E-commerce
June 26, 2026
On an e-commerce store, two doors help the customer find a product: internal search and the product discovery chatbot. They seem close, but they do not serve the same moment of the journey.
Internal search is excellent when the customer knows what to type: a brand, a reference, a category, a SKU, a color. The chatbot becomes more useful when the need is vague: "which model to choose?", "what is the difference between these two products?", "what do you recommend for my use?"
This article #14 compares the two tools with a functional angle absent from the silo: when to use search, when to use the conversational assistant, and how to combine them without creating a confusing UX.
Summary
What is the real difference?
Internal search ranks results based on a query. The discovery chatbot clarifies a need, sometimes asks a question, and then recommends an option.
Internal search
It responds well to: "I know what I am looking for". Example: "iPhone 15 case", "black dress size 8", "filter refill X200". The customer wants to go fast.
Product discovery chatbot
It responds well to: "I know my need, but not the right product". Example: "I am looking for a gift for a beginner runner", "which cream for sensitive skin?"
The point is therefore not to replace search with a bot. The point is to direct each intent to the right tool.
When is internal search unbeatable?
Known reference: the customer types a name, a brand, or a SKU
Recurring purchase: they want to quickly find an already known product
Large catalog: searching avoids browsing through deep menus
Mobile: typing three words can be faster than a conversation
Expert use: professional buyer or loyal customer who knows your offer
Shopify Search & Discovery focuses on semantic search, predictive search, typo tolerance, filters, synonyms, and product recommendations (Shopify Search and Discovery). This is the key foundation to take care of before adding more advanced layers.
When does the chatbot provide the most value?
Vague need: gift, usage, budget, constraint
Complex product: technical, beauty, sports, B2B, equipment
Comparison: two similar ranges or multiple variants
Objection: return, size, compatibility, delivery
Advice: the customer wants a recommendation, not a list
Gorgias describes Shopping Assistant as a pre-purchase assistant that answers product questions, recommends based on browsing activity, the cart, and what the customer says, and can then use discounts with caution (Gorgias Shopping Assistant).
How do you route intents without making mistakes?
Routing can remain simple. Start with three families.
Short and precise query: internal search
Natural language question: chatbot
No-click or zero-result search: offer the assistant
Practical examples
“Nike Pegasus 41”: internal search
“running shoe beginner fragile knee”: chatbot
“summer wedding guest dress”: search then assistant if there are too many results
“compatible iPhone 15 Pro?”: chatbot on product page
What data makes both tools better?
Search and chatbot rely on the same foundation: a clean catalog.
Product titles: clear, consistent, without unnecessary naming variations
Attributes: size, material, compatibility, usage, color
Metafields: structured data that filters and assistants can exploit
Synonyms: customer words vs. internal vocabulary
Stock: do not promote unavailable items
Policies: return, delivery, warranty to address objections
Shopify reminds us that Catalog structures product data so that AI agents and channels better understand products, their variants, pricing, and availability (Shopify Catalog).
Clean up the common foundation
Before judging the tools, clean up what both of them read.
Synonyms: example: trainers, sneakers, running shoes
Naming: the same color should not be called midnight blue, navy, and dark blue depending on the product page
Metafields: compatibility, usage, size, material, dimensions
Collections: categories designed for the customer, not just for the internal team
Out-of-stock products: suggested alternatives rather than dead results
This work improves search, filters, recommendations, and chatbot answers. It is often the best investment before adding a new app.
How of to build a hybrid model?
The most robust model combines search, filters, recommendations, and a chatbot.
Visible search bar for precise queries
Useful filters on collection and results
Suggested questions on product pages
Assistant after search with no click or zero results
Human handoff on high basket values or highly specific cases
Example: the customer searches for "cabin travel bag". The search displays the models. If they click on two products and then hesitate, the chatbot offers: "I can compare volume, weight, and cabin compatibility in 3 points."
Which scripts should be used for the chatbot discovery?
Vague need
“To guide you quickly, tell me the main use, your budget, and the most important constraint. I will propose a maximum of two options.”
Comparison
“Model A is better suited if you are looking for lightweight. Model B is more durable for everyday use. For your case, I would choose B.”
Unsuccessful search
“I couldn't find an exact result. Are you looking more for a specific use, brand, or feature? I can suggest an alternative.”
Product sheet
“Are you hesitant about size, compatibility, or delivery? I can check before you add to the cart.”
Which KPIs should be compared?
Search used: share of sessions with search
Zero results: queries that return nothing
Results CTR: click on a product after search
Post-search conversion: purchase after internal search
Chatbot open rate: usage on PDP, PLP, cart
Clicked recommendation: product suggested by the bot then opened
Assisted add to cart: action after conversation
Assisted conversion: order after interaction
The right diagnosis bridges both worlds. If a query often yields zero results and also comes up as a chatbot question, first correct the catalog, synonyms, or attributes.
Which mistakes should be avoided?
Replacing too quickly: a chatbot does not compensate for a broken search
Pushing everything to chat: precise customers want to go fast
Separated databases: search and bot do not read the same attributes
Too many widgets: quiz, pop-up, chat, filters, and search compete with each other
No separate measurement: impossible to know what is actually helping
No context: the bot ignores the customer's last search
A good rule: first fix searches with no results, then use the chatbot for needs that are impossible to express in three keywords.
How do I test for 30 days?
Choose a pilot category, not the whole site.
Select a high-traffic category with high user hesitation
Identify the 30 most frequent search queries
Identify searches with no clicks or no conversions
Add 5 chatbot prompts on PDP or PLP
Measure search, chat, cart additions, and conversion
Decide: improve search, enrich bot, or fix catalog
Example: in beauty, launch on a "sensitive skin" collection. Search handles brands and categories. The chatbot handles routine, active ingredients, allergies, frequency of use, and comparison.
Decision at the end of the test
If searches with no results decrease after cleanup, continue with search. If chatbot conversations convert better on vague needs, extend prompt to other categories. If neither progresses, the problem likely lies in the product sheets.
How does Qstomy complement internal search?
Qstomy does not replace your Shopify search engine. It completes the customer journey when the buyer needs advice, a comparison, or a product answer before purchasing.
DTC sport scenario
Shopify store, 650 product references. Internal searches for "trail jacket", "hydration pack" and "road shoe" generate many clicks but few conversions. Qstomy is added to three collections with advisory prompts.
Pilot hypothesis: 1,400 assistant openings/month, 38% comparison requests, 24% size or compatibility questions, 110 assisted cart additions, assisted AOV €82 vs €69 unassisted. The team also improves 18 product sheets thanks to recurring questions.
See AI sales assistant, Shopify integration, train a Shopify chatbot and request a demo.
Which playbooks should be applied this week?
Playbook 1: searches with no results
Export queries with no results. Add synonyms, tags, or attributes before adding a chatbot prompt.
Playbook 2: too many results
If a search returns 100 products, add useful filters and a prompt: "Need help choosing?"
Playbook 3: comparison
Identify the two most compared products. Write a simple answer: choose A if, choose B if.
Playbook 4: context
On PDP, configure the chatbot to take the viewed product into account. A question "is it suitable?" must be understood within the context of the product page.
Playbook 5: routing
Add a simple rule: after a search with no clicks, suggest the assistant. After a conversation asking for a specific product, redirect to the results or the exact product page.
Sources and useful links
Shopify: Search & Discovery app
Shopify Dev: Product Recommendations API
Gorgias: AI FAQs product pages
Qstomy: contextual recommendations
Enzo
June 26, 2026
