Skip to content
Supa Brain

The most common wrong answer in our tests

Your chatbot quotes a price that isn’t on your price list

The bot finds the right table and reads the wrong row: the pricier room, the member rate, the other plan. Customers act on it.

A customer asks what something costs. The bot answers at once, in a friendly sentence, with a number. The number is on your site, just not for that product: it belongs to the row above, the members-only rate or last season’s plan.

Nobody on your team sees it happen. The customer either books expecting that price, or decides you are too expensive and leaves.

How much is a triple room with breakfast?

A triple room is €175 per night, breakfast included.

Your page says

Triple Room €145 per night, breakfast included

Rates table. €175 is the Family Suite row below it

Wrong pricesA mistake we found in a live bot, retold with a made-up business.

How much is the annual membership?

The annual membership is £39 a month.

Your page says

Annual £399 paid upfront · Monthly rolling £39 a month

Membership page. £39 is the monthly plan next to it

Wrong pricesA mistake we found in a live bot, retold with a made-up business.

Why it happens

It isn’t random

01

Neighbouring rows look alike

Price lists repeat the same words: room names, sizes, “per night”, “per person”. The bot retrieves the right table and picks the line next to the one asked about. In our tests most wrong numbers worked this way.

02

Discounts read like prices

A member, student or early-bird price sits right beside the regular one. Without context the bot treats it as the price.

03

Old pages stay in the knowledge base

Prices change on the website, but the bot was trained on a copy from months ago, or on a PDF nobody updated.

What it costs you

  • A customer quoted too little expects that price at checkout or at the desk. You either honour it or argue with someone who has a screenshot.
  • A customer quoted too much leaves quietly. It never shows up as a complaint.
  • In 2024 a Canadian tribunal ordered Air Canada to pay a passenger the fare discount its chatbot had wrongly promised. It treated what the bot said as what the airline said.

How we check

  1. 01We read your price pages first and find pairs of similar rows: two room types, regular and member price, monthly and annual plan.
  2. 02We ask about one specific row from each pair, the way a customer would, and follow up with “and for two?” or “what about weekends?”.
  3. 03Every number in every answer is compared with your page. A wrong one is logged with the transcript, the page, the row and the screenshot.

What the fix looks like

  • Rewrite the price content the bot reads as one line per product with its full name, so neighbouring rows stop looking alike.
  • Add a short rule to the bot’s instructions: quote prices only with the product name, and link the price page.
  • Re-test the same questions to prove the fix holds.

Where we see it

Other mistakes we find

See your assistant answer your customers first

Email us your store’s address. Within a working day you get five questions your customers ask and the answers your assistant would give, each with the page it comes from. Free, no call needed.