What happens if the AI gives a customer wrong information?

If the AI is wrong, a customer heard a fee, an hour or a slot that was not true. That line should have come from a list you approved, or the question should have become a message. I will not promise a dental clinic or a coaching centre that this never happens.

Is a voice agent safe for a medical clinic is the ban list. Will they know it is an AI is the first sentence. Are chatbots useful is the loop in text.

Where a wrong line usually comes from

The card was last season. Holi hours. A batch that filled. A colour price you raised on Monday and never typed into the list. The AI voice agent and the AI chatbot will repeat the card. That is the design. It is also the failure.

Someone asked off the list and the system guessed anyway. That is not the product I sell. Off-list should be a number and a message on voice, or a handed thread on chat. If a transcript shows a guess, treat it as a break. Write to me.

A person misheard a calm voice and filled a gap. The agent said the door and they heard a diagnosis. The first sentence, that it is an assistant, is there to reduce that. Some people will still fill the gap. I cannot give you a percentage.

A Preet Vihar coaching desk that raised the two month fee on Monday, and left the old rupee on the card, will hear that old rupee spoken all week. A Jangpura clinic that shut for a wedding and forgot the spoken Sunday will seat nobody and still sound open. Neither miss is a city rate I can quote. Read your own transcripts.

I have no city-wide count of stale fees spoken on a line. I have not run that study. Your week of transcripts is the study.

What you do when it happens

Believe the customer first. Honour the chair or the batch if you said you would, or explain and make it right the way you already would after a receptionist error.

Then open the transcript. Match the sentence to the card. If the card was wrong, edit the card the same day. If the card was right and the spoken line was not, that is a defect in the build. I want that recording.

Do not wait for a monthly report I will not invent. Put the transcript hour next to the till hour. Preet Vihar parents will tell you the fee they heard. Jangpura patients will tell you the Sunday they trusted. Those two sentences are the audit.

What I will not put in writing as a guarantee

Zero errors. A model that “knows dentistry”. A bot that “understands CUET”. Those sentences are how shops get a wrong fee at scale.

Floor Rs 25,000 for voice and Rs 15,000 for chat, on pricing. Those floors are builds. They are not insurance.

I will not write that wrong answers are rare in Delhi NCR. Rare is a word that needs a count I do not have. I will write that a short card, a ban list, and a transcript you actually read are the hygiene I will stand behind.

How to reduce the chance, without a slogan

• Keep the card short. Short cards get edited.
• Ban diagnosis, drugs, and any result you have not seen. Write the ban.
• Read transcripts on a fixed hour, the way you already read the till.
• Change Holi and wedding shut days on the card before you change the sign.

None of that is a statistic. It is the same hygiene you already use with a new person at the desk.

Honest downsides

A wrong line on a voice is faster than a wrong line on a form. More people can hear it before you notice. That speed is why the product exists, and why the card matters.

Some owners will want me to “just let it be helpful”. That is how guesses start. I will refuse that setting.

A coaching owner who wants the bot to “just explain the course” off the card is asking for a paragraph that can drift. Write the paragraph. Put it on the list. If the question is not that paragraph, hand the thread.

What to do after a bad line this week

Save the transcript, the card as it stood, and what you told the customer.

Send those three. I will say the card was the hole, the build was the hole, or the line they heard was never on either and we need the recording before anyone guesses again.