Recommend more like this

With the catalogue embedded, recommendations are one call: "give me the records most like this one". Then a small extra step — recording clicks — lets the list learn from real choices.

Ask for the neighbours

Point the related lookup at any record and it returns the closest others, ranked by similarity — the "you may also like" strip you'd show under a product or article:

GET https://api.searchstack.dev/related/Demo/books/1/{record-id}?size=6
X-API-Key: {your key}

{
  "results": [
    { "name": "North Light",      "@text_score": 0.91 },
    { "name": "The Salt Path Home","@text_score": 0.88 }
  ]
}

Let clicks sharpen it

Similarity is a good start, but what people actually pick is better. When a visitor clicks a recommendation, record that click against the query. Over time the results lean toward what readers really choose next — the recommendations improve themselves from real behaviour, with no model to train by hand.

That's the difference between a static "similar items" box and one that gets better the more it's used — and it's the same click-recording the box at the top of this lesson does as you browse.

That's the lesson. An embedding model gives you "more like this" for free from your own catalogue, and recording clicks turns it into a recommender that learns what people actually pick.

Go deeper: Related results and recording clicks in the reference.