Skip to content
AI-SNS.IO — CHAT SUPPORTAI-SNS.IO — CHAT SUPPORT
Knowledge Base

Answering Product Questions from Your Catalog in Chat

AI SNS CHAT EditorialPublished 7 min read

Prepare your product catalog so chat answers on web, WhatsApp and Messenger stay accurate: structure data, set limits, test and hand off.

Customers rarely ask big questions in chat. They ask "Does this come in navy?", "Will it fit a 15-inch laptop?" or "Is the large size in stock?" These are catalog questions, and a wrong answer to any of them costs a sale or creates a return. This article shows how to prepare your catalog, set boundaries for an AI agent, and test the result so product answers in website chat, WhatsApp and Messenger can be trusted.

Why catalog questions are different from FAQs

A general FAQ changes a few times a year. A catalog changes constantly: new variants, discontinued items, revised dimensions, seasonal bundles. The questions are also narrower. The customer wants one fact about one product, and a vague answer is useless.

Three things make catalog chat harder than ordinary support chat:

  • Precision matters. "Fits most laptops" is not an answer to "Will it fit my 15-inch laptop?"
  • Facts age quickly. A price or stock level that was right on Monday can be wrong on Friday.
  • Products look alike. Two items with similar names and different specifications are easy to confuse unless the source text keeps them clearly apart.

An AI agent answers from the material you give it. If that material is messy, the answers will be messy. Most of the work is preparing the catalog, not configuring the bot.

Decide what the agent should and should not answer

Start by sorting product questions into three groups before touching any settings.

Stable facts the agent can state directly

Materials, dimensions, compatibility, care instructions, what is in the box, warranty terms, and how products differ from each other. These change rarely and can live in your knowledge base.

Facts that change often

Price, stock levels, delivery estimates and promotions. Decide whether the agent states these from a stored document, looks them up from a live system, or points the customer to the product page. Stating them from an old document is the riskiest choice.

Advice and judgement calls

"Which one is better for me?" is a legitimate question, but the agent should answer only using the differences you have documented. Where the catalog does not say, the right behaviour is to say so and offer a human.

Shape your catalog data for chat

A spreadsheet export with 40 columns is not a good knowledge source on its own. Chat answers work best when each product has a short, self-contained description that names the product, states its key attributes in plain sentences, and mentions what distinguishes it from close siblings.

A simple structure per product

  1. Product name and any common alternative names customers use.
  2. One-sentence summary of what it is and who it is for.
  3. Specifications written as sentences or short labelled lines.
  4. Variants (sizes, colours) and which attributes differ between them.
  5. Compatibility and "not suitable for" notes.
  6. Care, warranty and return conditions specific to this product.

Raw export versus chat-ready text

Raw export
Chat-ready entry
SKU 4471, col: NVY, mat: PU, dim: 38x28x10
Metro Backpack in navy: PU material, 38 x 28 x 10 cm
Fits laptop: Y
Fits laptops up to 15 inches; not suitable for 17-inch models
Notes: see PDF
Warranty and returns are written out in the entry itself
Similar item names in one cell
Each variant has its own line, with differences stated
Catalog data before and after preparationIllustrative example

The right-hand column costs more effort once, but it removes the guesswork the agent would otherwise do at the moment of the question.

Include the questions customers actually ask

Look at past chats, emails and reviews. If customers keep asking "Is it waterproof?", add an explicit line: "Water-resistant, not waterproof; avoid submerging." Direct phrasing lets the agent give a direct reply.

Handle price, stock and other fast-changing facts

You have three honest options, and the right one depends on how often the data changes.

ApproachGood forRisk
Store in the knowledge baseRarely changing prices, fixed bundlesGoes stale if nobody updates it
Look up from a live systemStock, order status, current pricesNeeds the system to be connected and reliable
Send the customer to the product pageHighly volatile or complex pricingAdds a click, but never misleads

For a small shop whose prices change twice a year, a stored price with a named owner for updates may be fine. For a shop with daily changes, avoid stating price or stock from static text. Tell the agent in its instructions to send the customer to the live page for those two facts, or use a read-only lookup against the system that holds them.

Write answers that suit each channel

The same fact reads differently on different channels.

  • Website chat: customers have the product page open beside the chat. Short answers with a pointer to the relevant page work well.
  • WhatsApp: conversations are personal and often on mobile. Keep replies brief, one idea per message, and avoid long tables.
  • Messenger: similar to WhatsApp, with customers often arriving from an ad or a post. The first message may be as vague as "How much?", so the agent should ask which product they mean.

Across all three, a good product reply does four things: it answers the question first, adds the one detail that prevents a return, offers the next step, and never claims more than the catalog says.

Ask a clarifying question when the product is ambiguous

If the catalog has three backpacks, "Does it come in black?" is unanswerable until the agent knows which one. Instruct the agent to ask which product, and to confirm the variant before quoting specifics.

Test before launch, then review real chats

Build a test set of 30 to 50 questions drawn from real customer wording, and include difficult ones: two similar products, a question the catalog cannot answer, a misspelled product name, and a request for a discount you do not offer.

  1. 01Prepare product entries
  2. 02Test with real questions
  3. 03Launch on one channel
  4. 04Review chats and fix gaps
A simple catalog-chat improvement loopIllustrative example

Then work through a loop. Check each answer against the source. Where it is wrong, fix the source text first, not the instructions, because bad data is the most common cause. After launch, read actual conversations weekly at the start, and sort what you find.

Answered from catalog
180
Needed live price or stock
70
Not in catalog, handed to a person
35
Off-topic or unclear
15
Example: how 300 monthly product chats might be sorted (hypothetical)Illustrative example

The numbers above are invented for illustration. The value of the exercise is the shape: the "not in catalog" group tells you which product information to write next, and the "live price or stock" group tells you where a lookup or a page link is worth setting up.

Doing this in AI SNS CHAT

Here is how the steps above map to the platform.

1. Add your catalog as a data source. Open Data sources in the sidebar. You can upload Files, add Text snippets, point the agent at your Website, or enter Q&A pairs for the questions customers ask most. If your product documentation lives elsewhere, the connectors under Channels include Google Drive, OneDrive, Dropbox, Box, Notion and others. See the full list on connections. Keep in mind that storage is limited by plan, from 30 MB on Free up to 3 GB on Corporate.

2. Write chat-ready entries. Use Text snippets or Q&A for the high-traffic products and for the "not suitable for" notes. Avoid putting internal material such as supplier costs into the knowledge base, because passages can be shown to people outside your team.

3. Set persona and guardrails. In the agent's persona, add your "must not guess" list: no invented stock levels, no estimated delivery dates, and a clarifying question when the product is unclear.

4. Connect live systems where needed. Integrations include a built-in Shopify action and read-only lookup tools such as WooCommerce, Google Sheets and Airtable. Integrations require Hobby or higher, and the number of Actions per agent depends on the plan.

5. Test in Playground. Run your test set before the agent goes live. Playground also has an A/B model compare if you want to see how two models handle the same questions.

6. Deploy. From Channels, add the Chat widget, WhatsApp or Messenger. Details are on the features page.

7. Capture leads and hand off. The Collect leads action stores contact details in Activity → Leads and creates a contact. For questions the catalog cannot answer, use Helpdesk → Scheduling to set shifts, so handoffs go to whoever is on duty, with alerts to Slack, Teams, email or other targets. The customer sees a "what happens next" line based on your rota, and the server never invents a wait time.

8. Improve from real chats. In Activity → Chat logs you can filter, rate and review conversations. You can take over a chat, and you can edit a reply into the ideal answer to train the agent. Email conversations appear in Inbox, and every other channel appears in Chat logs.

Each reply uses one credit, so plan sizes on the pricing page map directly to monthly chat volume. If you want help finding a screen, Synthia, the built-in fleet manager, can walk you through it.

Frequently asked questions

Can the agent quote live prices and stock levels?

Only if the data reaches it. Stored text goes stale, so for volatile facts either link to the live product page or use a read-only lookup against the system that holds the data, such as WooCommerce or Google Sheets. Integrations need the Hobby plan or higher.

How should I format a large catalog?

Give each product a short entry with a name, summary, key specifications and notes on what differs from similar items. Split by product rather than pasting one huge export, and write out variant differences in plain sentences.

What happens when a customer asks about something not in the catalog?

Instruct the agent to say it is not sure and offer a person. With Helpdesk scheduling set up, the handoff goes to whoever is on shift. The conversation then appears in Chat logs, where you can see the gap and add the missing information.

Will the same catalog work on WhatsApp, Messenger and website chat?

Yes, the agent uses one knowledge base across channels you deploy it to. You may want shorter reply instructions for WhatsApp and Messenger, where customers read on mobile.

#product catalog#knowledge base#ecommerce support#WhatsApp#chatbot

Put this into practice with AI SNS CHAT

Build an AI agent trained on your own content and connect it to your website, LINE, WhatsApp and more. Free to start, no credit card required.