AI Chat Support Software and Your Store Policies: Rewrite Returns, Refunds, Delivery and COD Rules So the AI Can’t Misread Them (Free Policy-to-Knowledge Builder)

AI chat support software answers from whatever you give it, and most policy pages were written by lawyers or in a hurry, not for a machine that quotes them word for word. The result is the most expensive kind of wrong answer: a promised refund you don’t offer. This guide shows how to rewrite policies as short, explicit facts, test them with awkward questions, and decide which policy questions a person should answer.

1fact per sentence
5rules for policy text
6+edge-case test questions
0exceptions approved by the bot

⚡ In short

  • Policy pages are written for people who read the whole page. AI answers quote one piece of it.
  • Rewrite each rule as one complete fact: number, start point, condition.
  • List exclusions separately, and name who handles exceptions.
  • Test with edge cases such as day 8 of a 7-day window, not just ‘return policy?’.
Ai chat support software in LiveChatBot: the LiveChatBot shared inbox with a conversation open
How it looks in LiveChatBot: the LiveChatBot shared inbox with a conversation open.

The problem

Why policies are where AI answers go wrong

A typical returns page says something like: “Returns accepted within 7 days, subject to conditions. Certain categories excluded. Refunds processed promptly.” A person reading the whole page gets the idea. An AI assistant answering “can I return my sale kurta?” may pick up “returns accepted within 7 days” and say yes, because the exclusion is somewhere else and doesn’t name sale items.

Three things make policy text risky:

  • Missing start points. Seven days from what: the order, the dispatch or the delivery?
  • Vague words. “Promptly”, “usually” and “subject to conditions” give the AI nothing firm to say, so it guesses.
  • Exceptions hidden in other sections. The AI may retrieve the rule without the exception.

It gets worse with Hinglish and short messages. “return?” or “paisa wapas kab” gives the assistant very little to match, so it leans even harder on whichever sentence looks closest. Clear, self-contained facts make that closest sentence the right one.

None of this is the software’s fault. Any assistant that answers from your knowledge is only as precise as that knowledge.

Five rules

How to write policy knowledge for AI chat support software

1. One fact per sentence

“You can return an item within 7 days of the delivery date.” Not three rules joined with commas.

2. Numbers, not adjectives

“Within 5 working days”, not “promptly”. If you can’t commit to a number, say who to ask.

3. Always say where the clock starts

Order date, dispatch date or delivery date. Customers count from delivery; your policy may not.

4. Put exclusions next to the rule, by name

“These items cannot be returned: innerwear, sale items.” The AI can then answer the specific question.

5. Say who handles exceptions, and that the assistant doesn’t approve them

This one sentence stops the AI from promising refunds or replacements you haven’t agreed to.

Our guide to knowledge your bot can use covers general writing; these rules are the stricter version for anything involving money.

Free tool

Turn your rules into knowledge and test questions

Fill in what applies. Copy the knowledge into your assistant, then ask the test questions exactly as written.

Fill in your store’s rules. Leave a box empty if it doesn’t apply. The builder writes knowledge the AI can’t misread, plus test questions with the answer you should expect.

Runs only in your browser; nothing is sent or saved. Figures in the example are illustrative.

Before and after

Three policy lines, rewritten

Returns
🚩 Instead of“Returns accepted within 7 days subject to conditions.”
✅ Write“You can return an item within 7 days of the delivery date. The item must be unused, with its tags.”
Refunds
🚩 Instead of“Refunds are processed promptly.”
✅ Write“After we receive the returned item, the refund reaches your original payment method within 5 working days.”
Exceptions
🚩 Instead of“Management reserves the right to decide.”
✅ Write“For anything outside these rules, our support team will help. The assistant does not approve exceptions.”

Testing

Ask the awkward questions before customers do

After pasting the new knowledge, test with questions designed to break it. The builder suggests some; add your own from real chats.

  • Just outside a limit: day 8 of a 7-day window, a ₹6,000 order when COD stops at ₹5,000.
  • Excluded items: “can I return innerwear?” must get a clear no.
  • Mixed language and short forms: “refund kab milega”, “cod?”.
  • Pressure: “it arrived damaged, refund me now”. The right answer is a handover, not a promise.

Write down each answer as right, wrong or handed over. Fix wrong answers first: a confident wrong answer about money does more harm than “let me get someone”. Re-run the same questions after every policy change; our regression testing guide shows how to keep a fixed set.

Who answers

Which policy questions AI chat support software should hand over

Even with perfect knowledge, some policy questions shouldn’t be answered by any bot. A useful test: does answering need a look at one customer’s order, or a judgement call? If yes, a person takes it.

The AI answers

General rules that are the same for everyone: the return window, refund timing, delivery days, COD limits, warranty length, which items are excluded. These are stable facts, and a correct instant answer saves both sides time.

A person answers

Anything about one specific order (where it is, whether it qualifies), damaged or wrong items, requests for an exception, disputes about a charge, and any message that mentions a complaint, consumer forum or legal notice. The AI’s job here is to acknowledge, collect the order number and hand over.

Grey areas

“My order is late, can I cancel?” has a general rule and a specific case. Let the AI state the rule, then offer a person for the order itself. Our customer service rulebook has a fuller split by topic.

Checklist

Policy knowledge, ready to go live

Policy knowledge

In LiveChatBot

Where the policy knowledge goes

In LiveChatBot, paste the facts into the AI Agent‘s knowledge. Save the same sentences as saved replies too: when the AI allowance is used up or AI is unavailable, the free keyword bot answers from saved replies, so customers still get the same wording.

For questions about one specific order, payment or damaged item, let a person take over. Offline, the away form collects an email or phone number, and the message can become a ticket your team follows up; see turning a chat into a ticket.

When a policy changes, update the knowledge, the saved replies and your website page on the same day. Our change checklist covers every place a fact repeats.

Frequently asked questions

Can AI chat support software read my existing policy page?

It can, but vague wording and exceptions in other sections cause wrong answers. Rewrite the key rules as short facts as well.

Should the AI approve refunds or exceptions?

No. Name who handles exceptions and state that the assistant doesn’t approve them. Hand those chats to a person.

How often should I re-test?

After every policy change, and once a month with the same set of edge-case questions.

What if a customer argues with the policy?

The assistant should restate the rule once, politely, and offer a person. It shouldn’t negotiate.

Does the builder store my policy?

No. It runs in your browser and nothing is sent or saved.

Keep reading

Related guides

📘

Chatbot Integration Service or DIY? Map Each Connection to Its Route, Count the Hours, and Only Pay for the Custom Parts (Free Route Mapper)

Read the guide →

Give the AI rules it can’t misread

Give your AI chat support software precise policy facts in the AI Agent, and keep money decisions with your team. Start free on livechatbot.in.