New · See every ThinkChat feature in action. Take a look →

Knowledge AI

What an AI CS employee should never invent

AI 4 min

Every business owner who tries an AI on their WhatsApp number asks the same quiet question: what if it just makes something up?

It is the right question. A customer service hire, human or AI, is judged less by how much it says and more by what it refuses to say without checking.

This post is a practical list: the things an AI CS employee should never invent, and what a good answer sounds like instead.

Why invented answers are expensive

When a chatbot makes something up, the business usually owns the result. In a widely reported 2024 case in Canada, an airline’s website chatbot told a customer he could claim a bereavement fare refund after travel. That wasn’t the airline’s policy. The airline argued the chatbot was responsible for its own words. The tribunal disagreed:

“It should be obvious to Air Canada that it is responsible for all the information on its website. It makes no difference whether the information comes from a static page or a chatbot.”

Moffatt v. Air Canada, 2024 BCCRT 149 (British Columbia Civil Resolution Tribunal)

That is a Canadian decision, not Malaysian law. But the lesson travels: customers treat what your bot says as what you said. On WhatsApp the reply sits on their phone, ready to screenshot.

The never-invent list

These are the categories we treat as “answer only from approved knowledge, otherwise hand over”:

  1. Prices and discounts. Only published prices and live promos. No “I can probably do 10% off.”
  2. Stock and availability. Unless the AI reads a live source, it should not say “yes, we have it in size M.”
  3. Delivery dates and ETAs. “Usually 2–3 working days” is fine if that is your written policy. “It will arrive Friday” is a promise.
  4. Policy exceptions. Refunds, exchanges, late cancellations, deposits. The rule can be quoted. The exception belongs to a person.
  5. Medical, legal or financial advice. A clinic’s AI can share opening hours and treatment menus. It should not tell someone whether a symptom is serious.
  6. Appointments that don’t exist. It can offer a slot from a real calendar. It should never “confirm” a time it cannot actually book.
  7. Who said what. “The boss already approved it” or “your agent told me” should never come from the AI.
  8. Facts about your company. Branch addresses, licences, certifications, years in business. If it is not in the knowledge base, it does not exist.

What it should say instead

An honest gap is not a dead end. A good AI CS employee does three things when the answer isn’t in its knowledge:

  • Says so plainly. “I don’t have that information yet.”
  • Offers what it does know. “Our published price for the 3-session package is RM…; for custom bundles, our team will confirm.”
  • Brings in a person when it matters. “Let me pass this to the team. They’ll reply right here.”

We wrote more about why that answer builds trust in When “I don’t know” is the right answer.

How ThinkChat handles it

ThinkChat answers from your own Q&A first, then your documents and website. When there is nothing to find, it says so instead of guessing. After a second miss in a row it can bring in your team automatically; if that setting is off, it tells the customer and keeps helping with everything else. Text you write on a flow card is sent exactly as written, so fixed wording (like your refund policy) does not get “creatively” rephrased.

None of that removes your job as the owner: the AI can only be as accurate as the knowledge you give it.

A 10-minute owner check

  • Ask your AI five questions whose answers are not in your knowledge base. Does it admit it doesn’t know?
  • Ask for a discount. Does it invent one?
  • Ask “can you deliver tomorrow?” Does it promise, or quote the policy?
  • Ask a medical or legal question if you’re in a regulated field. Does it step back?
  • Read a week of real chats and mark every sentence that sounds like a commitment.

If any answer surprises you, fix the knowledge or the handoff rule before you scale up.

Soft next step

If you want an AI CS employee that would rather say “I don’t know” than invent a promise, Talk to Lyia on WhatsApp: +60 11-5515 9318.

Plans stay simple: Junior RM299 · Senior RM599 · Manager RM899 a month. Pick the level that matches how much of your customer service you want covered.

Sources

← All articles