How to Control Which Products Get Recommended (Filters and Exclusions)
A great recommendation engine surfaces products customers actually want. A bad one surfaces a gift card next to a $200 jacket, or a clearance item that is one size away from selling out. The difference usually is not the algorithm - it is whether you have told it what to leave out. Product recommendation filters are how you draw that line, removing the items that should never appear so the engine only ever picks from a clean, relevant pool.
Most stores spend their energy making recommendations smarter. Just as valuable, and far easier, is making them cleaner. This guide walks through using hard filters and exclusions to keep the wrong products out, with concrete scenarios across collection, tag, vendor, and product type.
Why Product Recommendation Filters Beat Hoping the Algorithm Figures It Out
Recommendation engines learn from your orders and catalog. They are good at spotting what gets bought together, but they have no opinion about your margins, your brand strategy, or which items you are quietly clearing out. To the engine, a gift card that frequently appears in carts looks like a strong recommendation. To you, it is a wasted slot.
There are two ways to influence what an engine shows:
- Filters are hard rules. They remove products that do not qualify, full stop. A filtered-out product simply cannot appear, no matter how strong its purchase signal is.
- Boosts are soft rules. They re-rank products up or down without removing anything, useful when you want to nudge a category higher but still allow everything else.
This article is about filters and exclusions, the hard side. Reach for a filter when a product must never appear, and a boost when you only want it to appear less often. Our merchandising rules for product recommendations overview covers how the two work together.
The Products Most Stores Should Filter Out First
Before you get clever, clear the obvious noise. These are the categories that almost never belong in a recommendation slot:
- Gift cards. They show up in orders constantly, so the engine loves them, but recommending a gift card as an add-on rarely converts and looks careless.
- Out-of-stock-prone items. Products that hover near zero inventory create a bad experience when a customer clicks through to a sold-out page.
- Clearance and end-of-life stock. Useful for a dedicated sale section, distracting as an upsell next to full-price items.
- Low-margin filler. If a product barely breaks even, giving it premium recommendation real estate works against you.
- Off-brand or third-party items you stock but do not want to promote alongside your hero products.
- Irrelevant accessories that technically sell but pair poorly with most carts.
Each of these maps cleanly to a filter. The trick is choosing the right attribute.
Include vs Exclude: The Core Logic
Every filter works in one of two directions, and getting this distinction right is most of the battle.
Exclude logic removes a defined set and recommends from everything else. This is the right default for the cleanup list above. You are saying "recommend anything except these." Exclude a clearance tag, exclude the gift card product type, exclude a specific vendor.
Include logic does the opposite. It throws away everything except a defined set, so the engine may only recommend from that pool. This is far more restrictive and is the right tool when you want tight control - for example, "only ever recommend from the Bestsellers collection."
A simple rule of thumb: use exclude to remove a known list of problems, and include when you want to whitelist a small, curated source. Reaching for include when you only meant to remove one bad apple will silently starve your recommendations, because everything outside your included set disappears too.
Filtering by Collection
Collections are the most intuitive attribute to filter on because you have already done the grouping work in Shopify.
Exclude a collection to keep a whole category out. If you run a Clearance collection, excluding it means no clearance product can ever be recommended, regardless of how often it sells. Same logic for a Wholesale or Sample collection that should never reach retail shoppers.
Include a collection to constrain recommendations to a curated source. A store with a tightly merchandised Recommended Add-Ons collection can include only that collection, turning the engine into a smart re-ranker over products you have already vetted. You get machine-learning ordering with human-controlled inventory. We go deeper on this pattern in collection-scoped product recommendations.
Filtering by Tag
Tags are the most flexible filter because you control them entirely. They are ideal for cross-cutting attributes that do not map neatly to a single collection.
Add a tag like no-upsell or exclude-from-recs to any product you want held back, then create an exclude filter on that tag. Now keeping a product out of recommendations is a one-tag edit on the product page, no engine reconfiguration required. This is the cleanest way to manage a moving list of exceptions.
Tags also work for positive curation. Tag your proven add-on performers recommend and include only that tag when you want maximum control. For a fuller treatment of tag-based and vendor-based curation, see using product tags and vendors to curate recommendations.
Filtering by Vendor
Vendor filtering is the fastest way to manage brand-level decisions. If you carry several brands but only want to actively promote your own private label, an include filter on your house vendor keeps every recommendation on-brand. If a particular supplier has thin margins or unreliable stock, an exclude filter on that vendor removes their entire catalog from consideration in one move.
This is far less tedious than tagging hundreds of individual products. When the decision is "this whole brand, yes or no," vendor is the right lever.
Filtering by Product Type
Product type is the natural home for structural exclusions. Filtering out the Gift Card product type is the single most common filter stores set, and it should usually be your first. Other useful type filters include excluding Warranty, Shipping Protection, or Service types - items that exist in your catalog but make no sense as a product recommendation.
Because product type is a built-in Shopify field, these filters are durable. New gift card products you add later are caught automatically as long as they carry the right type.
Putting It Together: Worked Scenarios
Filters are most effective when combined. A few realistic setups:
Apparel store clearing seasonal stock. Exclude the Clearance collection and the Gift Card product type. Recommendations now pull only from current full-price inventory, and the end-of-season rack stays out of the upsell flow.
Multi-brand marketplace protecting margin. Exclude a low-margin vendor and a dropship tag. The engine still has hundreds of products to choose from, just not the ones that cost you money to promote.
Curated boutique wanting tight control. Include only a Recommended Add-Ons collection and add a price ratio rule so recommendations stay proportionate to the cart item. The engine becomes a precise re-ranker over a hand-picked shelf.
You can combine conditions inside one rule with And or Or, so a single rule can whitelist a collection and still exclude a stray tag inside it, and you can bracket separate rules into groups when you need mixed logic like "(A and B) or C".
Where EliteCart Fits
In EliteCart, this is exactly what Fine-tune EliteAI™ Ultra is built for. You start from one of two base engines - frequently-bought-together or cross-category complements - and then layer your own filters on top. You can filter by price, price-to-cart-item ratio, tag, vendor, product type, collection, or title, and each filter supports include or exclude logic. The setup is documented in the Filters and Boosts reference and the fine-tuning setup guide.
Your fine-tuned version trains on your own orders and catalog, and you assign it per surface, so your product page, cart, and checkout can each run different rules. See our EliteAI fine-tuning announcement for the full release.
If you only need to pull a handful of products out of recommendations and do not need a full fine-tuned engine, there is a simpler path: EliteCart also lets you exclude individual products from EliteAI entirely with a single setting.
Start with exclusions, not inclusions. Pull out gift cards, clearance, and low-margin filler first, then watch how much cleaner your recommendations get before you reach for anything more advanced. The wrong products doing nothing is worse than the right products doing more - and a few well-chosen product recommendation filters fix the first problem in minutes.