AI Sales Assistant Chatbot for Product Recommendations: Write Product Notes the AI Can Match to What Buyers Actually Ask (Free Match Tester)
AI sales assistant chatbot tools promise to recommend the right product, but they can only recommend what your product information describes. Buyers don’t ask for “Hydra cream 50 ml”; they ask for “something for dry skin in winter” or “a gift for a new mom”. This guide shows how to write short product notes in buyers’ words, test whether real requests find the right product, and decide when the assistant should ask a question instead of guessing.
⚡ In short
- An AI assistant recommends from the knowledge you give it, not from your whole catalogue by magic.
- Write one short note per product: who it’s for, what problem it solves, when people buy it.
- Use the buyer’s words: problems, occasions and people, not just product names and specs.
- Test with real requests; when two products tie, the assistant should ask one question.

The gap
Why product pages don’t make good recommendations
Most product pages are written to describe the product: name, size, ingredients, specification, price. Buyers start from the other end. They describe a problem (“my hair keeps falling”), a person (“gift for my dad, he cycles”), or an occasion (“something for a wedding in December”).
An AI assistant reading your product pages has to bridge that gap. Sometimes it can. Often the bridge isn’t there, because no page mentions “gift”, “winter” or “new mom”, and the assistant either guesses or says it doesn’t know.
The fix isn’t a bigger model. It’s better notes. A two-line note per product that says who it suits, what it helps with and when people buy it gives the assistant what it needs to match a request to a product.
This matters more for small shops than big ones. With 20 to 200 products, you can write these notes in an afternoon, and an assistant that recommends well feels like a knowledgeable shop assistant instead of a search box.
Anatomy
What a good product note contains
Who it’s for
Skin type, age, size, skill level, household. “Beginners”, “kids under 5”, “oily skin”.
The problem it solves
In the buyer’s words: “pimples”, “hair fall”, “back pain from long drives”.
When people buy it
Seasons and occasions: “winter”, “Diwali gift”, “first apartment”.
Who it’s not for
One line that stops a wrong match: “not for very sensitive skin”.
The one fact people check
Price band, size, delivery time or a key ingredient.
A close alternative
“If you need fragrance-free, see Kids mild soap.” Helps when it ties.
Writing the notes
How an AI sales assistant chatbot learns your catalogue
Start with your top 20 products by sales, not your whole catalogue. Those are the ones people ask about.
Collect the buyers’ words first
Scroll through WhatsApp chats, reviews and emails, and note how customers describe their need before they name a product. Those phrases go into the notes.
Keep each note short
Two or three lines. A long note mixes many facts, and short notes are easier for an assistant to match and for you to keep accurate.
Write the “not for” line
It prevents the worst recommendation: the right category but the wrong person. A strong face wash recommended to someone with sensitive skin costs you a return and a review.
Add the notes to the assistant
In LiveChatBot, product notes go into the AI Agent’s knowledge as text. It doesn’t sync your store catalogue automatically, so update the notes when products change. Our guide to a chatbot trained on business documents covers the format, and training your AI chatbot shows the steps.
Free tool
Test whether buyers’ requests find a product
Paste a note per product and a few real requests. The tester shows which product each request matches, flags ties and lists requests that find nothing.
1. Your product notes, one per line: product | who it's for and what it's good at
2. Things buyers ask for, one per line, in their words.
Runs only in your browser; nothing is sent or saved. This is a simple word match, harsher than a real AI: if a request matches nothing here, your notes are probably missing the buyer’s words.
Reading the result
Fix gaps, ties and idle products
The tester uses plain word matching, which is stricter than a real AI. That makes it a good early warning: if a request finds nothing here, the words buyers use simply aren’t in your notes.
No match. In the example, “gift for a new mom” finds nothing, because no note mentions gifts or new mothers. Add “gift for new parents” to the kids’ soap note, and it matches.
Ties. “Which sunscreen for oily skin” matches both the face wash and the sunscreen on “oily skin”. A real assistant might pick either. Either add more specific words to each note, or let the assistant ask one question: “Is this for daily washing or sun protection?”
Idle products. A product nobody’s request reaches is either rarely wanted, or described in words buyers don’t use. Check its note against real chats.
Once the tester looks good, paste the notes into the AI assistant and ask it the same requests. Then ask a few new ones you didn’t test with.
Asking, not guessing
When the assistant should ask a question
A good shop assistant asks one or two questions before suggesting. So should an AI sales assistant chatbot, but not more than two, or it feels like a form.
Useful questions narrow the choice in one step:
- “Is it for you or a gift?”
- “Oily, dry or sensitive skin?”
- “What’s your budget roughly?”
- “When do you need it by?”
Write these into the knowledge as guidance: “If someone asks for skincare without saying their skin type, ask whether their skin is oily, dry or sensitive.” Then test it.
Hand over to a person when the request is unusual, the order is large, or the buyer asks about something your notes don’t cover. In LiveChatBot the free keyword bot answers from saved replies when the AI allowance is used up, so keep your top recommendations there too.
Examples
Product-first versus buyer-first notes
Measuring
Is the AI sales assistant chatbot recommending well?
Each week, read ten chats where the assistant suggested a product. Note whether the suggestion fit, whether it asked a sensible question, and whether the visitor clicked through or bought. When a recommendation was wrong, fix the note, not the prompt. For a wider view of which chats turn into sales, see our guide to answering buying doubts and start free at livechatbot.in.
Frequently asked questions
What does an AI sales assistant chatbot need to recommend products?
Short notes per product that say who it’s for, what it solves and when people buy it, written in the words buyers use.
Does LiveChatBot import my product catalogue?
No. You add product notes to the AI Agent’s knowledge as text, and update them when products change.
How many questions should it ask before recommending?
One or two at most. Pick questions that split your products cleanly, such as skin type or budget.
What if the assistant recommends the wrong product?
Look at the request and fix the product notes: add the buyer’s words or a ‘not for’ line.
Does the match tester send my product list anywhere?
No. It runs entirely in your browser and nothing is sent or saved.
Keep reading
Related guides
Sales Chatbot for Website Doubts: Answer the Six Questions That Stop People Buying, Page by Page (Free Objection Reply Builder)
Customer Engagement Chatbot Without the Nagging: Set an Interruption Budget for Every Visit (Free Interruption Counter)
Online Customer Chat Software for Returning Customers: Greet First-Timers, Repeat Visitors and Known Customers Differently (Free Greeting Planner)
Recommend like your best shop assistant
Give the AI Agent product notes in your buyers’ words and test what it suggests. Start free on livechatbot.in.