harvards of the world, [5:11] pretty much all big large financial institutions,  commercial organizations around the table. And one [5:19] of the things that was consistent across all  these customers was just the sheer volume of [5:25] VIP coding tools that are being used across all  these organizations. And as they're using these [5:29] W coding tools, the volume of content that is  being generated that is not even on the brand, [5:36] let alone making sure that content is accessible  and let alone making sure that content is visible [5:40] and let alone making sure the content services and  products are showing up accurately in all these [5:45] AIdriven searches. So that asked the literally  pretty much everyone around the table said, [5:51] "Nayaki, if you can solve one problem for us,  please figure out a way to integrate with all [6:00] these AIdriven tools or wide coding tools so  that you can detect these issues early in the [6:05] process and remediate them." That's what we call  shift left. You know it's not enough to remediate [6:15] these issues after the content is created after  the website is live after all the uh thousands [6:23] and thousands of pages are already up and running.  If you try to remediate and after afterwards just [6:29] the sheer volume manpower and time it takes is  very very consuming and in the meantime you are [6:35] struggling to keep up with all the regulatory  deadlines and timelines. And the answer is we [6:42] have to shift left at the time of creation. At  the time when a developer is building the apps, [6:49] it doesn't matter whether they're using V coding  tools to build apps, lovables of the world, [6:52] claws of the world, VS codes of the world. As they  are building the apps, can we detect issues in [6:58] real time? Remediate those issues. As content  developers are building content in the design [7:04] tools like Pigmas of the world, detect those  issues, remediate those issues at the time of [7:10] the creation, not as an afterthought. That's what  we did. We launched our vision and strategy around [7:18] what we call agentic content intelligence platform  last year. And since then we have released a whole [7:25] bunch of agentic use cases and agents to detect  issues early in the process and auto remediate [7:33] these issues. Across our platform we have four key  pillars. We have accessibility analytics search [7:42] and content strategy. And across each one of these  pillars in accessibility, we have accessibility [7:47] agents to detect your compliance issues or  accessibility issues early in the process and auto [7:55] remediate those issues. In analytics, we have what  we call conversation analytics agents where you [8:02] can just ask a question and the agent generates  all your reports and dashboards. Imagine a day [8:07] in an age where we used to build these reports and  dashboards manually in every tool, right? just the [8:13] sheer volume of manpower it used to take. Now with  these agents in our platform, you can just ask a [8:19] question. It automatically generates all your  reports and dashboards and give you the right [8:23] answers and insights that you're looking for.  And the search agents we have both traditional [8:28] SEO search agents and AEO what we call advanced  AEO search agents where it ensures your content [8:36] is showing up is visible in all these AIdriven  tools where chart GP rock AI say the world period [8:44] of the world and make sure you're visible your  brand is visible your services are visible your [8:49] products are visible and a content strategy  offering where when you see that content is [8:57] not being represented, your brand and your product  and services are not being represented accurately, [9:01] you can remediate them. You can update your  content briefs, update your content descriptions, [9:06] update your content strategy documents and make  sure all that content is showing up accurately in [9:12] all these AIdriven searches. So this is the entire  platform. I call it Sesame Street simple. You can [9:18] you know one slide you can see what our vision  strategy and the journey is. But most importantly, [9:24] it all comes to life with some of the use cases  and the demos that Kelvin will do right after [9:29] this. With that, let me hand it over to Kelvin. A  big round of applause for Kelvin. Thanks, Mackie. [9:37] I'm Calvin. I lead product at Site Improve. I'm  going to talk a little bit about how we bring [9:41] this whole flywheel to life. So, I'll talk about  how we've embedded agents into each one of these [9:47] pillars of our platform and what that has done  for our customers. So as we've started building [9:51] AI into the platform, it's really condensed our  customers workflows into three main questions and [9:57] it goes around the wheel. First with identifying  what are the issues that exist. These are the [10:02] accessibility issues, the compliance issues, the  search issues and telling you what to do about it. [10:07] The second once you know what issues there are  you need to take some kind of action. So this [10:13] is you know humanled action as well as agentled  action that we enable within the platform itself. [10:19] And then lastly, prove. How do you know any of  those actions that you took actually worked? So, [10:25] um, one thing that is very important about  our platform is the center of this workflow [10:29] itself. We deposit everything, the finding, the  changes, the impact into one system of record, [10:35] one platform that stores all of that so  that our agents can build on that context, [10:40] take more specific action on behalf of your brand,  but also show you what the impact was over time. [10:48] So, let's go one by one. Everything I show  you will just center on each one of those [10:52] questions. The first is where am I invisible?  Where are things working? Where are things not [10:56] working? Um are am I being cited by AI agents that  search for me? Uh is my brand showing up? Is that [11:03] accurate? Is it on brand? Is it is it actually  um in increasing over time or am I decreasing [11:08] over time? So, I'll show you how that works in  the platform itself. So, if we go into this, [11:14] this is our product. This is our advanced AEO  insights uh that we have built and it's not just [11:21] another SEO dashboard. We show you share of voice  across all of these AI models, Chat GBT, Gemini, [11:27] Perplexity, etc. We track share of voice over time  as well as topic authority. How are you trending [11:35] on the topics that you're tracking and and how  is that relative to the competitor set that [11:40] you go against? There are two main numbers that  matter in this mentions. This is the frequency [11:46] that you're mentioned in a conversation as well as  citations. How many times is your content cited in [11:51] these answers and both of those are very important  when we think about the score over time with the [11:57] content that you're creating. Now, it's not just  enough to show you this as a dashboard. You know, [12:03] the biggest thing is what do you do about it? So,  one of the things that we dive deep into into this [12:08] dashboard is showing you what are the biggest  opportunities with the revenue and the impact [12:14] tied to it if you were to make adjustments or  changes to the content over time to show up [12:19] better in these answer engines. So, diving deep  into this allows you to go into that detail. So, [12:25] once you know what to do and what those biggest  opportunities are, the next question to answer [12:31] is what do I do about it? How do I take action  on this? So, there are two pieces to this. The [12:37] first is creating great content that's both  accessible and compliant and performant. Um, [12:43] but it's also about preventing that content  from ever being generated with those issues [12:47] in the first place. So, I'll walk you through what  we've built uh in this uh example in this demo for [12:53] creating content more effectively and efficiency  efficiently once you know what to change. So, this [12:59] is our content agent that we've created. It takes  all of those findings that we have uh surfaced to [13:06] you in the platform itself and puts them to work  by creating content that will actually perform. [13:16] It looks at all the topics that you track, how  it's performing today, where it's performing, [13:21] and then takes that as a baseline to then build  content that will fill those gaps and raise [13:26] your ranking in your share of voice in AI search  engines that look for you and helps increase your [13:32] topic authority over time. Now this is created as  a browser extension so it can work directly into [13:38] um document tools like Google Docs uh Microsoft  Word and some of these others over time. Um but [13:44] it's right there in the editing flow and in  the writing flow so that you as a writer can [13:47] actually create this content with the agent  collaboratively and then as that content is [13:52] being created you can see the score of  that content. How is it showing up in [13:57] answer engines and what is the prediction  of how this will perform once it is live. [14:04] Okay, the second half of taking action is  the prevention side. So Nyaki talked about [14:08] this as more and more people are developers  and building content, building applications, [14:13] how do you ensure that those agents create content  that is performant at the start without thinking [14:17] about fixing it after a publication. So this is  one of the biggest shifts that I think we've made [14:23] as a company. um instead of thinking about  you the workflow we've had for two decades [14:28] uh a customer of ours publishes a website we scan  it we show you the issues they go and fix it that [14:35] worked it was all after the fact now we're  thinking about shifting left and going into [14:40] the source of these tools these AI builders you  know these these agents that are creating on your [14:44] behalf we did this through MCP so we published our  own MCP server so all of these agents that we have [14:49] in our platform can now take action in the tools  in the workflows that you may have within your [14:54] organization. So, an example of this is with  Lovable. Um, no better place to to start than [14:59] here. Um, this is a uh our agent working within  the Lovable interface. We've built a website here [15:05] with Unovable. Um, we're now asking an agent, what  can you do with site improves agents? Um, it comes [15:11] back and says, I can scan, I can detect issues, I  can take action. Now, a lot of this is us exposing [15:17] our tools, our agents to lovables agent to create  content to create these web experiences that are [15:25] accessible and performant from the very beginning.  Now, our agent on this website can scan and find [15:31] those issues, present them back to you, and show  you what you're where you are in terms of your [15:37] goals of EAA compliance or weak compliance and  what you should do to take action. But instead [15:43] of you as a human taking action, it's now agent  to agent. So our agent is now collaborating with [15:49] Lovable's agent to create that content that will  perform and be compliant from the very beginning. [16:00] Now lastly, and probably most importantly  is once you can take all of these actions, [16:06] how do you actually prove that it worked?  that this had an impact on your visibility, [16:10] on your performance, that this investment  was worthwhile. So, as I mentioned before, [16:16] we deposit all of these findings, all of  this information back into the platform, [16:19] into our system of record. And we have analytics  built into the platform itself. We've invested in [16:25] conversational analytics and that just allows you  to ask plain language questions of your data and [16:31] get really simple answers back, but you know that  they're grounded in the source of truth. It gives [16:36] you the data that it pulled, where it pulled  it from, and shows you where you can reference [16:41] that data in the platform itself. So this is our  analytics module within the platform. You can see [16:46] there's a conversational interface. You can ask  plain language questions about past performance, [16:51] region performance. You can go as granular as  you'd like to go and it will give you your data [16:56] back in whatever format you ask for it for.  Maybe just a a list like this. Um, if it's [17:02] something that is graphical, it will present a  graph. Um, very similar to these experiences in, [17:08] you know, cloud, open AAI, some of these  others that you're that we're accustomed [17:11] to. But all of this is pulling from the data  within your content experiences that we track, [17:16] your traffic analytics, the performance,  and the impact all within the platform. [17:24] So, I'll leave you with a couple of final thoughts  of where does all of this start to go? And for us, [17:31] we think about content must be born compliant and  must be born performant. Otherwise, it's going to [17:38] be a liability. So, we think about prevention  at creation, embedding our agents into the time [17:46] of creation so that the content you create is  performant and compliant as as you create it. We [17:52] verify that content in production. It's very hard  to test personalized real-time experiences that [17:58] might be generated at runtime. So, we track that  in production and show you how it's performing [18:04] over time and what those look like as that as  those experiences are live. And then lastly, [18:09] we prove that those experiences are working.  They're driving the business impact that you [18:13] hope they do. So, for us, if you haven't put  it together yet, we think about content must be [18:21] compliant and performant. It's not an eitheror,  it's a both and. And laying the foundation for [18:28] accessibility is the same foundation that you need  to lay for agents, for you know bots, for these [18:35] LLMs to actually ingest and read your content and  site it within answer engines. So these two work [18:42] handinand to create per uh content that performs  for your brand. So I think we are at time here, [18:49] but we have a booth set up. Come visit us. We're  at booth 318. Continue the conversation. We'd love [18:55] to chat with you more and answer any questions  you might have about Sight and Proof. Thank you.