Recommendation Algorithms — What Each One Picks

Every recommendation widget is a frame. What goes in it is decided per visitor, at the moment it renders, by the algorithm you pick. This page is every algorithm the picker offers today: what it picks, what it is anchored to, which pages it works on, and what it needs before it can return anything.

On personyze.com: Every recommendation algorithm, page by page →

Four videos go deeper into what the algorithms pick and how to shape it:

Video: Product recommendations: what to show each visitor (2:02) — it plays at the top of Recommendations — Where to Start.

On this page

1. How an algorithm thinks: the crowd, times this visitor
2. Pick the page first
3. Reading a card
4. Product algorithms — the 25 cards in five groups
5. Content recommendations — the 20 cards for articles
6. After you pick: output, filters and fine-tuning
7. Looking for an older name?

How an algorithm thinks: the crowd, times this visitor

Most algorithms are a hybrid: they combine crowd data — what everyone does on your site — with this visitor’s own interest. Most Popular with Taken from set to from their latest reported interest is the most-viewed items (the crowd) in the interest this visitor showed last (the individual). Change either half and you have a different recommendation.

• Crowd-only cards — Most Popular, Best Sellers and Trending Now on their site-wide setting — work on a visitor’s very first page view. That is what makes them the usual fallback.
• Visitor History cards are lists, not models. Buy it Again, Recently Viewed and Items in Cart show what this person did, so they show nothing to someone who has not done it yet.
• When the visitor’s half is empty — no interest reported, no product viewed — a card that depends on it has nothing to return, and your fallback serves. Give every personal card a crowd fallback.

Pick the page first — it changes the algorithm, not just the placement

The picker asks where the widget will appear before it shows a single card: Home, Category, Product, Cart, Thank you, Search results for products; Home Page, Category Page, Article Page, Search Page for content. There is no 404 tab: for an error page, pick Home and target the error page in the campaign.

The tab is not a filter on where the widget shows — the Target step decides that. Behavioural algorithms anchor to something different on each page: the product being viewed on a Product page, everything recently viewed on Home, the cart contents on Cart. The same card on two pages is two different questions.

When the anchor is an event on the page — a product view, an add-to-cart, an article read or comment, a search — the widget can only show something where that event happened. The editor says so in a yellow strip (“…on pages where no product is known the widget renders nothing”) and sets the widget’s Restrict to page group for you: Product viewed on this page, Product added to cart on this page, Article viewed on this page, Article commented on this page or Searched on this site. Pick another group, or All pages, to run it everywhere — where there is nothing to recommend from, only its fallbacks fill it.

Starter recipes

A safe first choice per page, and a fallback that needs no history. Content recipes are further down.

Reading a card

Product algorithms

Twenty-five cards in five groups. Anchored to is the badge the card shows on each tab; a dash means it has none. Pages are the tabs the card is offered on.

Personalized & Behavioral

Good defaults for home and category pages, where the visitor’s intent is not clear yet.

Cross-Sells, Co-Views & Upsells

Relations between items — what goes with what, what people looked at together. The co-view and cross-sell cards share two menus: Prefer most (strongest match, transactions, users, recommended + bought) and Require at least, a minimum count behind each pair. Several also carry Recommend: any price, cheaper or more expensive than the viewed item, or within a price difference of it.

Visitor History

This person’s own past. Best for returning visitors and post-purchase pages.

Catalog Events

Things that happened to the catalog: a price drop, a restock, a new arrival.

Custom

Trending Now — and whose trend

Trending Now compares today’s activity against the item’s own weekly average, so a steady best-seller does not trend and a small item catching fire does.

• Trending by picks the signal: Views, Purchases, Add to cart, or Wishlist.
• Taken from picks whose trend it is: the whole site, categories of what they engaged with, or interests of what they engaged with. The last two are personal trending — what is taking off in the corner of the catalog this person actually cares about.

Site-wide is the default on purpose: it does not depend on the visitor, so it is also safe for email and product ads. A scoped trend gives that up — do it deliberately.

Back in Stock

The engine records the moment an item’s stock crosses in either direction, automatically, from your normal feed imports. There is nothing to configure on the feed. The window (Returned) is since last visit, or a fixed span from yesterday up to the last 3 months — the same windows Price Dropped offers.

It is offered on all six page types, and Cart and Thank you are deliberate: a restock of something a visitor abandoned is one of the strongest recovery triggers there is. It combines with every filter and fallback, and with Skip items the visitor already… — but leave has purchased before off if you are writing a restock reminder.

Similar Items (AI), and Similar Content (AI)

Every other algorithm here needs traffic first: an item nobody has viewed, carted or bought is invisible to them. Similar Items (AI) reads the catalog itself — titles, brands, descriptions, categories (for articles: title, topic, author, description) — and computes each item’s nearest neighbours by meaning, within its own category. A brand-new product gets recommendations on day one.

Switch it on in any of three places — they are one switch:

• A recommendation campaign’s Catalog step, under Optional: Similar items (AI) → Turn on.
• Settings → Recommendation setup → Product Catalog (or Article Catalog) → the catalog card’s ⋯ More options → Similar items (AI) — semantic similarity from your catalog text, then tick Compute similar items for this catalog.
• Settings → Site profile → Recommendations → Similar items (AI) → Configure.

One switch covers the whole account, products and content alike. Use it through the Similar Items (AI) card under Cross-Sells, Co-Views & Upsells, or Similar Content (AI) for articles.

• It runs nightly, only over items that were added or changed; after activation the first pairs appear within a day.
• It uses AI credits and respects the account’s daily AI cap — over the cap, it resumes the next night.
• Until it is on, the card serves nothing and the widget’s fallback covers.
• As a fallback it fills with on-topic items instead of generic bestsellers.

Matches Their Search

Every other algorithm recommends from behaviour. None of them can answer a phrase, and a phrase is the most explicit statement of intent a visitor ever makes.

Matches Their Search reads the phrase Personyze keeps on the visitor’s profile — typed into your own site search, or carried in from a search engine — and matches it against your catalog’s own words: title, brand and short description for products, title, topic and description for articles.

• It answers a phrase the first time anyone types it. There is no nightly job and no phrase list to build up.
• Matching is natural language, not a query syntax, so a visitor typing C++ or an unbalanced quote cannot steer or break it. A phrase that matches nothing returns nothing.
• Relevance decides who is in the list; personalization only reorders it. Among equally relevant items, a newly added item, a price drop or the visitor’s usual price range decide the order.
• It needs a phrase. Until the visitor has searched, the widget renders nothing — so the editor sets the page group Searched on this site for you, and says so on the widget.

What is captured, how it differs from the Keywords rule, and how to target on it: targeting and recommending on the last search.

Video: Show visitors what they searched for (0:48) — it plays at the top of Personalize by What Visitors Search For.

Content recommendations

A Content Recommendations widget has the same picker over your article catalog, with four page tabs — Home Page, Category Page, Article Page, Search Page — and no Cart or Thank-you. On an Article page the anchor is the article being read. Twenty cards, in five groups.

Video: Content recommendations: what to show each reader (2:11) — it plays at the top of this page, beside the 4:16 in-depth one.

Personalized & Popular

Co-Reads & Related

Visitor History, Catalog Events and Custom

Content starter recipes

“More from this author” is not a card: it is Most Popular (or any card) with a filter rule Author · Exactly matches · Viewing article → Author — see filters.

After you pick: output, filters and fine-tuning

Beside the settings, the What this widget recommends rail counts what they leave against your real catalog — the pool filter by filter, what happens per visitor, the data the settings need, and what picks the items. See What this widget recommends.

Output — how many, and what happens when it comes back short

Match the visitor

Under the algorithm cards, Match the visitor puts the items that fit each visitor first: the brands, sizes, categories or authors in their own history, their gender, the prices they look at, where they are, their language, or one of their visitor fields. It re-orders the algorithm’s top 200 and never adds items. See Match the visitor.

Filters

Skip items the visitor already… (viewed, has in cart, saved, purchased), Restrict the catalog (Match visitor’s gender and age, discounted, with an image, link or price, from given categories) and Custom filter rules narrow what any algorithm may return. Out-of-stock items are excluded automatically. Match visitor’s gender and age need gender and age in your feed and on the visitor’s profile. All of it, with examples: setting filters on recommendations.

Fine-tune the algorithm

Gentle multipliers on the chosen ranking — the algorithm still leads; these nudge. Each one, once on, has a strength: Slightly, Moderately or Strongly.

Each boost says whether your catalog carries the field it needs — “No items have the profit field set” means it will do nothing yet. These apply to this widget only; the card’s footer shows the account-wide setting live (“Account-wide blend now: …”), so you never tune one widget blind to the account it sits in. See the score blend.

Looking for an older name?

Earlier editors and guides used other names. Here is where each one lives today.

The cards on this page are the ones the website picker offers, and recommendations for third-party emails, JSON feeds and apps use the same picker — only their page choices differ. An Email & SMS Personalization campaign is the exception: its recommendation blocks keep the older list, grouped General, Wish list, Cart Abandonment and Made a purchase (Personal and General for content), with entries such as Left in Cart Items and Wishlist Now Discounted.

Related

• Recommendations — Where to Start — the whole journey, and what to check when a widget shows nothing.
• Category & interest recommenders — the same engines, recommending a category or topic instead of an item.
• Filters on recommendations
• Recommendation fallbacks
• Score blend & presets — the account-wide weights behind every ranking.
• Interests — what the interest-scoped options draw on.
• Quick add to cart — add to the cart straight from a recommendation.
• Placement, triggers & frequency — any recommendation can also open as a popup, on exit intent, idle or scroll.