Internal links do two jobs that have nothing to do with passing authority: they let a crawler find a page at all, and they tell anything reading the link what the target page is about. Both jobs matter more when the reader is a retrieval system working from fetched pages than when it is a person clicking around. A page no internal link points to may never be fetched. A page every link calls “learn more” is described, by your own site, as nothing in particular.
SEO practice has long treated internal linking as a way to move link equity toward priority pages. That model is not wrong, but it is narrow. This post covers the other half: anchors that describe, clusters that mirror the questions people ask, pages for the sub-questions a broad query breaks into, and the orphan pages that fall out of all of it. It sits with Chunking: why passages have to stand alone, which handles structure inside a page; this post handles structure between pages. Both are part of the page-level work described in Answer Engine Optimization: a page-level guide.
What is documented about internal links?
Google’s documentation says three things about links that apply directly here. Links need to be crawlable: “Generally, Google can only crawl your link if it’s an <a> HTML element” with an href attribute. Every page that matters needs at least one: “Every page you care about should have a link from at least one other page on your site.” And anchor text should be “descriptive, reasonably concise, and relevant to the page that it’s on and to the page it links to,” with generic text such as “click here” given as the example of what not to do (Google Search Central: link best practices).
Google’s guidance on AI features names internal links explicitly. Among the ordinary practices it recommends for AI Overviews and AI Mode is “making your content easily findable through internal links on your website.” The same page states that these features may use a “query fan-out” technique, “issuing multiple related searches across subtopics and data sources” to build a response (Google Search Central: AI features and your website).
That is the whole of what is documented. Nothing published says that an answer engine reads your anchor text when selecting a passage, or that a well-linked cluster is preferred over a poorly linked one. What the documentation does establish is that discovery runs on links, and that the questions behind an AI answer are split into related sub-searches. The rest of this post follows from those two facts.
Why do descriptive anchors matter beyond PageRank?
A descriptive anchor is a short, accurate label for the target page, written by the site that knows the page best. Generic anchors waste that label. “Read more”, “here”, and “this guide” describe nothing; a link that says “how to check whether AI crawlers can reach your site” describes the target in the same words someone would use to ask about it.
There are three practical reasons to care, none of which require a claim about how any engine weighs anchors:
- It is the documented guidance. Google asks for descriptive anchor text and gives the reason as making it “easier for people and Google to make sense of your content.”
- It survives extraction. A link inside a paragraph that is lifted into an answer carries its anchor text with it. “See our guide” in an extracted passage is a dangling reference. “See our guide to robots.txt rules for AI crawlers” still says something.
- It forces you to know what the target is for. If you cannot write a specific anchor for a link, the target page may not have a specific job. That is a content finding, not a linking one.
Two details are easy to miss. Image links use the image’s alt text as their anchor, per the same Google page, so a linked logo or thumbnail with empty alt text is a link with no label. And links injected by JavaScript are crawlable when they render as real a elements, but Google’s documentation recommends confirming in the URL Inspection Tool that they appear in the rendered HTML. Crawlers that do not render scripts will not see them at all; JavaScript content and what never makes the trip covers that gap.
How should a cluster be linked?
A cluster should be linked so that its structure matches the structure of the questions it answers. The hub-and-spoke pattern does this when it is applied strictly:
| Link | Direction | Anchor should describe |
|---|---|---|
| Hub to every spoke | Down | The specific sub-question the spoke answers |
| Every spoke to the hub | Up | The broad topic the hub covers |
| Spoke to sibling spoke | Across | Only where one answer depends on the other |
| Template navigation | Everywhere | Section names, kept short |
The hub states the broad topic and answers each sub-question in a sentence or two, then links to the spoke that answers it fully. Each spoke answers one sub-question completely and links back up. Sibling links are added only where the reader genuinely needs the other answer to use this one; a spoke that links every sibling is a second navigation menu, and it adds link-heavy text without adding meaning. The boilerplate ratio post explains why link-heavy blocks are the first thing main-content extractors discount.
Put the contextual links in the body copy, next to the sentence that needs them. A link in a paragraph carries the surrounding sentence as context; a link in a footer or a related-posts rail carries none, and both are usually dropped as boilerplate.
Why give sub-questions their own pages?
A broad question is rarely answered by one page, and Google documents that its AI features split a question into related sub-searches. A site that has a clear page for each sub-question, linked from the page for the broad question, has a candidate for each of those sub-searches. A site that answers everything in one long page has one candidate, and its relevant passage has to be found inside a page about something broader. Query fan-out: why one question hits many of your pages develops that mechanism.
This is reasoning from the documented technique, not a documented preference, and it has a limit. A sub-question earns a page when it has enough substance to answer on its own. Splitting a short answer into its own URL produces a thin page, which is worse than a clear section on the parent page. The test is the one chunking uses for passages: could this answer stand alone, with the rest of the site removed? If it can, and people ask it, it can be a page. If it cannot, keep it as a section with a question heading and link to the section’s anchor.
To find the sub-questions, use the same sources as for headings: question-phrased queries in Search Console, recurring questions in support tickets and sales calls, and the questions your hub already answers in a sentence when it should answer them in a page. Question headings vs. keyword headings covers phrasing them.
How do you find orphan pages and weak anchors?
An orphan page has no internal links pointing to it. It can still be discovered through a sitemap (sitemaps in AI search covers what those still do), but it has been given no context by its own site, and Google’s documented rule is that every page you care about should have at least one internal link. Find orphans and weak anchors by hand first, then across the site.
By hand. Pick a priority page. Search your site for its URL path, or use your CMS’s “what links here” feature if it has one. Read each anchor that points to it. If the list is empty, or every anchor is “read more”, you have the finding.
In Search Console. The Links report has a “Top linked pages” table for internal links, which shows “which of your pages is linked the most from within your own site,” and selecting a URL lists the pages that link to it (Search Console Help: Links report). Tables are limited to 1,000 rows, so on a large site use it for priority pages rather than as a complete inventory.
In a crawler. A crawl that starts from the homepage and follows links finds every linked page; your sitemap lists every page you want indexed. Export both URL lists to a spreadsheet and compare them with COUNTIF: sitemap URLs the link crawl never reached are your orphans. For anchors, most crawlers export every internal link with its source, target, and anchor text. Filter that export to anchors such as “read more”, “click here”, “learn more”, “here”, and empty strings, and pivot by target URL. The result is a list of pages and the generic labels your site gives them.
From a terminal. A developer can do the same comparison by parsing the sitemap and the crawl export with a short script. It adds nothing the spreadsheet does not, except speed on very large sites.
GEOBee raises orphan pages as a site-wide finding, alongside redirect chains and duplicate titles (findings & fixes), and its crawler can filter results to the URLs listed in your sitemap (crawl settings). Each page’s detail panel has an Internal Links tab listing the links on that page. Its SEO lens scores anchor text and link count at 2 points each (the three lenses), which is a small weight on purpose: these are hygiene checks, and the structure they hint at is a judgement the scores do not make for you.
A linking checklist for one cluster
Run this on one cluster at a time. It is a by-hand list; every item can also be checked from the crawl export described above.
- The hub links to every spoke from the body copy, with an anchor that names the spoke’s question.
- Every spoke links to the hub within its first few paragraphs.
- Sibling links exist only where one answer depends on another.
- No priority page in the cluster is an orphan.
- No internal link in the cluster uses a generic anchor.
- Linked images have alt text that describes the target.
- Every link is an
aelement with anhrefin the raw HTML, visible in view-source. - Each spoke answers its question well enough to stand alone; if it does not, merge it back into the hub as a section.
What better linking does and doesn’t buy you
Clear internal links make every page discoverable by a crawler that follows links, and they label each page in words that match the question it answers. That is verifiable: run the checks above before and after the change and compare the orphan list and the anchor list.
They do not guarantee that any page is retrieved or cited. No engine documents a reward for internal link structure in its answers. What the structure does is remove two failure modes you control: pages that are never found, and pages your own site never describes.
To find orphan pages and generic anchors across a whole site in one crawl, download GEOBee. The 7-day trial includes every feature.