She spent 26 hours talking to a chatbot. Your shoppers will not wait that long.

On 1 October, CBC News reported the story of Jasmine Ault, a Toronto resident whose Instagram, Facebook and a linked dog account were deleted overnight. Her Facebook account dated from 2006. The notice said she had broken community rules but not which ones.

She stayed awake for 26 hours trying to fix it. Meta did not respond to CBC's request for comment. Her story is about a social platform, not a store, but the way it failed is one your own chatbot can fail in too.

Four closed doors, one after another

First, the notice named no specific rule. Second, her appeal was closed as soon as she sent it, with no further option. Third, she paid $21 CAD for Meta Verification, expecting to reach a person, and got another AI assistant. Fourth, that assistant closed her ticket and told her she had reached her interaction limit.

None of these steps was a person deciding anything. Each one was a system reaching the end of its script and, with no instruction for that case, shutting the conversation.

Why companies build it this way

Professor Ela Veresiu told CBC that companies adopt AI for the savings, even when customers push back, and that chatbots struggle with "why" questions. That matches the pattern above: a bot can confirm that a rule was broken, but it cannot explain a decision to a person who disagrees.

The article cites a survey of 34,000 respondents in which 50% named the lack of empathy as their main frustration with AI customer service. It also reports that 87% of people prefer phone support, while only 7% of executives plan to prioritise it. Those two numbers sit far apart.

What a stuck shopper does instead

A Meta user has few alternatives. Your shopper has many. When your assistant closes the chat on a question about sizing, a delayed parcel or a charge they do not recognise, they do not stay awake for 26 hours. They leave, and the last thing they remember is the message that said they had used up their turns.

You also may not see it happen. A closed ticket looks like a resolved ticket in most reports. Shion Guha, quoted in the article, says automated systems need a clear pathway to a human agent and a fair appeals process. A store can read that as a short list of requirements.

Three exits to build into your chatbot

Start with a visible way to reach a person on every screen of the chat, not only after three failed attempts. A shopper who can leave the script early is less likely to feel trapped.

Next, remove the interaction limit as a dead end. If the limit is reached, the bot should pass the full conversation to a person, so the shopper does not have to explain again. Our post on when a chatbot should hand the conversation to a person lists the signals worth watching.

Finally, count the chats that end without an answer. If a ticket was closed by the bot, call it closed, not solved, and send a person to read a sample every week. Whoever reads them will find the "why" questions the bot cannot answer, and those are the ones a live conversation is for.