Core Web VitalsPage ExperienceSEO IssuesGEO

Core Web Vitals Issues That Hurt SEO and AI Rankings

How slow LCP, poor INP, and layout shift become SEO and GEO issues: they suppress Google rankings and leave AI crawlers with an incomplete page to cite.

A page can be indexed, unique, and still lose the query because it is slow or unstable on a phone. Core Web Vitals are the page experience measures Google publishes for that problem: largest contentful paint (LCP), interaction to next paint (INP), and cumulative layout shift (CLS). They are SEO issues when real visitors wait, and GEO issues when an AI crawler times out or stores a shell with no answer to quote.

This guide follows common SEO and GEO issues that hurt rankings. Use the indexing and crawl guide when the URL is not in the index, and duplicate content and keyword cannibalization when the wrong URL is the one that ranks. Use this one when the page is indexed and the field data for speed or stability is poor.

What Core Web Vitals measure

Google evaluates page experience from Chrome field data, at the 75th percentile, separately for mobile and desktop. A lab score from Lighthouse is a clue. It is not the number Search uses.

MetricGoodWhat visitors feelUsual cause
LCP2.5 seconds or lessThe main content appears lateHero image, web font, slow server, render-blocking files
INP200 milliseconds or lessTaps and clicks feel stuckHeavy JavaScript, third-party tags, long tasks on the main thread
CLS0.1 or lessThe page jumps while readingImages without size, banners, fonts that swap late

“Needs improvement” and “poor” sit above those lines. A template that fails on mobile can drag every URL that uses it. A ranking drop after a redesign often shows up here, as a page experience issue, before it shows up as a keyword issue.

Page speed as an SEO ranking problem

Page experience does not outrank a clearly better page. It does decide close contests, and it decides whether Googlebot finishes rendering a JavaScript-heavy URL before it moves on. A slow website ranking is often this, not a missing keyword.

On Hong Kong sites the pattern is familiar:

  • A full-width hero image, uncompressed, and lazy-loaded even though it is the LCP element
  • Traditional Chinese webfonts downloaded before any text paints
  • Tag managers, live chat, and a maps embed competing with the article on the main thread
  • Cookie bars and sticky headers that push content down after the first paint
  • The same heavy template on both the English and Traditional Chinese URLs

Check field data in Search Console’s Core Web Vitals report, not a single Lighthouse run on a desktop office connection. Mobile on a cellular network is the view that matches most local searches.

Fix the template. Compressing one blog image will not repair a font or a chat widget that ships on every page.

Why page experience is a GEO issue

Generative engine optimization needs a page a crawler can fetch and a passage it can quote. How to rank in AI search covers the shape of that passage. Speed and rendering decide whether the crawler ever sees it.

Many AI crawlers request the HTML and do not run the page’s JavaScript the way Chrome does. If the answer, the headings, or the comparison table appear only after hydration, the crawler stores a thin shell. A slow time to first byte makes that worse: the client may close the connection before the document finishes, so the citation is missing the section you wrote for it.

Layout shift matters less to a crawler than to a person. The shared failure is the same stack of scripts and images. A page that is heavy enough to fail INP is often heavy enough that the first response is not the article.

Do not “fix” this by blocking AI crawlers in robots.txt. Blocking Googlebot is an indexing problem, covered in the crawl guide. Blocking GPTBot, PerplexityBot, or Google-Extended can be a product choice, and it is a GEO issue if those products are where you wanted to be cited. It is not a page-speed fix.

A practical order of fixes

  1. Put the answer in the initial HTML. The title, the H1, the opening paragraph, and the table you want quoted should not depend on JavaScript.
  2. Give the LCP image a real size, a modern format, and a high priority. Do not lazy-load the element that is supposed to paint first.
  3. Reserve space for images, ads, and banners so CLS stays at or below 0.1. Set width and height, or an aspect ratio, before the file arrives.
  4. Cut main-thread JavaScript. Delay chat, maps, and non-essential tags until after the first interaction, or remove them from article templates.
  5. Subset or self-host the Chinese and English fonts you actually use. A full CJK family is a common LCP delay on bilingual Hong Kong sites.
  6. Read the Core Web Vitals report after the deploy, on mobile, and wait for new field data. A lab score can improve the same day. The ranking input moves when Chrome users have generated a new 75th percentile.

Then re-check the URL in Search Console and fetch the raw HTML the way a simple crawler would. If the passage is in that HTML and the field metrics are in the “good” range, a remaining citation gap is a content and entity problem, not a speed problem. Content formats AI platforms cite is the next step.

Questions teams ask

Do Core Web Vitals affect AI rankings?

They affect whether the page is retrievable and pleasant enough to compete. Google AI Overviews still lean on indexed pages, and other AI products cite pages their crawlers could fetch. A slow or JavaScript-only page can fail that step. Passing the three metrics does not by itself earn a citation. The page still needs a clear answer, stable facts, and an author, as described in the broader SEO and GEO issues guide.

What counts as a good LCP?

2.5 seconds or less at the 75th percentile of real visits. Between 2.5 and 4 seconds needs improvement. Above 4 seconds is poor. Measure the URL template your visitors actually hit, on mobile, not a throttled lab test of a cached desktop load.

Why is the mobile score worse than desktop?

The phone has a slower CPU, a less stable network, and often the same oversized image and script bundle. Hong Kong searches are heavily mobile. A desktop “good” rating can hide a mobile “poor” rating, and the mobile URL group is the one that matches those queries.

Will fixing page speed recover a ranking drop?

It recovers the part of the drop that page experience caused. It will not repair a noindex tag, a duplicate canonical, or a thin page. If Search Console shows an indexing exclusion, start there. If several URLs target the same query, fix cannibalization before you spend the month on milliseconds.

What to review this month

Open the Core Web Vitals report and list URL groups that are poor on mobile. For each group, name the LCP element, the longest JavaScript task, and any banner that shifts the layout. Ship one template change, confirm the answer is in the raw HTML, and leave the crawlers you want AI answers to use unblocked.

Page experience is a technical SEO issue with a GEO consequence: visitors never reach the answer, and crawlers never store it. Treat LCP, INP, and CLS as part of the same audit as indexing and duplicate URLs.

核心網頁指標頁面體驗SEO問題GEO

核心網頁指標問題:LCP、INP 與版面位移如何影響 SEO 與 GEO

說明最大內容繪製過慢、互動延遲與累積版面位移,如何成為拖低 Google 排名的 SEO 問題,並讓 ChatGPT、Perplexity 與 AI Overview 的爬蟲抓不到完整可引用內容。

頁面可以已被索引、內容也不重複,卻因為在手機上又慢又不穩而輸掉查詢。核心網頁指標是 Google 公開的頁面體驗量度:最大內容繪製(LCP)、下次繪製互動延遲(INP)與累積版面位移(CLS)。訪客要乾等,這就是 SEO問題;AI 爬蟲逾時,或只存下一頁沒有答案的空殼,這就是 GEO問題。

本篇接在常見 SEO 與 GEO 問題之後。網址不在索引裡,用索引與爬取問題;排名的是錯誤網址,用重複內容與關鍵字互搶。頁面已索引,但速度或穩定度的實地數據偏差,就用本篇。

核心網頁指標在量什麼

Google 用 Chrome 的實地數據評估頁面體驗,取第 75 百分位,手機與桌面分開計算。Lighthouse 的實驗室分數只是線索,不是搜尋使用的數字。

指標良好訪客的感受常見原因
LCP2.5 秒或以下主要內容很晚才出現主視覺圖片、網頁字型、伺服器慢、阻擋繪製的檔案
INP200 毫秒或以下點按像卡住JavaScript 過重、第三方代碼、主執行緒上的長任務
CLS0.1 或以下閱讀時頁面亂跳圖片沒有尺寸、橫幅、字型太晚才替換

超過這些線就是「需要改善」或「不良」。手機上失敗的範本會拖累每一個使用它的網址。改版之後的 SEO排名問題,常常先出現在頁面體驗,而不是關鍵字。

網頁速度如何變成排名問題

頁面體驗不會壓過明顯更好的另一頁。它會決定接近的比較,也會決定 Googlebot 能否在離開之前,把依賴 JavaScript 的網址繪製完。網站速度慢而沒有排名,往往是這個原因,不是少了一個關鍵字。

香港網站常見的模式:

  • 全寬主視覺又大又未壓縮,而且被延遲載入,但它其實就是 LCP 元素
  • 繁體中文字型在任何文字繪製之前就先下載
  • 代碼管理工具、線上客服與地圖嵌入,跟文章搶主執行緒
  • Cookie 列與固定頁首在第一次繪製之後把內容往下推
  • 英文與繁體中文網址共用同一套過重的範本

看 Search Console 的核心網頁指標報告,不要只在辦公室的桌面連線上跑一次 Lighthouse。手機加上流動網絡,才接近大多數本地搜尋的情況。

修的是範本。壓一張網誌圖片,修不好每一頁都載入的字型或客服小工具。

為什麼頁面體驗也是 GEO 問題

生成式引擎優化需要爬蟲抓得到的頁面,以及引用得了的段落。如何在 AI 搜尋中提升可見度講的是那段文字的形狀。速度與繪製方式決定爬蟲看不看得到它。

不少 AI 爬蟲只請求 HTML,不會像 Chrome 那樣執行頁面上的 JavaScript。答案、標題或比較表若要等水合之後才出現,爬蟲存下的就是薄薄的空殼。首位元組時間太慢會更糟:連線可能在文件傳完之前就斷了,你為引用而寫的那一節根本不在抓取結果裡。

版面位移對爬蟲的影響小於對人的影響。共同的失敗是同一疊指令碼與圖片。重到過不了 INP 的頁面,第一個回應往往也不是那篇文章。

不要用 robots.txt 封鎖 AI 爬蟲來「修好」速度。封鎖 Googlebot 是索引問題,見爬取指南。封鎖 GPTBot、PerplexityBot 或 Google-Extended 可以是產品決定;若你希望那些產品引用你,這就是 GEO 問題。它不是網頁速度的修法。

實務上的修正順序

  1. 把答案放進最初的 HTML。標題、H1、開首段,以及你希望被引用的表格,不應依賴 JavaScript。
  2. 給 LCP 圖片真實尺寸、現代格式,以及較高的載入優先級。不要延遲載入那個本應最先繪製的元素。
  3. 為圖片、廣告與橫幅預留空間,讓 CLS 維持在 0.1 或以下。檔案到達之前就設定寬高或長寬比。
  4. 削減主執行緒上的 JavaScript。把客服、地圖與非必要代碼延到第一次互動之後,或從文章範本移除。
  5. 只子集化或自行託管你真正使用的中英文字型。完整的中日韓字型家族,是雙語香港網站常見的 LCP 延遲。
  6. 部署之後看核心網頁指標報告的手機數據,並等待新的實地數據。實驗室分數當天就可以變好。排名所用的輸入,要等 Chrome 使用者產生新的第 75 百分位才會動。

然後在 Search Console 再查一次該網址,並像簡單爬蟲那樣抓原始 HTML。段落已在 HTML 裡,實地指標也在「良好」範圍,剩下的引用缺口就是內容與實體問題,不是速度問題。下一步見AI 平台引用的內容格式。

團隊常問的問題

核心網頁指標會影響 AI 排名嗎?

它們影響頁面是否抓得到、以及是否順暢到有競爭力。Google AI Overview 仍然倚賴已索引的頁面,其他 AI 產品則引用其爬蟲抓得到的頁面。又慢、或只有 JavaScript 才有內文的頁面,會在這一步失敗。三項指標都達標,本身不會帶來引用。頁面仍需要清楚的答案、穩定的事實與作者,如常見 SEO 與 GEO 問題所述。

怎樣才算良好的 LCP?

真實訪問第 75 百分位在 2.5 秒或以下。2.5 到 4 秒是需要改善。超過 4 秒是不良。量的是訪客真正打開的網址範本、而且是手機,不是已快取的桌面負載再加模擬限速。

為什麼手機分數比桌面差?

手機的處理器較慢、網絡較不穩,卻常常載入同樣過大的圖片與指令碼。香港搜尋以手機為主。桌面「良好」可以蓋住手機「不良」,而對上那些查詢的是手機那一組網址。

修好網頁速度就能找回排名下跌嗎?

它找回的是頁面體驗造成的那一部分。它修不好 noindex、重複的 canonical,或內容單薄的頁面。Search Console 若顯示索引排除,先處理那裡。若多個網址搶同一個查詢,先處理關鍵字互搶,再花一個月去擠毫秒。

這個月要檢查什麼

打開核心網頁指標報告,列出手機上屬於不良的網址群組。每一組寫下 LCP 元素、最長的 JavaScript 任務,以及會把版面推開的橫幅。推出一項範本修改,確認答案在原始 HTML 裡,並保持你希望 AI 答案使用的爬蟲不被封鎖。

頁面體驗是帶有 GEO 後果的技術 SEO 問題:訪客到不了答案,爬蟲也存不下它。把 LCP、INP 與 CLS 放進與索引、重複網址同一輪審計。