​

How Personalization Works: One Store, a Million Versions

Two shoppers open the same store at the same moment. One has been browsing trail-running shoes all week; the other just bought a coffee grinder. They land on the same homepage — but they don’t see the same products. The runner sees trail shoes and hydration packs. The coffee lover sees filters, beans, and a matching kettle. That’s personalization, and it isn’t magic. Glood.AI quietly does three things: it listens to what each shopper does, it understands what your products are and how they relate to each other, and then it reorders what every shopper sees — in recommendations, search results, collection pages, emails, and checkout — so the most relevant products come first. This page explains how that works across the whole platform, in plain language.

​

The Building Blocks

Everything personalized in Glood.AI is built from two ingredients: a taste profile for each shopper, and product intelligence for your catalog. Here’s how they come together:

​

The signals: what Glood.AI listens to

The Glood.AI Web Pixel — a small, privacy-sandboxed tracker that Shopify itself supervises — records the shopping journey as it happens:

• Page and product views: which products a shopper looks at, and in what order
• Searches: what they type into the search bar, and which results they engage with
• Cart actions: what gets added (and removed)
• Checkouts and purchases: what they actually buy
• Widget interactions: which recommendations they see, click, or ignore

All of this is first-party data from your own storefront. Nothing is bought from data brokers, and nothing is sold onward.

​

The taste profile: a memory that leans recent

Every shopper gets a lightweight profile — think of it as a short-term memory of their taste. It’s deliberately biased toward the recent past: the last few products someone viewed count far more than what they browsed last month. That’s a feature, not a shortcut. Shoppers change missions. Someone who spent last week browsing cookware and today starts looking at camping gear is on a camping mission now — and within a couple of page views, their recommendations lean toward camping gear too. The profile follows the shopper’s intent instead of clinging to their history. Identity works in layers:

• Anonymous visitors are remembered per browser, so personalization starts from the very first product view — no login required.
• Logged-in customers get a profile tied to their customer account, which unlocks cross-device recall: view a navy running shoe on your phone at lunch, and it’s waiting in “Recently Viewed” when you open your laptop that evening.
• Headless stores and email channels can identify shoppers by email address or phone number, so personalization follows the shopper even outside the standard storefront.

​

Product intelligence: knowing your catalog like a good salesperson

The taste profile is only half the story. Glood.AI also builds a deep understanding of your products:

• What each product is: AI reads every product’s title, type, tags, and description — and with AI Product Enrichment, even its images — to understand it the way a shopper would (“a slim-fit, floral, midi summer dress”), not just as a row in a database.
• How products relate: the crowd teaches the system. Products that shoppers view in the same session become “similar.” Products bought in the same order become “frequently bought together” — weighted by how often and how profitably the pair sells. If 6 out of 10 orders containing a particular phone also contain the same case, that case earns its spot under the phone.
• What’s working right now: sales velocity and trending signals surface products gaining momentum, while products that get shown a lot but ignored are gently pushed down. Being ignored is data too.

​

Personalization, Feature by Feature

​

Product Recommendations

Recommendation widgets are where personalization is most visible, and different widgets use it differently:

• Personalized For You is the visitor-first widget: the shopper’s own browsing history is the starting point. A shopper who has been circling three different yoga mats sees yoga gear front and center — even on the homepage, where there’s no “current product” to anchor to.
• Similar Products, Frequently Bought Together, and Cross-Sell are anchored to the product on the page — and become personalized when you switch their ranking to Personalized in the dashboard. The balance matters: the product being viewed still dominates, and the shopper’s history acts as a tiebreaker. If five jackets are equally good matches, the one closest to this shopper’s taste rises to the top. Personalization nudges the widget; it never hijacks it.
• Recently Viewed and Recently Purchased are inherently personal — a private trail of the shopper’s own journey, cross-device once they log in.

What about brand-new visitors? Someone with zero history never sees a broken or empty widget. They get bestsellers or trending products as a smart default — and the moment they view their first product, personalization kicks in and starts steering. Dive deeper into each engine: Personalized For You, Frequently Bought Together, Similar Products, and Trending Products.

​

AI Search

Two shoppers type the exact same query — say, “shoes” — and get results in a different order. The runner with a week of trail-shoe browsing sees trail runners first; the shopper who’s been browsing formal wear sees oxfords and loafers. The results are re-ranked against each shopper’s taste profile, on top of Glood.AI’s typo-tolerant, intent-aware search. Search is also a two-way street: what a shopper searches for feeds their taste profile. Someone who searches “linen shirt” has just told the system something valuable — and their recommendations across the rest of the store adjust accordingly, even if they never click a result. You decide the recipe. The Search relevance settings let you tune exactly what makes one product rank above another. Ranking is a blend of six signals, each with its own share:

• Keyword match — how much the shopper’s actual words matter
• Semantic similarity — how much meaning matters when the exact words are absent, so “warm jacket” can surface a parka whose title never says “warm”
• Personalization — how much this shopper’s own history shifts the results
• Trending — how much recent sales momentum matters (tuned further by velocity and acceleration)
• Recency — how much newly published products are favoured, with a half-life you choose
• Price match — used only when the query implies a price, like “dress under 2000”

Presets such as Balanced, Best sellers, New arrivals, Personalized, and Strict relevance give you a sensible starting point before you touch a single slider. The shares are also honest about when they apply: personalization only kicks in for shoppers who actually have browsing history, and price match only when the query implies a price — when a signal can’t apply, its share is redistributed across the rest instead of silently distorting results. The personalization share is itself a blend you control. You decide how much weight goes to what the shopper has browsed, searched, purchased, and currently has in the cart — so a store where carts signal strong intent can lean on cart contents, while a discovery-heavy store can lean on browsing.

​

Collection Merchandising

A collection page is prime real estate, and the default sort order treats every shopper identically. With personalized collection merchandising, the order of products inside a collection adapts to each shopper: in your “Footwear” collection, the trail runner sees trail shoes in the first row, while the sneakerhead sees lifestyle sneakers first. Collection pages use the same tunable ranking engine as search — the relevance settings have a dedicated Collection pages tab. There’s no query to match on a collection page, so the recipe blends the behavioral signals:

• Personalization — how much this shopper’s own history (browsed, searched, purchased, in cart) reorders the collection
• Trending — products with recent sales momentum climb toward the top
• Recency — newly published products get a boost that decays as they age
• Sales volume — steady sellers over your chosen window (off by default, so a few bestsellers don’t flatten discovery)

Because shares redistribute when a signal can’t apply, a first-time visitor simply sees the collection ordered by trending and recency — and the moment they build a little history, personalization starts earning its share. Your merchandising rules always win. If you pin products to the top, boost a product line, or bury out-of-stock items, personalization works within those rules — it re-ranks what’s left, it doesn’t override your strategy.

​

Email

Email personalization is keyed to the one identifier email already has: the email address. When your email tool (Klaviyo or any ESP) renders a Glood.AI recommendation block, the products are chosen for that specific recipient at the moment they open:

• Recent buyers see products that complement what they just bought — the tripod after the camera, the conditioner after the shampoo.
• Recent browsers see products similar to what they’ve been looking at on the store.
• Brand-new subscribers with no history see fresh arrivals — rotated weekly, so your Tuesday newsletter never shows the same four products it showed last Tuesday.

Because the blocks are delivered as images and links, they work inside your existing email templates with no re-platforming. Glood.AI can also sync recommended products directly onto Klaviyo customer profiles for use in flows and segments. Customer-based personalization via the Recommendations API. For full control — custom email pipelines, CRM flows, or any backend integration — your system can query the headless Recommendations API directly with a customer ID (an email address or phone number works too). Glood.AI resolves that customer’s complete history and uses AI to assemble their set:

• What they’ve purchased anchors complementary picks — the AI finds products that pair naturally with what the customer already owns, the way a tripod follows a camera.
• What they’ve browsed and added to cart anchors similarity — products close in style, category, and intent to what they’ve been circling but haven’t bought yet.
• No history at all falls back to bestsellers relevant to the customer’s region, so the response is never empty.

The same query works for one customer at a time (personalizing a transactional email as it’s sent) or in batch (pre-computing picks for an entire campaign list), and the identity resolution is the same one the storefront uses — so a customer’s email picks stay consistent with what they see on the site.

​

Checkout & Post-Purchase Upsells

Checkout upsells blend your rules with per-cart intelligence:

• You decide when an offer appears: campaigns trigger on cart contents — subtotal thresholds, specific products or collections, item counts.
• AI decides what fills the smart slots: AI-powered slots are filled with frequently-bought-together products seeded from this shopper’s actual cart. A cart holding a yoga mat gets a strap and a block offered at checkout; a cart holding running shoes gets performance socks. Post-purchase offers work the same way, seeded from what was just bought.

The shopper sees a stable offer throughout their checkout — it won’t reshuffle between page reloads and erode trust.

​

Customer Account Pages

Logged-in customers get the most personal surfaces of all: a full Personalized For You page in their account (their picks, their recently viewed items, follow-ups to their last purchase), plus recommendation blocks on order pages anchored to what they last bought. This is where cross-device profiles pay off most — the account is the identity.

​

You Stay in Control

Personalization never overrides your merchandising judgment. Every personalized surface respects the same hierarchy:

1. Pinned products come first. If you pin it, it shows — personalization ranks everything after it.
2. Your rules set the boundaries. Merchandising rules can act as gentle boosts or hard requirements. Exclude tags, cap price ranges, keep recommendations inside a collection — the AI works within the fence you draw.
3. Segments target the right widget to the right group. Show one section to first-time visitors and a different one to returning buyers.
4. A/B Experiences prove what works. Test a personalized setup against a non-personalized one; every shopper is stably assigned to one variant (no flip-flopping mid-session), so your results are clean.
5. Geo-targeting localizes the defaults. Bestsellers and trending fall back gracefully from city to region to country, so a shopper in Toronto sees what’s actually selling near Toronto.

​

Privacy, Honestly

• First-party only: signals come from your own storefront, collected through Shopify’s strict web-pixel privacy sandbox.
• No data sales: shopper behavior is used to personalize your store — nothing is shared across merchants or sold to anyone.
• Graceful by design: if there’s no history, no consent, or no identity, shoppers simply see smart defaults like bestsellers. Personalization is an enhancement layer, never a point of failure.

​

Explore Further