作者: Ido Finder, Assaf Elovic, Gad Shalev, Liad YosefarXiv 2609.34951

AX is the New AEO(AX 是新的 AEO)于 2026-09-28 提交到 arXiv,作者 Ido Finder, Assaf Elovic, Gad Shalev, Liad Yosef。去 arXiv 读原文,或下载 PDF。

为什么和被引用有关

最重要的一点。 面向智能体时代的优化核心在于让自身官网能够被机器直接抓取与解析,能够被智能体顺利阅读比在全网各处散布品牌讨论更能决定推荐成败。

摘要

In 2023, AI models answered from training data and hallucinated when it ran out, and businesses were told to seed that knowledge. Models’ training knowledge has since given way to live web search, and the advice followed it there: answer-engine optimization, or AEO, now tells businesses to scatter breadcrumbs across forum threads, listicles, and off-site citations, so AI engines are likelier to surface and recommend them. But being surfaced is no longer enough: an agent opens the results and reads them before deciding, and one buyer question sends it through several rounds of search and fetch. What decides the outcome at this drill-down step is whether the agent can fetch and read the business’s own site: agent experience (AX). We argue that AX is the new AEO. We run 37,927 agent journeys, each a buyer question about a business, across four independent harnesses over 1,056 real businesses, matched on fame, prior model knowledge, and two AEO proxies, then split based on their AX level. Only 7-10% of the finished answer comes from the model’s training knowledge, whether or not the site is readable. Agent-ready businesses have answers built from their own pages 78% of the time against 56% and are clearly recommended 1.9x more often, while a grounded answer about a not-agent-ready business costs the agent 64% more on average. Holding business, harness, and question fixed, answers built from the site are 41% more accurate on average. The dominant failure is not fabrication but omission: web-built answers are 3.7x more likely to contain none of the facts the buyer asked for. Baselines differ sharply across the four harnesses, with clear-recommendation rates varying sevenfold from stack to stack, yet the recommendation gap holds in every one. In the agentic web era, being readable beats being talked about, and improving a site’s AX is the strongest lever a business has.

译文:在 2023 年,AI 模型依据训练数据进行回答并在数据不足时产生幻觉,企业当时被告知要向这些知识库中灌输内容。此后模型的训练知识逐渐让位于实时网络搜索,相关建议也随之转变:现在的回答引擎优化(AEO)建议企业在论坛帖子、清单推荐和站外引用中散布线索,以便 AI 引擎更容易发现并推荐它们。但仅仅被检索出来已经不够了:智能体在做出决定前会打开并阅读这些结果,一个买家问题就会触发多轮搜索与抓取。在这一深入探索阶段决定结果的,是智能体能否抓取并阅读企业自身的网站,即智能体体验(AX)。我们认为 AX 就是新的 AEO。我们在四个独立测试平台上运行了 37,927 次智能体任务,每次任务都是针对某家企业的买家提问,覆盖了 1,056 家真实企业。这些企业在知名度、模型先验知识和两项 AEO 指标上进行了匹配,随后根据其 AX 水平进行划分。无论网站是否易读,最终回答中只有 7-10% 来自模型的训练知识。具备智能体就绪状态的企业,其回答有 78% 的时间基于自身页面构建,而未就绪的企业这一比例为 56%;前者的明确推荐率达到后者的 1.9 倍,同时智能体为未就绪企业生成有据可依回答的成本平均高出 64%。在固定企业、测试平台和提问的情况下,直接基于自身网站构建的回答在准确率上平均高出 41%。主要的失败表现为遗漏:基于外部网络生成的回答遗漏买家所需全部事实的概率要高出 3.7 倍。四个平台的基线差异显著,明确推荐率在不同框架间存在高达七倍的差距,但推荐率差距在每个框架中均稳定存在。在智能体网络时代,自身可读胜过被外界谈论,提升网站的 AX 是企业拥有的最强杠杆。

关键术语

论文原文

现存的大多数网站在技术规范上并没有为自动化智能体做好准备。

Across nearly 100,000 sites, fewer than 1% earn its top grade of A or above: most of the web is not ready for the consumers it now has, and a standards effort is under way to define what readiness requires ( Anthropic, 2024 , Microsoft, 2025 , agentready.org, 2026 ) .

译文:在近 10 万个网站中,获得 A 级或以上最高评级的比例不足 1%:大部分网络尚未为它现在迎来的这类消费者做好准备,旨在明确就绪要求的标准化工作正在推进中。

在搜索引擎中被发现与决定最终回答内容是两个截然不同的步骤。

A single buyer question sends an agent through several rounds of search and fetch, and what it reads in those rounds is what the answer gets built from. SEO and current AEO both govern the surfacing step; the drill-down step decides what the answer is made of.

译文:一个买家问题就会触发智能体进行多轮搜索与抓取,而它在这些轮次中读到的内容决定了最终回答的构成。SEO 和目前的 AEO 掌管着暴露展示环节,深入抓取环节则决定了回答究竟由什么内容构成。

当官网无法被智能体读取时,系统会转而采纳不受控的第三方数据源。

If the agent can read you, it grounds its answer in your first-party facts. If it cannot, because the page is JavaScript-only, because the site’s bot controls block the agent ( Prince, 2025 ) , or because access is otherwise gated ( Longpre et al., 2024 ) , the agent does not stop. It has the open web to fall back on: third-party pages, competitors, listicles, aggregators.

译文:如果智能体能够阅读你的页面,它就会把回答建立在你的第一方事实之上。如果由于页面纯依赖客户端渲染、站点防护拦截或访问受限而无法读取,智能体并不会停下来,它会退回到开放网络中寻找替代来源:第三方页面、竞争对手、清单汇总和聚合网站。

官网的可读性是企业在生成式搜索生态中真正能够自主掌控的要素。

Being found still matters. Being read is what decides the answer, and it is the one input the business owns. AX is the new AEO, and it is where the investment belongs.

译文:被搜索引擎找到依然重要。被智能体读懂才是决定回答的关键,而且这是企业完全自主掌控的一项输入。AX 就是新的 AEO,也是资源应该倾注的地方。

论文里的图

评测显示在所有测试系统中,智能体就绪度良好的网站获得明确采购推荐的比例接近未就绪网站的两倍。

Figure 4: One in five answers clearly recommends an agent-ready business; one in nine when it is not agent-ready. Bars a

图来自论文 arXiv:2609.34951,版权归作者。原图说明:Figure 4: One in five answers clearly recommends an agent-ready business; one in nine when it is not agent-ready. Bars are the share of answers that both blind judges rate a clear recommendation, all four stacks pooled; both judges must award the top grade for an answer to count (permutation p < 0.0001 p<0.0001 ).

作者自己说的含义

可以怎么做

  1. 保证核心产品功能与定价列表能够通过普通文本格式直接获取,无需依赖前端脚本加载。 依据:基于官网第一方信息生成的答案在事实准确度上相比外部信息来源高出 41%。
  2. 检查网站的反爬与防火墙策略,允许可信的智能体研究抓取工具读取公开信息。 依据:智能体尝试为未就绪站点寻找佐证信息时,由于额外搜索跳转导致调用成本平均高出 64%。
  3. 将内容优化重点从外部渠道铺量转移到提升官网本身的机器解析能力上。 依据:具备良好智能体就绪度的企业在评估中获得明确推荐的几率要高出 1.9 倍。

注意。 本项研究仅测试了以英文为主的软件服务与电商类企业的客观事实查询场景,未涉及复杂多轮交互任务与事务处理场景。

一句话

现在的 AI 系统在回答用户提问时主要依靠实时检索网页信息。当官方网站无法被程序直接阅读时,智能体只能参考第三方二手资料,导致事实严重遗漏并降低推荐意愿。建设易于机器抓取的网站体验,是让企业在生成式搜索中获得准确推荐的核心手段。

去 arXiv 读论文 →

这篇解读根据论文原文写成,发布前有人看过。引用的段落是原话,数字都是作者自己报告的。方法和完整结果请读论文。