Two models put you in the driver's seat. Hand-Picked replaces a model's output with your own picks for one anchor product. Custom Blend (Pro plan and up) is a model you define yourself.
Hand-Picked: your picks per anchor product
A curation is your own list of recommendations for one anchor product. When a shopper views that product, your picks replace what the model would have shown. The picks show first, in the order you set.
A curation overrides these models when they run for your anchor product:
- Similar Products
- Complete the Look
- Smart Mix
Curations don't change Frequently Bought Together. To choose bundle partners, pin them on the Pairs tab of the Bundles page instead. See Bundles & bundle discounts.
Three rules keep curations predictable:
- When a widget needs more items than you picked, the model fills the remaining slots.
- Your "never recommend" tags remove any excluded pick before your curation serves. If every pick carries an excluded tag, the model serves instead of your curation. See Model settings that matter.
- Analytics report these results as Hand-Picked, so the model tables stay honest.
Create a curation
- Open the PersonalizerAI dashboard.
- Go to Recommendations.
- Open the Manual Curations tab.
- Select the anchor product under Source product.
- Add products under Recommended products.
- Drag the products into the order you want.
- Click Save curation.

A curation in progress: one anchor product, hand-picked recommendations in your order.
Edit a curation
Saved curations appear under Existing curations on the same tab.
- Click Edit on the curation's row. Selecting the same product under Source product does the same.
- The form loads your saved picks in their order. Add, remove, or drag picks.
- Click Save changes.
The list you save replaces the old one. If a saved pick is no longer in your catalog, the form says so, and saving drops it.
Remove a curation
- Find the curation's row under Existing curations.
- Click the remove button (the trash icon) on the row.
- Click Remove curation to confirm.

Every saved curation is listed with its pick count and a remove button.
Custom Blend (Pro plan and up)
A custom blend is a recommendation model you define. You describe which products it starts from, which products it may show, and how it matches them. Use it for logic only you know: "show our own brand next to resale items", or "pair every espresso machine with our grinders".
A custom blend has three parts:
- A source filter — the anchor products it applies to.
- A target filter — the products it may recommend.
- A similarity method — how it ranks the targets: by co-purchase, by popularity, or (Custom) by embedding similarity.
Your plan sets how many custom blends can be active:
Plan | Active custom blends | AI similarity methods |
|---|---|---|
Pro | 5 | — |
Scale | 10 | — |
Custom | Unlimited | Included |
Create a custom blend
- Open the Custom Blends tab of the Recommendations screen.
- Click Create Custom Blend. The builder has five steps: Name & intent, Source, Target, Similarity, and Preview.
- Type a Blend name (required). You can also describe the blend in plain words and click Pre-fill with AI — the app fills the next steps for you.
- Set the source filter: which products the blend starts from.
- Set the target filter: which products the blend may show.
- Select a similarity method: Co-purchase frequency, Popularity ranking, or Embedding similarity. Embedding similarity shows Custom plan only on other plans. You can also set Results per source product.
- Check the preview. It shows sample recommendations for three of your source products.
- Click Publish blend, or click Save as draft to finish later.
Next checks each step: a type or tag filter needs at least one value, and a price range needs a minimum or maximum. If you close the builder with unsaved changes, it asks first. If publishing fails after the blend was saved, click Publish blend again — the app publishes the same blend and does not create a second one.
After you publish, the app computes the blend. The status moves from Queued to Computing to Validating to Active. This usually completes in less than a minute. The list updates on its own. If the status shows Error, read the message under the status, adjust the blend, and click Retry.

The Custom Blends tab: each blend shows its slug, method, status, and actions.
Put a custom blend on your store
- Copy the blend's slug: click Copy next to the slug under the blend's name on the Custom Blends tab.
- In the Shopify theme editor, add or select a Recommendations block on a product page. You can also use the Cart recommendations block, or the cart drawer settings in the PersonalizerAI app embed.
- Paste the slug into the Custom model slug (optional) field. The field overrides the model dropdown.
An Active blend shows only where a theme block uses its slug. A custom blend computes its recommendations per product. On a product page it shows that product's picks. On the cart, it combines the picks for every cart product that is inside the blend's source filter. On pages without a product or a cart, the widget shows the fallback models instead.
Pause, edit, or remove a custom blend
- Click Pause to stop an active blend. The widget shows the fallback models until you click Activate again.
- An edit that changes the source filter, the target filter, or the similarity method returns an active blend to Draft. The builder warns you when your edit does this, and asks before Save as draft. Publish it again to apply the change. The widget shows the fallback models until the publish completes.
- A name-only change keeps the blend active. The blend's slug never changes — the app sets it once, when you create the blend.
- Click Delete to remove a blend, then confirm. The confirmation names the blend and its slug. Remove its slug from the theme editor too.
- The app refreshes every active blend's recommendations about once a day, on its own.
When you move to a lower plan, the app warns you first. Custom blends above the new plan's limit pause on their own — the app deletes nothing. Blends that use an AI similarity method pause when you leave the Custom plan. See Plans, billing & limits.
Related pages
- Models: product relationships — the models a curation can override.
- Model settings that matter — the "never recommend" tags that also filter your picks.
- Merchandising rules — pin, boost, and hide across search and recommendations.
- Plans, billing & limits — what pauses on a downgrade.