The AI Search Assistant on Search Configuration hands your search set-up to the Configuration Agent. It changes what the search page shows and finds — the tabs, the filters in the sidebar, the result cards, the sort order and dates, and the categories the tabs are built from — from a description in plain words.
This is separate from Catalog Search on the same screen, which configures which filters AI catalog agents may use, refreshes allowlists, and tests structured search_catalog() plans. See Catalog Search.
The assistant is evidence-driven. Before it proposes anything it checks what is actually indexed for your site, what your pages expose that a filter could be built from, and what each tab currently returns. It then presents a plan for you to approve, makes the edits, checks them, and asks for your approval again before anything goes live.
Opening the assistant
- Save or discard any unsaved edits on the Search Configuration page — the assistant works from the saved configuration.
- Click AI Search Assistant at the top of the page. The agent opens in a side panel with three starting points:
- Describe a change — "add a Campus filter to the Courses tab", "show news newest first", "rename Programs to Courses", "show the CRICOS code on course cards".
- Fix an empty or wrong tab — "the Events tab shows nothing", "the News tab is full of course pages". These are almost always a category-mapping problem, and the assistant finds which one.
- Design search for our website — the assistant looks at what has been crawled, your sitemap and navigation, and your current search page if you have one, then proposes a set of tabs and filters in plain language and waits for your agreement before planning any change.
- You can also ask for search changes from any admin page through the agent button in the header; the same skill answers.
What the assistant can change
| Area |
Examples |
| Tabs (scopes) |
Add, rename, hide or reorder tabs; change which filters each tab shows; set the default tab and sort |
| Filters (facets) |
Enable a filter on a field that has usable values; cap how many values show; remove filters that duplicate the tabs |
| Fields |
Add a field extracted from page metadata, JSON-LD or a stable part of the page, when the assistant has seen it on a real page |
| Result cards |
Choose a layout per tab, add badges (dates, type, location) and fact rows, show a summary |
| Sort and dates |
Newest-first for news, next-event-first for events, date presets on a tab |
| Categories (page types) |
Add or reorder the URL rules in the crawl configuration that decide which tab a page belongs to |
| Search UI settings |
Theme, Ask tab, AI answer, chat agent and backend API flags held in the descriptor |
The assistant is conservative by design:
- It edits only the setting your request needs and leaves the rest of the configuration as it is.
- It never adds a filter it cannot fill: a new filter needs a field the crawler can extract from your pages (the assistant inspects a real page of that type first), or one that is already indexed with a sensible number of values.
- A tab is only added when the category behind it exists or is added in the same change. Validation blocks a tab that would be empty because nothing produces its category.
- It keeps the tab strip to a handful of visitor journeys rather than a list of every category, and it will not put the same split in the sidebar as a filter.
- It does not change how the search page looks (colours, fonts) — that is the AI Style Assistant on UI Design — or what the chat assistant says.
Live now, or after re-processing?
The assistant tells you which in its plan and again when it finishes:
| Change |
Takes effect |
| Tab labels, which filters a tab shows, filter limits, sort order, date presets, result cards, search UI settings |
As soon as the change is applied |
| A new field, a changed extraction rule, a value derived from the URL, a new or changed category rule |
After the website content is re-processed |
When a change needs re-processing, the assistant quotes how long it will take and offers to start it — it never starts a rebuild on its own. You are emailed when the job finishes. Until then the new filter or field stays empty.
How a change is checked
Before the assistant presents a change it:
- confirms the values a filter will show are the ones actually indexed for your site;
- inspects a real page of the relevant type for the exact field a new filter or badge is built from;
- cross-checks every tab against the categories that produce it, so a tab cannot be created empty;
- runs the changed tab through the live search API and confirms the counts, the filters shown and the result card;
- compiles the whole descriptor, exactly as Save does on this page.
Each proposed edit appears line by line in the panel's Changes tab, where you can keep or drop it before validation and apply. Nothing is written until you approve the apply; after that the Search Configuration page reloads with the saved configuration.
Good requests
- "Add a Study level filter to the Courses tab if the course pages carry it."
- "The News tab should show the newest stories first and badge the published date."
- "Rename Programs to Courses everywhere."
- "The Events tab is empty — find out why and fix it."
- "Design the search for our website: propose the tabs and filters and check them with me first."
Requests that belong elsewhere:
- "Make the search page match our brand colours" — use the AI Style Assistant on UI Design.
- "Crawl these extra pages" — see Crawler Settings.