A page can be well written and still never rank, because nothing on the site leads to it, or every path stops on a broken link or a chain of redirects. Internal linking is how Googlebot, and most AI crawlers, decide a URL belongs to the site. A redirect is how you retire an old URL without dropping that path. When either one is wrong, the SEO issue is discovery, and the GEO issue is a citation that never reaches the passage you published.
This guide follows common SEO and GEO issues that hurt rankings. Use the indexing and crawl guide when Search Console says the URL was discovered or crawled but is not indexed, including soft 404s. Use duplicate content and keyword cannibalization when two live URLs still compete and one redirect should collapse them. Use this one when the links themselves are broken, loop, or never land on the page you want cited.
What fails between two pages
| Fault | What the crawler sees | SEO issue | GEO issue |
|---|---|---|---|
| Broken internal link | A 404 or a soft 404 | The destination is never stored, or the URL that ranked is gone | No passage to quote |
| Redirect chain | Several hops before the final URL | Crawl is spent on URLs that should not rank | A short fetch may stop on an intermediate URL |
| Redirect loop | The same URLs point at each other | Neither URL is a stable canonical | The fetch fails |
| Orphan page | No HTML link from an indexed page | The URL depends on the sitemap alone | Retrievers that follow links never enter it |
| Weak architecture | Every URL is linked, but only from a noisy template | Important pages look the same as filters and tags | The clearest answer is not the URL that gets fetched |
A few 404s are normal. Google has said they are not a sitewide penalty. The defect is a pattern: important URLs returning 404, menus still pointing at them, and the replacement left with no links.
Broken internal links
Start with links your own templates emit. A navigation item, a related-article card, or a breadcrumb that still uses last year’s slug sends every crawl into a dead end.
These are the cases that show up on Hong Kong sites:
- A redesigned slug with no replacement. The article moved, the old path returns 404, and the homepage still links to the old path. The new URL is an orphan. The old URL, which had the impressions, is gone.
- Language paths that do not exist. A Chinese page links to
/zh/insights/plus an English slug, or an English page links to a Chinese slug without/zh/. The reader and the crawler both get a 404. Hreflang cannot repair a link that never resolves. - Links left on a staging host. Absolute URLs pointing at a staging domain, or at
http://, force a redirect or an error on every page that includes the template. - Buttons with no
href. A control that only runs JavaScript is not a link in the raw HTML. Crawlers that do not execute the script never see the target.
Point the href at the final URL. A redirect from the old slug is the safety net for bookmarks and external links. It does not replace an updated internal link. A 200 response with an empty “not found” layout is a soft 404, which the indexing guide treats as a coverage problem rather than a link problem.
Redirect chains
A redirect chain is two or more hops before the final URL. A common stack is http to https, then non-www to www, then a trailing-slash rule, then the new slug. Each rule is reasonable alone. Together they make the destination harder to reach.
Google’s documentation says Googlebot follows a limited number of hops, up to 10, and can stop before the content if the chain is longer. A chain that long is already a fault. Use one permanent redirect, a 301 or a 308, from any retired URL straight to the current canonical. Keep temporary redirects, a 302 or a 307, for short tests. A temporary redirect left in place after a move can keep the old URL eligible longer than you intended. Google may treat a long-lived temporary redirect as permanent. Do not leave the outcome to that guess.
When you merge pages, do not chain the retired URLs through each other. Each old URL should redirect once to the survivor. Choosing the survivor is the cannibalization guide. After the redirect exists, replace internal links, the sitemap, and ads so they use the final URL. A canonical on the old URL is a hint. A link that still points at the old URL keeps asking for it to be crawled.
A loop, where A redirects to B and B redirects to A, ends the crawl. Neither URL becomes a stable result.
Use a server-side HTTP redirect. A meta refresh or a JavaScript location change is easier to miss, and some AI crawlers never run the script. If the final page is also slow, the fetch is more likely to stop early. That speed problem is the Core Web Vitals guide.
Site architecture and same-language links
Site architecture is which pages exist and which links make them look important. A flat blog with no hubs, or a store that links every facet from every page, both produce SEO issues. The first leaves useful URLs undiscovered. The second spends crawl on URLs that should not rank.
On a bilingual Hong Kong site, check each language on its own:
- The English article index links to English articles. The Chinese index links to Chinese articles.
- In-content links stay in the same language, unless the sentence is explicitly offering the other language.
- The anchor names the topic. “Read more” and “click here” waste the words that explain why the target should rank for that query.
- One hub owns each topic. A service page and a guide can both exist, but they should not target the same query. That split is keyword cannibalization.
An orphan is a URL with no ordinary HTML link from an indexed page. Sitemaps help discovery. They do not show that the page sits inside the site. The coverage state is in the indexing guide. The fix is the link: at least one descriptive, same-language link from a hub, a parent guide, or the service section the page supports.
Click depth is a warning, not the defect. Five clicks from the homepage on a small site means the architecture is hiding the URL. Zero internal links means the URL is not in the architecture at all.
Do not mark your own guides nofollow or ugc. Nofollow is a hint, and on your pages it says you do not vouch for the target. Those marks belong on links you do not control.
How link faults change AI citations
Generative engine optimization needs a passage a crawler can fetch. How to rank in AI search describes that passage. Internal links decide whether the fetch happens.
Answer engines that browse the live web start from known URLs and follow links. A 404 ends the path. A long chain may stop on an intermediate URL whose title is not the page you maintain. A Traditional Chinese prompt that only follows English links retrieves an English page, then cites it for a Chinese question. The answer looks off, and the brand looks less local.
Script-only menus have the same effect as a missing link for any crawler that stores raw HTML. Put the href in the first HTML response. Empty shells are an indexing problem. The link graph is the part you can audit from a crawl export without rendering the page.
A practical order of fixes
- Crawl both languages and list internal links that return 404, 410, or a redirect.
- For each retired URL that should still resolve, add one permanent redirect to the final canonical. Remove the hops in between.
- Replace internal hrefs, breadcrumbs, and the sitemap so they use that final URL.
- Add one same-language, descriptive link to every indexable URL that currently has none.
- Confirm there is no loop, and that Chinese links do not point at English-only paths.
Then test a few customer prompts again. If the cited URL is an old slug, the chain or the internal link is still advertising it.
Questions teams ask
Do broken links hurt Google rankings?
A few 404s do not trigger a sitewide penalty. Rankings drop when the URL that used to rank now returns 404, or when the replacement has no internal links, so it looks unimportant. Repair the important broken links and point them at the page that should rank.
How many redirects are too many?
Keep internal links on a single hop. One permanent redirect from the old URL to the current one is enough for visitors and for crawlers that still request the old address. Update the links so most crawls never hit the redirect.
Should English pages link to Chinese pages?
Use the language switcher and hreflang to connect translations. Keep in-content anchors in the same language as the page. Linking every sentence across languages mixes the two indexes and still leaves the Chinese URL looking orphaned to a Chinese crawl.
Will a redirect fix an AI citation that names the old URL?
Only after the old URL permanently redirects to the page with the current answer, and internal links no longer prefer the old URL. If the old page stays live and the new page is an orphan, systems that already stored the old URL will keep citing it.
What to fix this month
- Export internal 404s and redirecting hrefs for the English and Traditional Chinese templates.
- Collapse every chain to one permanent redirect.
- Repoint navigation, related links, and the sitemap at the final URLs.
- Add a same-language link from a hub to every orphan that should be indexed.
- Re-check prompts that cited an old or redirected URL.
Search and answer engines both follow the path you publish. Broken internal links, thin site architecture, and redirect chains are how that path stops before the page you meant to rank.