# How AI Guided Selling Works

> How AI guided selling works: constrain, retrieve and explain. Ask the right questions, ground every claim in catalog data and keep tradeoffs visible.

- Type: AI in Ecommerce guide
- Level: Intermediate
- Reading time: About 7 minutes
- Audience: Written for store owners, builders, and designers refining high-performance ecommerce stores.
- Last updated: 2026-10-30
- Canonical page: https://academy.getzenbasket.com/ai-in-ecommerce/ai-how-ai-guided-selling-works/
- Part of: [AI in Ecommerce](https://academy.getzenbasket.com/ai-in-ecommerce/) on ZenBasket Academy

---

## What you will learn

- The three jobs guided selling does, and why they must stay separate
- How to choose which constraints to ask about, and which to skip
- How to keep the assistant from inventing product facts
- How to explain fit and tradeoffs without manufacturing certainty

### Guided selling is three jobs, not one

**Constrain.** Turn "something for traveling with a laptop" into size, weight, capacity, and budget. This is the part that makes a conversation worth having, because it extracts criteria the shopper could not have entered into a filter.

**Retrieve.** Find the products that meet those constraints, from your catalog, using your current data. This is ordinary search and filtering underneath, driven by criteria the conversation produced rather than by the shopper clicking them.

**Explain.** Say why each option fits; in terms of the constraints the shopper gave. This is the only part most shoppers consciously notice, and it is the part most likely to state something untrue.

**Keep them separate when you build and measure.** If the shortlist is wrong, better explanations will not save it. If the explanation invents a feature, the retrieval was never the problem. Teams that measure the three together cannot tell which one is failing.

### Asking the right constraints

Ask only what changes the answer. Every question that does not narrow the result is a reason for the shopper to give up and go back to search. Three or four well-chosen questions beat a long intake, and the right ones are usually about the situation rather than the specification: what it is for, where it has to fit, what it has to work with, what the budget is and when it is needed.

The useful output of this stage is a small, visible set of criteria the shopper can see and change. If they cannot tell what the system thinks they asked for, they cannot correct it when it is wrong.

### Ground every claim in a fact you own

Recommendations must come from current catalog data: identifiers, attributes, variants, price, availability and fulfillment, each with an owner and a refresh rule. The failure that matters is not a poor shortlist. It is a confident specification that does not exist.

Set the rule explicitly: when a fact is unknown, say it is unknown. An assistant asked whether a case fits a particular laptop, with no dimension recorded, will produce an answer from the product name or the image. That answer reads exactly like a verified one, and the store is the party that told the shopper.

### Explain the fit, and keep the tradeoff

A good explanation connects a product fact to something the shopper said. "This one fits a 15-inch laptop and stays under your budget" is checkable. "This is our best option" is not, and it is a claim the evidence does not support when all you know is that the item met the stated constraints.

When several products fit, compare only the attributes that matter to this shopper rather than every field you hold, and preserve the tradeoffs. The cheaper one lacks a feature. The larger one will not fit the space. Hiding a disadvantage to close a sale converts a return into a refund and a complaint, which costs more than the order was worth.

### Design for correction

Shoppers change their minds mid-conversation, and the first set of constraints is often wrong. Let them adjust any criterion, reject a recommendation, and ask why an item appeared. Refresh the answer when price or stock moves, because a recommendation built on a five-minute-old cart is a promise you may not be able to keep.

Decide the fallback before launch. When the catalog cannot separate two products, send the shopper to the product pages or to a person rather than letting the assistant pick one and sound certain about it.

### When guided selling is a good fit

| **Your situation** | **Verdict** | **Why** |
| --- | --- | --- |
| **Attributes are complete and current on the products involved** | Good candidate | The system has real facts to reason from |
| **Products need nuanced fit or compatibility judgement** | Good candidate, if data is strong | This is exactly where filters fail shoppers |
| **The assistant would have to guess missing specifications** | Fix the data first | It will guess, fluently, and you will own the answer |
| **Shoppers already know the item or can filter for it** | Not worth it | Search is faster and already works |
| **Recommendations touch safety, health, finance or regulated claims** | Specialist review first | A wrong fit here is not a returned parcel |

The third row is the common one. Most stores sit somewhere between complete and sparse, so the practical route is to run guided selling on the categories where the deciding attributes are already recorded and leave the rest on search until the data catches up.

### Key takeaways

- Guided selling is a decision aid, not a chat interface bolted onto a catalog.
- Constrain, retrieve and explain are three jobs. Build and measure them separately.
- Ask only the questions that change the answer and show the shopper what you captured.
- An unknown fact must be stated as unknown, or it will be invented.
- Keep the tradeoff in the comparison. Hidden downsides come back as returns.

### Related learning

- AI-P012: AI Shopping Assistants vs Shopping Agents
- AI-P017: AI Shopping Assistant Errors and Human Handoff
- Existing Academy: How Do I Write an Honest Product Comparison?
- Existing Academy: How Do I Make Product Information Clear to AI Systems?
- Existing Academy: Ecommerce Product Recommendations
- Existing Academy: Helping Shoppers Choose Between Products