# 从「浏览模型」到「验证模型」:首屏要回答的四个问题(数据研究)

- ID: 95
- URL: https://www.stoyard.com/viewpoint/95
- English URL: https://www.stoyard.com/en/viewpoint/95
- Category: 网站数字化
- English Category: Digital website
- English Title: From "Browse Model" to "Validate Model": Four questions to answer on the first screen

## 中文主体

> 当 AI 摘要把点击率压掉四成、零点击率升到 73%,官网的访问逻辑已经变了。点进来的人不是陌生人,是来确认的。首屏要回答的,从"我们是谁"变成四个具体问题。

![](https://static.glabcms.stoyard.com/news/content/20261121/xpb5fyqbpje.png)

- 栏目分类:网站数字化

- 官网改版|辅助:信息架构、UX、零点击搜索、AI 搜索优化、AI Overviews

- 正文约 1400 字

# **
当 73% 的搜索不再产生点击,官网的第一次见面已经发生过一次了。**

一项预注册的随机实验给出了这组数字:当搜索结果页出现 AI 摘要时,用户点击外部网站的次数从每次 0.62 降到 0.37,降幅约 40%;不产生任何点击的搜索,从 54% 升到 73%。(该研究由 1,065 名美国桌面用户参与、观测两周;目前为工作论文、未同行评审,结论只在美国桌面端且仅在信息类查询上显著——这层限定要保留,否则容易被读成"官网没人看了"。)

数据真正的含义不是"流量没了",而是:**官网的第一次见面,已经在搜索结果里发生过一次。**

那么还会点进官网的人是谁?绝大多数是已经知道你的人:看过一次 AI 摘要的转述、在展会上拿过名片、在同行口中听过你的名字。他们带着一个具体疑问进来,只想确认一件事——然后决定是继续看,还是关掉。

# 浏览模型是从哪来的

过去二十年,官网设计共用同一套线性假设:吸引注意 → 建立兴趣 → 介绍能力 → 引导行动。首屏做成封面,讲一句品牌宣言,配大图或轮播;往下依次是业务、案例、联系方式。

这套模型在 2010 年代成立,因为入口就是搜索结果页:用户从一排蓝色链接里挑一个,对这家公司一无所知,需要被介绍、被说服、被引导。页面因此被组织成一条叙事线,像一本宣传册。

它的成立条件只有一个:**访客是陌生人。**

条件消失,每一环都开始失效。首屏在讲"我们是谁",访客想知道"你是不是我要找的那家";第二屏在讲愿景,访客在找"有没有和我一样工况的客户";案例页在铺 Logo 墙,访客在判断"这些人是不是和我同一类问题"。

更要紧的是,浏览模型默认访客有耐心。而验证型访客没有——他只关心某个答案在哪,找不到就走,走的成本几乎为零。

# 验证模型的三个特征

把这种访问单独命名,更容易讨论它的设计。我们称之为**验证型访问(Verification Visit)**:访客带着既有认知回到品牌官网,只为核实某一个具体事实。它与探索型访问的区别是——不需要被介绍,只需要被确认。

三个特征决定页面做法:

- **带着名字来**:首屏不必再做自我介绍,要做的是让他在两秒内确认"来对地方了"。

- **只查一件事**:一次访问通常只为一个问题服务,其余内容都是噪音。页面要能被"跳到答案",而不是只能被"读完"。

- **容忍度极低**:找不到答案,他不会问你,他会去问 AI,或者问同行。

这里划一条边界:验证模型不取代浏览模型。新品发布会页、品牌周年页、招聘主页这类有明确叙事任务的页面,浏览模型依然有效。问题在于很多企业把整个官网当成一张发布会海报来做——那是把少数场景的做法,套在了承担全部确认职能的基础设施上。

# AI 全面生成摘要之后,官网成了证据基础设施

只看点击,结论似乎是官网在贬值。但另一组数字要放在一起看:AI 摘要里的那些引用链接,只贡献约 7% 的外链点击。绝大多数人读了答案,不点来源。

这说明了一件更要紧的事:**用户不再通过访问网站获取信息,但仍需要通过官网确认信息。**

获取环节,官网输给了摘要;确认环节,官网是唯一还被信任的地方。这也解释了为什么同一家企业在生成式回答中的被引用比例,可以高于它在传统媒体上的曝光排名——官网作为结构化、可溯源的自有信源,权重正在被重新抬高。

于是官网的职能静默转移:它不再是获客渠道(获客发生在摘要、社群、口碑、展会),它变成了证据基础设施——所有在别处听来的说法,最后都要回到这里核对一次。

职能变了,衡量方式也要变。访问量、停留时长、跳出率考核的是一个"确认型"官网,等于用零售店的指标考核质检部门。更贴近真实的三个指标:目标问题上 AI 回答里有没有你(被引用度)、见过你一次的人会不会带着品牌词回来(品牌词回访)、销售在客户会前发的第一个链接是不是你的某一页(销售复用率)。

# 首屏要回答的四个问题

这四个问题和站点结构最基础的四层同构,只是顺序要从"展示顺序"改成"验证顺序"。每一条都配一个页面上该出现的证据,而不只是一句该写的话。

**第一问:是不是我要找的那家。** 需要的不是形容词,是一个可被复述的定义:你在产业链哪个位置、解决谁的什么问题、不做哪些事。三行以内。这里最常见的问题是"什么都想做"——定义里塞进五个业务方向,它就不再是定义,而是目录。

**第二问:做到什么程度。** 规模、客户构成、资质、关键里程碑,要能快速扫描。数字比形容词可信,时间点比"多年深耕"可信。这一问的难点是口径:同一个规模数字,在官网、财报、销售 PPT 里往往是三个版本。访客不会追究哪个对,他只会记住"这家公司说不清自己有多大"。

**第三问:有没有和我相似的情况。** 案例页的重点不是一墙 Logo,而是"工况":对方原来是什么状态、你们改了什么、结果如何衡量。案例要按问题分类,而不是按行业分类。一墙 Logo 证明的是关系,不是能力。

**第四问:下一步找谁,多久有回应。** 一个具体的人、一个具体的入口、一个可兑现的时间承诺。访客往往不是决策者,他需要把"下一步"带回去给老板。"联系我们"四个字不是答案:它没说找谁、没说多久回、也没给低门槛的第一步。

# 访客黄金八秒

因为它不是设计问题,是责任问题。

首屏由市场部定语气,案例由销售提供,数字由董秘和法务把关,联系方式散落在三个页面。每一方在自己的权限内都做对了,但没有人对"访客能不能在八秒内完成确认"负责。

在禾赛科技、非夕科技这类技术型项目里,我们花时间最多的往往不是视觉,而是把"我们是谁"这句话从不同部门的表述里收敛成一个版本。收敛完成之后,页面设计反而很快。

通常三步:开一次只有决策者参加的收口会,产出一句话定义;把它拆成三处必须对齐的表述——官网、投资者与财务材料、销售话术;指定一个人对一致性负责,并在业务变化时更新。

# 一个可以自己做的检查

不要找设计师,找一个合作三年以上的老客户,把首屏给他看八秒,然后问四句:我们做什么、做到什么程度、有没有和我情况接近的客户、下一步找谁。每答出一条计 2 分,低于 6 分说明首屏还在做封面,低于 4 分说明问题不在首屏,在定义本身。

官网不需要变得更漂亮,它需要变得更可被确认。

**

**

**参考资料

**
[1] Saharsh Agarwal(印度商学院)、Ananya Sen(卡内基梅隆大学),预注册随机实验,AEA RCT Registry 登记,2026 年 4 月工作论文(1,065 名美国桌面 Chrome 用户,2026 年 1 月观测两周):AI 摘要出现时外链点击 0.37 vs 0.62,零点击率 54% → 73%,广告点击无显著变化;未经同行评审,仅美国桌面端,仅信息类查询显著。 [https://sitech.ge/en/blog/google-s-ai-answers-reducing-search-traffic-by-40/](https://sitech.ge/en/blog/google-s-ai-answers-reducing-search-traffic-by-40/)

[2] 同组实验的另一项结论:AI 摘要内的引用链接仅贡献约 7% 的外链点击。 [https://www.omegatechnologysolutionsgroupinc.com/blog/google-ai-overviews-cut-publisher-traffic-40-study-finds-c2f852](https://www.omegatechnologysolutionsgroupinc.com/blog/google-ai-overviews-cut-publisher-traffic-40-study-finds-c2f852)

[3] DataForSEO 月度监测(2000 万关键词样本):AI 摘要已覆盖约 61% 的被追踪关键词。 [https://www.rswebsols.com/news/randomized-study-reveals-google-ai-summaries-reduce-outbound-link-clicks-by-40-with-ai-overviews-now-featured-in-61-of-search-results/](https://www.rswebsols.com/news/randomized-study-reveals-google-ai-summaries-reduce-outbound-link-clicks-by-40-with-ai-overviews-now-featured-in-61-of-search-results/)

[4] Cision,《2026 年全球头部企业传播热门话题报告》:企业官网在生成式 AI 回复中的权重显著提升,存在官网在 LLM 输出中占比高于传统媒体曝光的案例。 [https://www.docin.com/touch_new/preview_new.do?id=4980368555](https://www.docin.com/touch_new/preview_new.do?id=4980368555)
** **

## English Content

> When the AI summary reduced the click-through rate by 40% and the zero click-through rate rose to 73%, the access logic of the official website has changed. The person who came in was not a stranger, but came to confirm. The first screen to answer, from "

- Column classification: website digitization

- Main keywords: official website revision | auxiliary: information architecture, UX, zero-click search, AI search optimization, AI Overviews

- From browsing the model to validating the model: four questions to be answered on the first screen of the official website

- Text of about 1400 words

# **
When 73% of searches no longer generate clicks, the first meeting on the official website has already happened once.**

# **

![](https://static.glabcms.stoyard.com/news/content/20261121/xpb5fyqbpje.png)

**

A pre-registered randomized experiment gave this set of numbers: when a AI summary appeared on the search results page, the number of times users clicked on an external website dropped from 0.62 to 0.37, a drop of about 40%; Search without any clicks rose from 54% to 73%. (The study was attended by 1,065 U.S. desktop users and observed for two weeks; it is currently a working paper, not peer-reviewed, and the conclusions are only significant on the U.S. desktop side and only on information queries-this limit should be retained, otherwise it is easy to read as "no one on the official website.")

The real meaning of data is not "traffic is gone",:**The first meeting of the official website has already happened once in the search results.**

So who are the people who will click into the official website? Most of them are people who already know you: they have seen a AI summary of the report, took a business card at the exhibition, and heard your name in the mouth of their peers. They come in with a specific question and want to confirm only one thing-and then decide whether to keep watching or turn it off.

# Browse where the model came from

Over the past two decades, official website design has shared the same set of linear assumptions: attracting attention → building interest → introducing ability → guiding action. The first screen is made into a cover, with a brand declaration, with a big picture or carousel. Next are business, case and contact information.

This model was established in the 2010 s because the portal is the search results page: users pick one from a row of blue links, know nothing about the company, and need to be introduced, persuaded, and guided. The pages are thus organized into a narrative line, like a brochure.

It has only one condition:**Visitors are strangers.**

When the conditions disappear, each link begins to fail. The first screen is talking about "who are we" and the visitor wants to know "are you the one I am looking for?" The second screen is talking about the vision, and the visitor is looking for "are there any customers who work in the same condition as me?" The case page is laying a Logo wall, and the visitor is judging "are these people having the same kind of problems as me".

More importantly, the browsing model defaults to the patience of visitors. However, the verification visitor does not-he only cares about where a certain answer is, and if he cannot find it, he will leave, and the cost of leaving is almost zero.

# Three characteristics of the validation model

By naming this access separately, it is easier to discuss its design. We call it**Authenticated Access (Verification Visit)**: Visitors return to the brand official website with existing knowledge, only to verify a specific fact. The difference between it and exploratory access is that it doesn't need to be introduced, it just needs to be confirmed.

Three characteristics determine page practices:

- **Come with a name**The first screen does not need to introduce itself again, what needs to be done is to let him confirm "coming to the right place" within two seconds ".

- **Just check one thing.**: A visit usually serves only one problem, and the rest is noise. The page should be able to be "jumped to the answer", not only "read".

- **Very low tolerance**If he can't find the answer, he won't ask you. he will ask AI or his colleagues.

A boundary is drawn here: the validation model does not replace the browsing model. The browsing model is still valid for pages with clear narrative tasks such as the new product launch page, brand anniversary page and recruitment homepage. The problem is that many companies treat the entire official website as a press conference poster-that is to put a few scenes on the infrastructure that undertakes all the confirmation functions.

# After AI the full generation of abstracts, the website becomes the evidence infrastructure

Just look at the click, the conclusion seems to be the official website in the devaluation. However, another set of figures should be put together: the reference links in the AI summary only contribute about 7% of the external chain clicks. Most people read the answer, not the source.

This suggests something more important:**Users no longer obtain information by visiting the website, but still need to confirm information through the website.**

In the acquisition link, official website lost to the abstract. In the confirmation link, official website is the only place that is still trusted. This also explains why the same company's citation rate in generated answers can be higher than its exposure ranking in traditional media-official website, as a structured and traceable source of its own information, is being re-weighted.

As a result, the functions of the official website have shifted silently: it is no longer a channel for obtaining customers (obtaining customers occurs in abstracts, communities, word of mouth, exhibitions), it has become an evidence infrastructure-all statements heard elsewhere have to be checked back here at last.

As functions change, so too must the ways we measure performance. The number of visits, length of stay and jump rate are assessed by a "confirmed" official website, which is equivalent to assessing the quality inspection department with the indicators of retail stores. Three indicators closer to the truth: whether you are in the AI answer to the target question (cited degree), whether people who have seen you once will come back with brand words (brand words return visit), and whether the first link sent by sales before the customer meeting is one of your pages (sales reuse rate).

# The first screen to answer four questions

These four problems are isomorphic to the most basic four-layer structure of the site, except that the order should be changed from "display order" to "verification order". Each article is accompanied by a piece of evidence that should appear on the page, not just a sentence that should be written.

**First question: Which one am I looking.** What is needed is not an adjective, but a definition that can be repeated: where you are in the industrial chain, what problems you solve, and what you don't do. Within three lines. The most common problem here is "want to do anything"-with five business directions in the definition, it is no longer a definition, but a directory.

**Second question: to what extent.** Scale, customer composition, qualifications, key milestones, to be able to scan quickly. Numbers are more credible than adjectives, and time points are more credible than "years of deep cultivation. The difficulty of this question is the caliber: the same scale figure is often three versions in official website, financial report and sales PPT. The visitor will not investigate which pair, he will only remember that "the company can't tell how big it is".

**Third question: Is there anything similar to me.** The focus of the case page is not a wall Logo, but a "working condition": what was the original state of the other party, what you changed, and how to measure the result. Cases should be categorized by issue rather than by industry. A logo on the wall proves connections, not competence.

**The fourth question: who to look for next and how long to respond.** A specific person, a specific entrance, a time commitment that can be fulfilled. The visitor is often not the decision maker, he needs to take the "next step" back to the boss. The word "contact us" is not the answer: it does not say who to look for, does not say how long it will take, and does not give the first step to a low threshold.

# Eight seconds of visitor gold

Because it's not a design issue, it's a responsibility issue.

The first screen is determined by the marketing department, the case is provided by the sales department, the number is checked by the secretary of the board and the legal department, and the contact information is scattered on three pages. Each party did the right thing within its own authority, but no one was responsible for "whether the visitor could complete the confirmation in eight seconds.

In technology-based projects such as Wo Sai Technology and Fei Xi Technology, what we spend most time on is often not vision, but the phrase "who are we" converges into a version from the expressions of different departments. After the convergence is completed, the page design is very fast.

Usually three steps: hold a closing meeting with only decision makers and produce a one-sentence definition; break it down into three statements that must be aligned-official website, investor and financial materials, and sales speech; and designate a person responsible for consistency and update it when the business changes.

# A check that can be done on its own

Don't look for a designer, find an old customer who has cooperated for more than three years, show him the first screen for eight seconds, and then ask four sentences: what do we do, to what extent, are there any customers close to me, and who to look for next. Each answer counts 2 points, less than 6 points indicates that the first screen is still making the cover, and less than 4 points indicates that the problem is not on the first screen, but in the definition itself.

The website does not need to become more beautiful, it needs to become more identifiable.

**

**

**References

**
[1] Saharsh Agarwal (Indian Business School), Ananya Sen (Carnegie Mellon University), Pre-registration Randomized Experiment, AEA RCT Registry Registration, April 2026 Work Paper (1,065 US Desktop Chrome Users, Two Weeks in January 2026): When the AI summary appears, the external chain clicks 0.37 vs 0.62, zero click rate 54% → 73%, no significant change in advertisement clicks; Without peer review, only the US desktop side, only the information class query is significant. [https:// sitech.ge/en/博客/google-s-ai-answers-reducing-search-traffic-by-40/](https://sitech.ge/en/blog/google-s-ai-answers-reducing-search-traffic-by-40/)

[2] Another conclusion of the same group of experiments: the reference link in the AI abstract only contributes about 7% of the external chain clicks. [https://www.omegatechnologysolutionsgroupinc.com/blog/ google-ai-overviews-cut-publisher-traffic-40-study-finds-c2f852](https://www.omegatechnologysolutionsgroupinc.com/blog/google-ai-overviews-cut-publisher-traffic-40-study-finds-c2f852)

[3] DataForSEO monthly monitoring (20 million keyword sample): The AI summary has covered about 61% of the tracked keywords. [https://www.rswebsols.com/news/ randomized-study-reveals-google-ai-summaries-reduce-outbound-link-clicks-by-40-with-ai-overviews-now-featured-in-61-of-search-results/](https://www.rswebsols.com/news/randomized-study-reveals-google-ai-summaries-reduce-outbound-link-clicks-by-40-with-ai-overviews-now-featured-in-61-of-search-results/)

[4] Cision, "Global Head Enterprise Communication Hot Topics Report 2026": The weight of corporate official websites in generative AI responses has increased significantly, and there are cases where official websites account for a higher proportion of LLM output than traditional media exposure. [https://www.docin.com/touch_new/preview_new.do?id = 4980368555](https://www.docin.com/touch_new/preview_new.do?id=4980368555)
** **