July 2026
- Aug 12
- 4 min read
AI Shopping Agent
We recently shipped the biggest addition to Visual Quiz Builder since the quiz itself: an AI Shopping Agent that helps shoppers find products through a conversation instead of a fixed set of questions.
What it does
A chat widget sits on your storefront. A shopper types what they're after in their own words — "something for dry skin that won't break the bank" — and the agent asks a few clarifying questions before recommending products. It's an LLM experience with your guardrails around it: your products, your tone, your recommendation widget.
Why it matters
Quizzes are great when a shopper is willing to answer five questions. Plenty aren't. The agent catches the shoppers who would otherwise bounce off a landing page, and it does it without giving up the control a quiz gives you. You decide the persona, the tone, how many questions it's allowed to ask, and which products it can reach for.
How to set it up
1. Open the new Agents tab in the left sidebar and click Create Agent.
2. Under Assistant Identity, set the persona name, tone of voice (professional, casual or enthusiastic), subtitle, start button label and welcome message.
3. Pick a Recommendation Strategy — Best Sellers, Trending Products or Product Discovery — and set the maximum clarifying questions and maximum recommendations.
4. Use Widget Appearance to match the launcher, colours and position to your theme.
5. Under Advanced Settings, set the auto-open delay, how long conversations are remembered (0 disables history), and whether shoppers see the agent's thinking steps.
6. Click Save Agent. Saving unlocks the Recommendation Component and Live Preview.
7. When you're happy, click Enable in Shopify, switch on the Shopping Agent app embed in the theme editor, and save. Storefront Availability will flip to Live.
Review AI Shopping Agent in our knowledge base for the full walkthrough.
Recommendation Component
What it does
The Recommendation Component decides how recommended products actually look inside the chat. Rather than picking from templates, you describe the card you want in plain English and the agent builds it.
Why it matters
A chat window is a small, awkward canvas. Cards that look great on a result page fall apart in a 400px-tall bubble. Being able to say "compact mobile cards, big image on top, full-width add to cart" and get exactly that — then tweak it in one more sentence — is much faster than fighting a template.
How to set it up
1. After saving your agent, open the Recommendation Component tab.
2. Choose a recommendation type: Products, Variants, Collections or Regimen.
3. Write your prompt. Type @ anywhere to insert dynamic fields — product name, price, description, image, variant SKU, collection name, Add to Cart button, Buy Now button and more.
4. Click Generate. To refine an existing design rather than start over, edit the prompt and click Modify.
Review Recommendation Component in our knowledge base for the full walkthrough.
Agent Responses & Customer Data
What it does
Every agent conversation is recorded in its own section of the Responses tab, updating live while shoppers are still talking. Click any record to see the full conversation, what was recommended, and everything captured along the way.
The agent can also collect data as it goes and update the customer's profile in Shopify and Klaviyo:
Contact information — turn on Require Contact Information and the agent asks for an email at a natural point in the conversation.
Customer Properties — define values like skin_type or budget with a plain description, and the agent picks them out of the conversation and attaches them to the session.
Customer Metafields — map to metafields that already exist in your Shopify store and the collected values sync straight to the customer record.
An AI-generated summary of the shopper's preferences and the recommendations they saw is written to the Shopify customer's notes, so your support team has context without reading a transcript.
Why it matters
This is the difference between a chatbot and a channel. A conversation that ends in a Shopify customer record with tags, metafields and a summary is a segment you can email tomorrow — not a nice interaction that evaporates.
How to set it up
1. In your agent's Customer Data Capture section, enable Require Contact Information.
2. Click Add Property for each custom value you want, with a short description so the agent knows what it's looking for.
3. Click Add Metafield and enter the metafield name exactly as it appears in Shopify. Names must match exactly for the sync to work.
4. Check results in Responses.
Review Shopify Integration in our knowledge base and Klaviyo Customer Segmentation for the full walkthrough.
Quiz Performance Insights
What it does
The insight on your quiz list — the one that grades how a quiz is performing over the last three months — got considerably sharper this month.
Why it matters
The original version graded every quiz against one completion benchmark, which was unfair. A quiz that demands an email before showing results will never complete at the same rate as one that doesn't, and telling a merchant their 72% is "needs work" when it's actually strong for their setup is worse than saying nothing.
The benchmark now adjusts for what you ask of people. For example:
Email required — strong above 70%, good above 60%, needs work below 60%.
Email optional — strong above 75%, good above 65%, needs work below 65%.
No email — strong above 85%, good above 75%, needs work below 75%.
The insight also now names the single question costing you the most completions (outside of the email question). Quiz Performance Insights is an excellent place to peek before starting a conversation with your quiz analytics using the VQB Integration with ChatGPT and Claude.
How to set it up
Nothing to switch on. Open the Quizzes tab and look at the insight on any quiz with enough recent traffic.
