Skip to content

Suggestions

GiftWrapt has an optional layer of AI-assisted hints called Suggestions (internally: Intelligence). It looks at your own lists and items and surfaces a small, focused set of recommendations: items that look stale, items that might be duplicates, sets of items that could be grouped together, gaps a giftgiver would trip over, and lists that aren’t shaped right for an occasion that’s coming up.

This feature is opt-in and off by default. Your admin has to enable it.

Ten analyzers run, each looking for one specific thing. Several are pure heuristics with no AI involved at all.

KindWhat it spots
Primary listYou haven’t picked a primary list yet. The card gives you a picker to choose one. Pure heuristic, no AI.
List hygieneAn occasion is coming up and your lists aren’t shaped for it: no list matching the event, a matching list that’s still private, or the wrong list marked primary. The decision is rule-based; a model is only consulted to propose a new name when it suggests converting a list.
Relation labelsMother’s Day or Father’s Day is approaching and you haven’t told GiftWrapt who you shop for, so the reminders can’t fire. Heuristic.
Stale itemsItems you added a long time ago and haven’t touched - probably worth removing or refreshing.
Possible duplicatesItems that look like they refer to the same thing (“AirPods Pro” and “Apple AirPods Pro 2nd Gen”). Same-URL pairs are caught without asking a model at all.
GroupingSets of items that look like “I want one of these” or “this set in order,” and could become an item group.
Missing priceItems where you pasted a URL but never filled in a price. Heuristic.
Missing imageItems where the scraper found candidate photos but none was ever picked. Heuristic.
Stale scrapeItems with a URL that has never been scraped, or hasn’t been re-scraped in a long time, so the price and availability may have drifted. Heuristic.
Clothing detailsClothing items whose title and notes don’t pin down a size or color, which is the usual way a giftgiver ends up guessing.

Each suggestion shows up as a card with a short rationale. Most are advisory - click through and decide what to do. A few are apply-able in one click (see below).

Your admin can turn individual analyzers off, so a given deployment may surface fewer than all ten.

Every suggestion has three actions:

  • Apply (when available) - the change happens server-side without leaving the page. Grouping and stale items both ship apply actions today: “Group as Pick One” creates the item group and links the items immediately, and a stale-item card can delete the item it flagged, or roll a set of co-flagged alternatives into one group. Other cards carry a link that takes you to the right screen instead.
  • Dismiss - hide this specific suggestion. It won’t come back even when the analyzer re-runs, unless the underlying items change enough that GiftWrapt no longer recognizes it as the same suggestion.
  • Ignore - just close the card. It’ll likely reappear on the next run.

Dismissed suggestions stay dismissed across regenerations until the underlying items shift. So once you tell GiftWrapt “no, these aren’t duplicates,” that exact pair stays quiet.

By default suggestions refresh every 7 days per user, not daily. A background job works through whoever is overdue, oldest first, and you’re overdue once your last successful run is older than that interval. Your admin can change the interval.

Whether you’ve read the previous batch makes no difference. Suggestions regenerate on the interval either way, so an untouched pile of cards won’t hold up the next run.

A run is skipped outright only when the feature is switched off, when no AI provider is configured, or when a run for you is already in progress. Separately, each analyzer checks its own inputs before it calls a model: if nothing it looks at has changed since last time, it carries its existing cards forward rather than paying to regenerate the same answer. That’s per analyzer, not all-or-nothing, so an unchanged duplicates check doesn’t stop grouping from re-running.

You can also manually refresh from the Suggestions page, subject to a per-user cooldown (default 60 minutes between manual triggers).

Just your items. Specifically:

  • Item titles, notes, prices, URLs, priorities - the obvious stuff.
  • Item update timestamps - so “stale” can be a meaningful concept.
  • Item availability - unavailable items are excluded from duplicate / grouping consideration.

What the AI doesn’t see:

  • Anyone else’s items or lists.
  • Any claim data. GiftWrapt never sends gifter, claim, or purchase information to the AI. Prompts also explicitly instruct the model never to mention gifts or gifters in its output.
  • Revealed (archived) items. Those have moved past the active-list flow.
  • Gift-ideas or todo lists. Gift-ideas are spoiler surfaces for someone else; todos have no fields to analyze.

See Privacy for the full spoiler-protection rule.

The user menu has a Suggestions link (when the feature is enabled), or go directly to /suggestions. If the feature is off, the page shows a “feature offline” notice rather than a 404.

If you’re configuring this for a deployment, the admin side lives across two pages: