Limiting or customising the content used for Algolia search?

[jimjam-slam]

Description

I'm looking at ways to customise the search behaviour. I'm generally happy with the defaults, but I'd like to cut the terms of the search down to really just page titles. The reason is that our site has a bunch of dynamically generated pages that represent different places, and those pages have a lot of boilerplate text (I run a script before site render to generate the place pages). These are generated as:

places/
  canberra/
    index.qmd
  sydney/
    index.qmd
  melbourne/
    index.qmd
[...]

Where each index.qmd will have a pagetitle: Melbourne or some such.

I don't really want to pollute the search with all of this boilerplate content; I'd really prefer to just be able to have the search happen on these page titles. I could also implement the search as an OJS control, but I'd really like to put the search in the site nav, so I'm hoping I can just modify the existing search functionality.

I've been going through the current docs on this:

• Using Algolia (Guide > Websites > Website Search)
• Algolia options (Reference > Websites)

The index-fields key looks to me like it might be designed to allow you to modify where Quarto pulls its search content from. I can see that its keys match the keys of the objects in _site/search.json (eg. index-field.text: body implies to me that the body of the document is considered the content to display, and filter on, in the text of a search result, below the title).

If I understand this correctly, how can I modify this to use nothing for the search text, and to only show the title? Or to use some other metadata field for the body? For example, if I added the root metadata key stationid (eg. stationid: 123456) to my generated index.qmd docs, could I use index-field.text: stationid instead? Or null or similar to have it display nothing?

Alternatively (and this is the nuclear option), could I simply overwrite _search.json using a post-render script myself?

[jimjam-slam]

Can confirm that replacing _search.json post-render works wonderfully! For example, writing out a JSON file like this:

[
 {
    objectID: "009021",
    href: "places/perth",
    title: "Perth (009021)",
    section: null,
    text: null
  },
  {
    objectID: "004032",
    href: "places/port-hedland",
    title: "Port Hedland (004032)",
    section: null,
    text: null
  }
]

Gives you a search like this (obvs there's an extra entry in this screenshot):

And links correctly 😊