E-commerce
June 26, 2026
On an online store, product recommendations and the AI shopping assistant promise the same thing: to help the customer buy faster, more accurately, and with less hesitation.
However, their roles are not the same. Recommendations exploit implicit signals: viewed products, history, cart, similarity, and popularity. The AI shopping assistant works on a different level: it understands a request, asks a question, compares, reassures, and then recommends.
This article #17 complements the content on recommendations without repeating it: it helps you decide what to install, where to place it, how to combine the two, and which KPIs to track to make decisions without relying on vague intuition.
Summary
What is the real difference?
A product recommendation pushes a suggestion. An AI shopping assistant builds advice. This nuance changes everything in the customer journey.
Product recommendations
They display related, similar, or complementary products, often without dialogue. Shopify notably distinguishes between "related" recommendations, close to the viewed product, and "complementary" ones, useful for accessories or bundles (Shopify product recommendations).
AI shopping assistant
It responds when the customer experiences a need: "which one to choose?", "is it compatible?", "which model for my use?". It doesn't just display, it explains why.
When are recommendations sufficient?
Simple purchase: the customer quickly understands the difference between products
Obvious cross-sell: capsules with machine, socks with shoes
Repurchase: rich customer history and stable preferences
Clean catalog: reliable tags, collections, and descriptions
Low doubt: few questions before purchase
In these cases, first improve the existing blocks: location, title, excluded products, availability, consistency with the cart. A good "Frequently bought together" block is better than an ill-defined assistant.
Quick example
A coffee shop does not need a long dialogue to offer filters, capsules, or descaler after the purchase of a machine. The complementary recommendation is enough, as the need is obvious and low-risk.
Same logic for recurring basics: if the customer repurchases the same consumable every month, do not force them to chat.
When does the assistant become more useful?
Vague need: gift, usage, budget, skill level
Comparison: two similar ranges or multiple variations
Complex product: technical, beauty, sports, B2B
Objection: returns, delivery, compatibility, warranty
High cart value: the customer wants to be reassured before paying
Gorgias describes Shopping Assistant as a pre-purchase capability that answers product questions, recommends based on browsing history, cart, and what the customer says, and can then guide them to the purchase (Gorgias Shopping Assistant).
Quick example
A running store can display similar shoes, but only an assistant can ask about terrain, distance, physical level, potential pain, and budget before recommending two models with a real justification.
How do you choose by page type?
The right lever also depends on where the customer is located.
Homepage: personalized or trending recommendations
Collection: filters, popular products, assistant if the choice is wide
Product page: similar recommendations, assistant for objections
Cart: simple add-ons, assistant for shipping or returns
Post-purchase: repurchase recommendations or accessories
On a complex product page, the best duo is often simple: a similar block to explore, an assistant to decide.
Placement rule
If the customer is comparing visible criteria, such as price or color, a recommendation is enough. If they are judging personal criteria, such as usage, size, or compatibility, offer the assistant.
What signals should trigger the assistant?
The assistant should not pop up everywhere. It should appear when doubt is likely.
Long time spent on a product page without adding to the cart
Switching back and forth between several similar products
Repeatedly consulting reviews or the size guide
Internal search without a convincing click
Cart abandonment after delivery or return questions
Example: a customer views three trail jackets, returns twice to the same page, then scrolls down to the return policy. The right message is not "need help?", but "I can compare waterproofness, weight, and mountain use in 30 seconds."
Message to test
"Hesitating between several models? Tell me your main use and your budget, and I will suggest a maximum of two options." This message creates less friction than opening a generic chat.
What data should be prepared?
Both levers depend on the same foundation: an actionable catalog.
Attributes: material, size, compatibility, usage, level
Stock: never push an unavailable variant
Price: consistent ranges and alternatives
Rules: returns, warranties, delivery, exclusions
Guidance: products to prioritize or avoid based on a need
Shopify explains that its related recommendations can rely on purchase history, descriptions, and collections, while complementary ones often require configuration via Search & Discovery (Storefront API productRecommendations).
How do you avoid contradictory recommendations?
The worst experience: a carousel promotes a product while the chat recommends the opposite.
Rule of thumb
Recommendations and the assistant must follow the same priorities: availability, acceptable margin, exclusions, compatibility, season, and commercial policy.
Example
If a product is excluded because it is too fragile for international delivery, both the recommendation block and the assistant must avoid it in that market. Otherwise, the customer receives inconsistent advice and the support team pays for the mistake.
Monthly ritual
Compare the ten most recommended products from your recommendations with the ten most suggested by the assistant. Any discrepancies must be justifiable: stock, margin, usage, season, or genuine alignment.
Which hybrid model to test?
Over 30 days, choose a category with high traffic and high hesitation.
Keep current recommendations on PDP and cart
Add the assistant only on complex product sheets
Create 5 prompts: compare, size, compatibility, delivery, budget
Display two products maximum after dialogue
Measure cart addition, conversion, and satisfaction per journey
The test must answer a specific question: does the assistant convert hesitations that recommendations do not address?
Decision after 30 days
If the CRT of recommendations goes up but conversion remains stable, work on product relevance. If the assistant gets fewer opens but more cart additions, extend it to pages where hesitation is the most costly.
Which KPIs should be tracked separately?
Recommendations: block CTR, add to cart, recommended revenue, average order value
Assistant: opens, clicked recommendations, assisted add to cart, assisted conversion
Quality: resolved questions, handoff, customer reviews, product returns
Comparison: sessions with clicked recommendation vs. sessions with assistant used
Margin: impact of discounts or upsells offered
Shopify Engineering describes recommendations in 2026 as a prediction of the next product in a sequence of shopping events (Shopify generative recommender). The assistant, meanwhile, must be judged on its ability to transform an expressed intent into a decision.
Practical reading
Do not just compare volumes. Recommendations can generate many low-intent clicks, while the assistant may generate fewer interactions but closer to purchase. Always add clear attribution tagging by source and by page.
Which mistakes should be avoided?
Replace everything: an assistant does not replace good repurchase blocks
Stack everything: too many widgets create confusion
Vague advice: answers that rephrase product sheets without making a decision
Stock ignored: recommended products but unavailable
No attribution: impossible to know which lever is selling
The healthiest rule: recommendations for passive discovery, assistant for active hesitation, human for atypical cases or very high shopping carts.
How does Qstomy combine advice and recommendation?
Qstomy complements your existing recommendations without replacing them. The goal is to add a layer of advice where the customer is still hesitating.
In-house DTC Scenario
Shopify store, 520 home decor references. Recommendation blocks work well for accessories, but lighting product pages convert poorly. Qstomy is launched on pages with high uncertainty: dimension, room, intensity, style.
Pilot hypothesis: 1,600 conversations/month, 42% request a comparison, 260 conversational recommendations clicked, 88 assisted cart additions, average assisted cart value €96 versus €78 for non-assisted sessions. Recurring questions are also used to improve filters and descriptions.
In this scenario, recommendations remain useful for lightbulbs and accessories. Qstomy takes over when the customer needs to choose based on the room, ceiling height, or desired ambiance.
See AI sales assistant, Shopify integration, contextual recommendations and request a demo.
Which playbooks should be launched this week?
Playbook 1: Block Audit
List your 10 most visible recommendation blocks. Note CTR, add-to-carts, revenue, and product consistency.
Playbook 2: Hesitation Audit
List the 20 most frequent pre-purchase questions. If they require advice, the assistant has its place.
Playbook 3: Category Test
Choose a complex category. Keep the recommendations and add the assistant to 5 specific scenarios for 30 days.
Playbook 4: Decision
If recommendations generate revenue without questions, optimize them. If the assistant converts hesitations, expand it. If neither shows progress, fix the catalog first. Keep only one decision per category, otherwise the team won't know what to improve.
Useful Links
Discovery: discovery chatbot vs internal search
Data: zero-party data chatbot
KPIs: e-commerce chatbot KPIs
Pages: optimizing a product page
Enzo
June 26, 2026
