colab
Internal Link Opportunities
Find internal linking opportunities with embeddings and similarity scoring instead of page-by-page review, with an LLM writing the anchor text at higher tiers.
Finding internal links by hand means reading pages and guessing at relatedness. That stops scaling somewhere around a few dozen pages, and it misses the pairs you’d never have thought to compare. So these score every pair on embedding similarity instead.
Three tiers, same clustering in all of them. What changes is what happens to the candidates afterwards.
| Notebook | What it adds | Cost |
|---|---|---|
| Rising Tide | Embeddings, clustering, a ranked CSV of candidates by similarity | Free, fully local |
| Open Water | LLM validation that a link makes sense, anchor text, rationale, priority | ~$0.01 to $0.08 per run |
| Deep Sea | No LLM call cap, Screaming Frog embeddings, OpenAI embeddings option | A few dollars |
Rising Tide hands you a ranked CSV to review yourself. Open Water and Deep Sea send the top candidates to an LLM, which throws out the pairs that only look related (near-duplicates, shared boilerplate) and writes the anchor text for you.
Same caveat as anything with an LLM in it: those two tiers send your content off-site. Check whether your client and your security team are fine with that first. Rising Tide never leaves your session, so it’s the one to run while you wait for an answer.
And a high similarity score means two pages are about the same thing. It does NOT mean linking them is a good idea. The repo’s notes on reading the output are worth the two minutes before you bulk-implement anything.