Write descriptive anchor text
Google's own guidance is that good anchor text is descriptive and relevant to both pages. "Click here" fails that twice. So does repeating the identical exact-match phrase on every link to the same destination, because the surrounding sentence stops carrying any information.
The accessibility requirement points the same way. WCAG success criterion 2.4.4, Link Purpose (In Context), requires that the purpose of each link be determinable from the link text or its programmatic context, because screen reader users navigate by pulling up a list of links. GOV.UK's content guidance puts the practical version better than most style guides: front-load the relevant terms and avoid generic text like "click here" or "more."
So instead of "read our report," write "our 2024 review of sanitation spending in Malawi." Instead of "learn more about our work," write "how the community water committees are funded." The anchor tells the reader where they are going, and it tells a retrieval system what the destination covers. That is the cheapest part of AI-focused internal linking and the part most often skipped.
Audit existing links
You don't need a budget for this. A crawler and a spreadsheet will surface almost everything that matters on a site of a few hundred pages. Screaming Frog classifies any URL with no observed linking path from the crawl start point as an orphan, and you can spot them in the Internal tab by filtering for a blank crawl depth.
Work through the findings in this order, because the first two cost you the most:
-
Orphan pages that matter, especially research and impact content with no inbound link anywhere
-
Broken links and redirect chains, which waste crawl attention that is already thin on small sites
-
Ambiguous anchors, starting with anything that says "here," "this," or "more"
-
Clusters where the hub links down but nothing links back up
-
Pages carrying so many links that none of them reads as a deliberate choice
Record what you changed and when. An AI-focused internal linking audit is worth repeating twice a year, and the second pass is far faster when you can see what the first one touched. Entity relationship mapping notes from your original inventory make the re-check quicker still, because you already wrote down which connections were supposed to exist.
Avoid common mistakes
Linking by keyword alone is the first trap, and it produces pages that look connected while telling a machine nothing about how the subjects relate. The second is forcing a link into a sentence that was not going there, which readers notice immediately and which damages the copy more than the missing link ever did.
Overloading is the third. Analysis of Deep Research session logs found that pages with fewer than 20 links made up about 78% of what the agent read in full on its first pass. The figure fell to 33% for pages carrying 60 or more. Relying only on the main menu is the fourth, because navigation describes your org chart rather than your subject relationships, and a content topology strategy exists precisely to fix that gap for AI-focused internal linking.
Then there are the quieter failures. Circular journeys where three pages link to each other and nowhere else. Icon-only or empty links that leave screen reader users with nothing to go on. Automated linking plugins deployed without an editor reading the output, which is how a page about child protection ends up linking to a donation form on the word "support." NTEN's guidance on nonprofit traffic in an AI-first search world is blunt about the squeeze small teams are under, and automation without review adds cleanup work rather than removing it.
AI-focused internal linking checklist
Run through this before you call a cluster finished:
-
Every page has a defined role, either foundational or supporting
-
Relationships are documented in words
-
Anchor text describes the destination and makes sense read on its own
-
Links are server-rendered <a> elements with real href values
-
Each cluster has a hub, and the hub links out to all of its supporting pages
-
No page you care about is orphaned or more than three clicks deep
-
Link text meets WCAG 2.4.4, which also covers image and icon links
-
One named person owns the link map, and a review sits in the calendar twice a year
Your entity relationship mapping notes are the source of truth for the second item, so keep that file somewhere the whole team can reach it.
Snoika Foundation works with NGOs and nonprofits to make their content legible to ChatGPT and Perplexity, which starts with the structural work described here. If your research and programme pages are doing good work that nothing else on your site points to, book a call with our team and we'll walk your site's structure with you and show you where AI-focused internal linking is missing.