Braze Item Recommendations: AI or Rules-Based?

Choose Braze item recommendations by three questions: how much interaction data you have, whether you already know the logic, and how fresh each recommendation must be. AI Personalized suits large catalogs and audiences where tastes vary. Rules-based recommendations suit known patterns such as restock cycles, and they're the safer choice when a message follows a purchase within hours.

Key Takeaways

• Choose item recommendations per placement by data volume, whether you know the logic, and how fresh the item must be.
• Braze suggests AI Personalized works best with hundreds or thousands of items and typically at least 30,000 users with interaction data.
• Rules-based recommendations use fixed logic you write, which suits known patterns such as restock reminders.
• The option to exclude items a user already bought only updates weekly, or every 24 hours with Item Recommendations Pro.
• If a model fills fewer than 30 items, Braze appends most popular items.
• The same item can appear in several sections of one email, so deduplicate with Liquid.

What kinds of Braze item recommendations are there?

Every recommendation draws from a catalog, and only items in that catalog are recommended. Within that, Braze offers AI models and a rules-based approach you build yourself.

The AI models learn and adjust on their own. Rules-based recommendations do exactly what you wrote until you change them.

When should you choose AI Personalized?

Choose it when you have the data to support it. Braze suggests AI Personalized works best with hundreds or thousands of items and typically at least 30,000 users with purchase or interaction data, while noting that this is a rough guide.

Recommendations are generated for users who completed the event you configure, so pick the event that best represents interest in an item. Users without enough history receive most popular items instead, which keeps every message filled.

Two practical limits shape the setup. Selections that use Liquid can't be used in AI recommendations, and training can take between 10 minutes and 36 hours, so build that time into launch plans.

When are rules-based recommendations the better choice?

Use rules when you already know what to recommend. Braze describes rules-based recommendations as fixed logic you set manually, best for recommendations that don't need frequent updates.

Restock reminders are the clearest case. If a product has a predictable usage cycle, such as monthly vitamins, a rule based on the last purchase date is simpler to explain and test than a model.

Rules are also easier to audit. When a stakeholder asks why someone received an item, the answer is the logic you wrote, which matters in regulated or sensitive categories.

Braze's engine options differ in how fresh they are. Among the rules-based options, Connected Content is the one marked as updating recommendations in real time, at the cost of hosting and troubleshooting the data source yourself.

How fresh are AI recommendations?

AI Personalized can exclude items a user has already bought or interacted with. That option only works as well as the most recent update.

Without the Pro upgrade, updates happen weekly. With Item Recommendations Pro, they happen every 24 hours. Braze gives an example: if a user buys something and receives an email within 30 minutes, the item they just bought might not be excluded in time.

That shapes where AI recommendations belong. For an order confirmation or a same-day follow-up, use rules or exclude the purchased item with Liquid, rather than relying on the model's exclusion.

How do you add recommendations to a message?

Braze generates the Liquid for you from the Add personalization option in the composer. The pattern it produces looks like this, with field names that depend on your catalog.

{% assign items = {{product_recommendation.${recommendation_name}}} %}
{{ items[0].name }} for {{ items[0].price }}

Plan for two behaviours. If a model fills fewer than 30 items, Braze appends most popular items after the personalised ones. And because each recommendation works independently, the same item can appear in several sections of one email.

Braze's fix for duplicates is to track the item IDs you've already shown with Liquid and skip them in later sections. Build that in from the start if a message has more than one recommendation block.

How should you choose for each message?

Work through these questions for each placement, and stop at the first clear answer.

1. Does the message follow a purchase within a day? Use rules, or exclude the purchased item with Liquid.
2. Do you know the logic already, such as a usage cycle? Use rules-based recommendations.
3. Is the audience large, with enough interaction history? Use AI Personalized.
4. Is the goal to show momentum rather than individual taste? Use Trending.
5. Is the audience new or data-light? Use Most popular.

Mixing types in one message is fine. Give each section a clear job, and deduplicate across them.

Which mistakes should you avoid?

Name the mistakes so they're easy to spot. A thin-data model is AI Personalized running on too little history, where most users fall back to the same popular items.

A just-bought repeat recommends the item a customer bought minutes earlier, because the exclusion hadn't updated. A duplicate wall shows the same product in several sections of one email.

A forgotten rule keeps recommending from logic nobody has revisited since the catalog changed. Review rules-based logic whenever the catalog does.

Where does this connect to the rest of the programme?

Recommendations are one way to deliver dynamic content, as covered in seven ways to deliver dynamic content. They sit inside the wider approach in hyper-personalisation in Braze.

Post-purchase messages, where freshness matters most, are covered in automating transactional emails. The interaction events models learn from are set out in tracking user events in Braze, and product data often arrives through integrating CRM data into Braze.

As a Braze implementation partner, CustomerIK recommends choosing per placement, not per programme. Use the model where tastes vary and data is plentiful, and use rules where you already know the answer.

Recommendations after a recent purchase

For a fictional coffee shop, a receipt sent minutes after a purchase should not rely on a model refresh to exclude that item. Use a maintained compatibility rule or explicit item suppression for that placement. A weekly discovery email may be a better candidate for personalized recommendations.

Test the placement against its existing alternative with the same eligibility and observation window. Record purchases as well as clicks; do not assume a click-winning recommendation improves revenue. Give the catalog and rules a named owner.

The worked example and test cases are illustrative editorial guidance, not measured client outcomes or a claim that a live integration has been tested.

Frequently Asked Questions

1. How much data does AI Personalized need?

Braze suggests it works best with hundreds or thousands of items and typically at least 30,000 users with purchase or interaction data, as a rough guide.

2. Why do popular items appear in my AI recommendations?

If a model fills fewer than 30 items, Braze appends most popular items after the personalised ones. Users without enough history also receive most popular items.

3. Can AI recommendations exclude items a user just bought?

They can, but only as of the latest update: weekly without Item Recommendations Pro, every 24 hours with it.

4. How long does an AI recommendation take to train?

Between 10 minutes and 36 hours. Braze emails you when training finishes or fails.

5. When should I use rules-based recommendations?

When the logic is known and stable, such as restock reminders based on a usage cycle, or when a message follows a purchase closely.

Sources

• Braze, Item recommendations. Retrieved 18 Oct 2026.
• Braze, Create AI item recommendations. Retrieved 18 Oct 2026.
• Braze, Create rules-based item recommendations. Retrieved 18 Oct 2026.
• Braze, Use item recommendations in your messaging. Retrieved 18 Oct 2026.

CustomerIK is a Braze implementation partner working across onboarding, technical integration, marketing operations and customer data management.

If you're deciding where recommendations belong in your messages, let's talk.