Short answer, as of September 2026: internal links still matter for AI search, but for two reasons rather than one — they keep answer-bearing pages discoverable, and they supply the context that lets a passage be understood on its own. The second reason is the one classic internal-linking advice never had to consider, and it changes which links are worth adding.
Traditional guidance optimised for crawl depth and the distribution of link equity. Those still apply, because discovery has not stopped mattering. What is new is that a retrieval system frequently works with a portion of your page rather than the whole thing, and a portion that depends on context you put three sections earlier is a portion that cannot safely be quoted.
Two jobs, not one
Job one: reachability. A page nothing links to is a page that gets crawled late, crawled rarely, or not at all. That has always been true and it is unchanged. If a page is not in the index, it is not a candidate for any answer, regardless of how good it is. Orphan pages are the most common and most fixable version of this.
Job two: self-containment. When an assistant uses your content, it may work with a section rather than the document. If that section says "as described above, the process takes four to six weeks", the sentence is unusable on its own — the reader does not know what process, and neither does the engine. A link, or a restated noun, repairs it.
Job two is why the most useful internal-linking change most sites can make is not adding more links. It is rewriting the sentences around the links so each answer-bearing passage stands up alone.
What actually helps
Descriptive anchors that name the thing. "Read more" and "click here" tell a retrieval system nothing about the destination. An anchor that names the topic — the way the links in this article do — is simultaneously a navigational aid and a statement about what the target page is about. That is the cheapest form of context you can add.
Links placed inside the explanation, not in a footer block. A related-posts row at the bottom of every page is boilerplate: it appears identically on every page, so it says nothing specific about this one. A link inside a paragraph that explains why you would want the other page carries real information about the relationship between the two.
Hub pages that state the relationships. A page that gathers a topic and describes how the parts relate does three useful things: it gives the cluster an entry point, it creates a page whose subject is the topic itself, and it supplies a plain-language map an engine can use to associate your pages with each other. A bare list of links does much less of this than a short paragraph on each.
Restating the entity near answer-bearing claims. In the passage that answers the main question, name the subject rather than relying on a pronoun or an earlier heading. This is the single highest-value edit for citation, because it converts a dependent paragraph into a quotable one. It is also the core of the retrofit method in retrofitting existing pages for citation.
What does very little
High link counts for their own sake. Adding forty internal links to a page does not make it more retrievable; it dilutes the signal each link carries and makes the page harder to read. There is no target number worth chasing.
Exact-match anchor repetition. Using the identical keyword anchor for every link to a page is a pattern from an older optimisation era. Vary the anchor naturally — the target page's topic is established by many signals, not by anchor uniformity.
Footer and sidebar link farms. Site-wide blocks are treated as navigation furniture. They help reachability a little and context almost not at all.
Links inside headings or FAQ answers. These fragment the exact passages most likely to be extracted. Keep headings clean and keep FAQ answers self-contained prose.
| Pattern | Helps reachability | Helps retrieval context | Verdict |
|---|---|---|---|
| Descriptive in-body link | Yes | Yes | The default; use this |
| Hub page with explanatory text | Yes | Yes | Best structural investment |
| "Read more" / "click here" anchor | Yes | No | Rewrite the anchor |
| Site-wide footer link block | Slightly | No | Fine as navigation, not a strategy |
| Related-posts boilerplate row | Slightly | Barely | Low value; not harmful |
| Link inside an H2 or FAQ answer | Yes | Negative | Avoid; it fragments quotable text |
| Forty links on one page | Neutral | Negative | Dilutes and distracts |
Find your orphans before you add anything
Most sites have pages that nothing links to, and the owners are usually surprised by which ones. Old landing pages, posts published outside the normal flow, pages that lost their link when a hub was redesigned.
You can find them with a crawl: crawl from the homepage, list every URL discovered, then compare that list against your sitemap and your CMS. Anything in the sitemap but not reachable by crawling is orphaned. That single comparison is usually the highest-return hour in an internal-linking project, because it converts pages that cannot be selected into pages that can.
Two adjacent checks are worth running at the same time. If your navigation or links are injected client-side, a retrieval system may see none of them — the failure mode is covered in JavaScript rendering and AI visibility. And if your structured data disagrees with your link structure about how pages relate, tidy that too; the conventions are in structured data and llms.txt.
The migration case, where this breaks hardest
Site migrations and redesigns are where internal link structure silently degrades. A rebuild moves a hub, drops a section, or replaces contextual links with a generic related-posts widget. The pages still exist and still render, so nothing looks broken, but the connective tissue that made a passage understandable and a page reachable has gone.
This is one of the mechanisms behind citations disappearing after a release with no content change — a pattern documented in citation decay, and worth planning for explicitly using the checklist in protecting AI visibility through a site migration.
The defensive habit is simple: before a redesign, record which pages are cited and what links into them; after, diff it.
For multi-location and multi-service sites
Sites with many near-identical pages have a specific version of this problem. Location and service pages often link only upward to a hub and never to each other, which leaves each one isolated, and their similarity means nothing distinguishes them in context.
Two moves help. Link related service pages to each other where the relationship is genuine — the service somebody actually needs next, not every service you offer. And make sure each page's answer-bearing passage names its own place and service explicitly, since that is frequently the only thing distinguishing one from its forty siblings. The architecture that supports this is in the multi-location SEO playbook.
A short working checklist
Find and fix orphans first. Rewrite vague anchors to name their destination. Move links out of headings and FAQ answers. Add a hub page with explanatory prose for each real topic cluster. Then go through your top pages and make the passage that answers the main question self-contained — name the subject, restate the qualifier, remove the "as mentioned above".
That last step is the one that most directly changes whether a passage can be quoted, and it requires no new pages at all. The features overview shows how site-side checks and content scoring fit alongside visibility tracking, so the structural work and the measurement sit in the same place.
Frequently asked questions
Do internal links matter for AI search? Yes, for two reasons. They keep pages discoverable so a page can be a candidate at all, and descriptive in-body links supply context that helps a passage be understood on its own. The second job is new relative to classic internal-linking advice.
How many internal links should a page have? There is no useful target number. Adding links for volume dilutes the signal each carries and makes the page harder to read. Add a link where it genuinely explains a relationship a reader would want to follow.
Should I put internal links in headings or FAQ answers? No. Both fragment the exact passages most likely to be extracted and quoted. Keep headings clean and keep FAQ answers as self-contained prose, and place links in the explanatory body text instead.
What is an orphan page and why does it matter? A page nothing on your site links to. It gets crawled late or not at all, and a page that is not indexed cannot be selected for any answer. Find them by crawling from the homepage and comparing the discovered URLs against your sitemap.
Do related-posts widgets help AI visibility? Barely. Site-wide or templated blocks appear identically across pages, so they say nothing specific about the page they are on. They are acceptable navigation but are not an internal-linking strategy.
Why did my citations drop after a redesign with no content changes? Redesigns frequently replace contextual links with generic widgets, move hubs, or render navigation client-side. The pages still exist, but the connective tissue that made passages understandable and pages reachable is gone. Record which pages are cited and what links to them before a rebuild, then diff after.
Does anchor text still matter? Descriptive anchors help, because they tell both readers and retrieval systems what the destination covers. Repeating one exact-match keyword anchor everywhere does not help and belongs to an older optimisation era.
What is the single highest-value internal linking change? Making the passage that answers your main question self-contained: name the subject explicitly, restate the qualifier, and remove references like as mentioned above. It converts a dependent paragraph into a quotable one and requires no new pages.