Skip to Content
PlatformCatalogAI suggestions

AI suggestions

AI suggestions are proactive show recommendations that appear inside any collection you’re building. While the chat assistant waits for you to ask, suggestions show up on their own — based on the shows already in the collection — and offer one-click add or dismiss.

When you see them

Open any collection. A Suggestions card appears above (or alongside, on wide screens) the shows grid, showing up to 5 AI-recommended shows that match the collection’s pattern.

The suggestions are computed by the AI looking at:

  • The shows currently in this collection (names, networks, hosts, categories, audience profile).
  • The set of shows available in the active workspace (so suggestions never cross org boundaries you can’t see).
  • Excludes shows already in the collection.
  • Excludes shows you’ve dismissed individually.

What you can do

ActionWhat happens
Add (per show)The show is added to the collection. The card refreshes with a replacement suggestion.
Add allEvery visible suggestion is added in one go.
Dismiss (per show)That show is hidden from suggestions for this collection. Won’t reappear.
Hide all suggestionsThe Suggestions card collapses. This is global — see “Hiding globally” below.
Chat to add morePre-fills the AI chat panel with “Add more shows like these to this collection”.
Refresh / regenerateA small refresh button asks the AI for a fresh set.

One-click add

Clicking Add is fast — the action runs without a confirm. The added show appears in the collection’s grid immediately, the card refreshes, and a single replacement suggestion fills the slot.

The action is not undoable from the card — to remove an accidentally-added show, select it in the grid and use Remove from Collection.

Hiding globally

The “Hide all” / dismiss-the-card action sets a single global flag in your local storage. Once dismissed:

  • The card stays hidden on every collection, not just this one.
  • A small “Reopen suggestions” affordance appears in the collection toolbar so you can bring it back.
  • The reopen button shows an accurate count of currently-available suggestions for the current collection.

This is intentional — suggestions are useful but visually loud, so once you’re heads-down on a list, dismissing once shuts them up across the app until you opt back in.

Caching & remembered state

A few details that affect how suggestions feel:

  • First open of a collection runs the suggestion generation — this takes a couple of seconds.
  • Results are cached server-side per collection so subsequent opens are instant.
  • The cache invalidates when shows are added or removed from the collection (so the suggestions stay in sync with the current contents).
  • The last-seen suggestion list for each collection is also stored client-side, so the reopen affordance can show an accurate count without waiting for a fresh generation.
  • Individual dismissals are persisted server-side so the same show won’t suggest itself again, even from a different machine.

Best-fit collections

Suggestions work best on collections with 3+ shows already in them — the AI needs a few examples to infer a pattern. On a brand-new empty collection, the suggestion card either shows generic starter suggestions or stays empty.

The card also doesn’t appear on:

  • The unscoped All Shows view (no collection context)
  • The home grid (no collection context)
  • Shared collections you only have view permission on (you can’t add to it anyway)
  • Empty collections without enough context

Edge cases & gotchas

  • Slow first generation — the first open of a collection waits for the AI. The card shows a skeleton state for a couple of seconds.
  • Duplicate suggestions across sessions — if you dismissed the card globally and reopened it on a new machine, the dismissal flag is local-only. You may see the card again until you re-dismiss.
  • “Add all” then immediately Add another — the AI regenerates suggestions after each add, but the gap is small. Rapid-fire clicks can momentarily show the previous list.
  • PPN-only or scope-gated content — suggestions respect RLS. Shows you can’t see won’t be suggested. If a suggestion fails to add (e.g. the show was deleted between generation and add), a toast explains.
  • Cost — the AI runs each time the cache invalidates. For very large collections (hundreds of shows) the model context bloats; suggestions remain accurate but generation is slower.
  • Not personal — suggestions are derived from the collection’s contents, not from your individual history. Two users opening the same collection see the same suggestions (modulo their personal dismissals).
Last updated on