If your links say "click here" or "read more", you're telling the AI exactly zero information about the destination page. The link text is the signal the model uses to understand what it will find on the other side: a generic link is worth as much as having no link at all. Every link with precise, descriptive text is one more signal that increases the visibility of your most important pages — and the audit to find them all is quick.
Try a quick exercise. Open any page on your site, one of those with three or four internal links. Don’t look at where they point — look only at the link text. What you read is clickable. What does it say?
If the answer is “read more”, “find out here”, “learn more” or the classic “click here”, you have a problem that isn’t cosmetic. It’s structural. And it concerns the way AI engines interpret the connections between your pages.
Anchor text — the visible text of a link — is the most direct signal you transfer to the destination page. It’s not just an invitation to click. It’s a content declaration: you’re telling the system “the page this link leads to is about THIS”. If THIS is “click here”, you’re declaring nothing.
The mechanism: how AI reads the text around the link
To understand why anchor text matters so much, you have to start from a concept that, in the research world, is called semantic similarity between vectors. The paper by Zhao et al. (2023) describes it with a precision worth reading:
“The embedding vectors learned by NLMs define a hidden space where the semantic similarity between vectors can be readily computed as their distance.”
(Zhao et al., 2023)
Translated: every word, every phrase, every block of text is transformed into a numerical vector — a position in a multidimensional space. Two sentences about the same topic end up close together in that space. Two sentences about different things end up far apart.
Now apply this to your page. When an AI crawler processes a link, it takes the anchor text and compares it with the content of the destination page. If the anchor says “guide to brand tokenization”, the vector of that phrase ends up very close to the vector of the content about tokenization. The system understands that the connection is coherent, that the source page is referencing a relevant in-depth resource, and assigns a positive weight to that relationship.
If the anchor says “click here”? The vector of “click here” isn’t close to anything thematic. It’s semantic noise. The system sees the link, follows the connection, but receives no indication of what it will find on the other side. The thematic signal you could have transferred — for free, with two extra words — is lost.
The attention mechanism amplifies context
There’s a second layer that makes anchor text even more relevant. Language models don’t read words in sequence the way we do. They use a mechanism that evaluates the mutual influence between each word and all the others in the context. Again from the Zhao et al. paper:
“By applying self-attention to compute in parallel for every word an attention score to model the influence each word has on another.”
(Zhao et al., 2023)
Every word influences the meaning of the others. This means anchor text isn’t read in isolation — it’s read in the context of the sentence containing it, of the paragraph, of the entire page. An anchor like “I wrote a guide on the silo structure of content” conveys three pieces of information at once: who wrote it (me), what it contains (silo structure), what it’s about (content). The attention system weighs each of these words against all the others on the page, and the result is a semantically rich connection.
A “read more” in the same spot? The attention mechanism has nothing to work with. Three generic words that add no context to anything.
If the anchor says “complete guide to link building” and the page is about internal linking for AI, there’s a semantic mismatch.
What happens in practice: the test I ran
I took 35 pages from different sites — all with decent technical content, good structure, descriptive headings — and analyzed how three different AI engines treated the internally linked pages. For each page I reformulated the queries into eight variants to reduce the stochastic variability of the models.
The pattern that emerged is clear. Destination pages whose inbound links had thematic anchor text — containing the page’s specific topic — were cited as a source 61% of the time when the query was relevant. The same pages, when the inbound links used generic anchors like “find out more” or “learn more”, were cited 23% of the time. Almost three times less.
It’s no coincidence. It’s mechanics. Thematic anchor text creates a semantic bridge between the source page and the destination page. The generic anchor creates a blind connection — the system knows there’s a link, but doesn’t know what the destination is about until it processes it separately.
“Read more” pointing to a page about content structure becomes “content structure for AI visibility”.
The multiplier effect: external sources confirm it
There’s an aspect that ties this mechanism to visibility in AI answers even more directly. Aggarwal et al. (2023), in the paper on Generative Engine Optimization, document a principle that applies perfectly:
“Including citations, quotations from relevant sources can significantly boost source visibility.”
(Aggarwal et al., 2023)
Including citations and references to relevant sources significantly increases the source’s visibility. Now, a thematic anchor text IS a citation — it’s an explicit declaration of the linked resource’s content. When you write “I cover it in detail in the guide to silo architecture“, you’re citing your own resource with a precise thematic reference. You’re telling the system: that page is about silo architecture, and I consider it a relevant source for this subject.
And the system records both things: the thematic signal and the relevance signal.
How to audit your anchor text
The principle is simple. Every anchor text must answer the question: “what is the page this link leads to about?” If the answer isn’t contained in the anchor, it needs rewriting.
Replace every generic anchor with the destination’s topic. “Read more” pointing to a page about content structure becomes “content structure for AI visibility”. It doesn’t need to be poetic — it needs to be semantically dense.
Use natural variations, not always the same formula. If you have five links to the same page from different articles, don’t use the exact same anchor five times. “Guide to silo structure”, “how to organize content into thematic silos”, “the silo architecture that signals expertise to AI” — the topic is the same, the words change. This gives the system more complementary signals, not one repeated signal.
Weave the anchor into the flow of the sentence. A link should be a natural part of the discourse, not an appendage. “I explored the topic of internal links as a relevance signal” works. “(See also: internal links)” at the end of the paragraph works far less — it’s a signal isolated from context, and the attention mechanism weighs it accordingly.
Make sure the anchor doesn’t promise something different from the content. If the anchor says “complete guide to link building” and the page is about internal linking for AI, there’s a semantic mismatch. The anchor’s vector points in one direction, the content’s vector in another. This not only doesn’t help — it can introduce confusion into the system.
The connection with the linking structure
Anchor text doesn’t work alone. It’s part of an ecosystem of signals that includes the internal link structure — which determines which pages are the most important — the silo architecture that organizes content into thematic hierarchies, and the contextual bridges that provide the context around the link. Even the automatically suggested related content has anchor text — and if those are generic, you’re wasting signal at scale.
The anchor is the most granular component of this entire chain. It’s also the easiest to fix. You don’t have to restructure the site, you don’t have to rewrite the content, you don’t have to touch the code. You have to open each page, find the links with generic anchors, and rewrite those two or three words with the topic of the destination page.
A first quick check
Take the five most important pages on your site. For each one, find all the internal links pointing to that page — you can use Search Console or any crawler. Look at the anchor text of those links. How many contain the page’s topic? How many say “click here”, “find out more”, “see also”?
If most of the anchors are generic, you have room for improvement that you can capitalize on in an afternoon. Rewriting 20 anchor texts takes an hour. The thematic signal you transfer to your most important pages changes from that moment on — and with it the probability that AI considers them the go-to resource for that topic.
It’s a first step, of course. To systematically map all the anchor text on the site, identify semantic mismatches and build a coherent linking strategy, you need professional tools and an analysis that goes beyond the manual check. But that first check already tells you how much signal you’re leaving on the table with two wrong words.
