82% of Shoppers Want a Refund When Your AI Breaks a Promise

Yousif Alias · 4 min read

82% of Shoppers Want a Refund When Your AI Breaks a Promise

What did the Forethought holiday survey find?

82% of US shoppers expect a business to give a refund or credit when its AI makes a promise the business doesn't keep. That's the headline from Forethought's State of AI in CX 2026 holiday report, released November 4, based on an online survey of 1,039 US adults fielded September 4 to 8, 2026.

The same shoppers are open to AI. 62% would let AI make some purchases for them. But the patience runs out fast when a bot gets it wrong. From the full report:

FindingShare of shoppers
Expect a refund or credit when the AI breaks a promise82%
Have been frustrated by a bot that couldn't understand their issue72%
Would let AI make some purchases directly62%
Would be more loyal if the AI warned them about a delay and offered a fix50%
Will wait less than 30 minutes for support before canceling or buying elsewhere44%
Abandoned a cart because a chatbot couldn't answer a basic shipping or stock question41%
Would let AI make purchases over $50018%

Forethought sells AI customer service agents, so read it as a vendor survey. The questions are still the right ones, and the numbers line up with what I see when a business puts a bot on its site and walks away from it.

Does a business have to honor what its AI agent says?

In practice, often yes. Customers already expect it, and at least one tribunal has ruled that way. In 2024 a British Columbia tribunal ordered Air Canada to pay a customer after its website chatbot described a bereavement refund policy that didn't exist. The airline argued the chatbot was responsible for its own words. The tribunal rejected that.

Now regulators are writing it down. On October 20, New York City's Department of Consumer and Worker Protection proposed a rule on automated customer service. It would require businesses to tell people up front when they're talking to an automated system, give them an easy way to reach a human, and bar businesses from blaming unlawful conduct on the technology. It covers phone voice menus and rules-based chatbots, not just AI. Fines would start at $525, and a Ballard Spahr analysis says repeat violations could reach $3,500. A public hearing is set for November 19, 2026, and comments are due that day.

It's still a proposal, and it only covers businesses serving New York City consumers. Plan as if your AI agent's words count as your own anyway, because your customers already treat them that way.

What does this mean if your AI answers leads and books calls?

If an AI agent answers your phone or your texts, everything it quotes is a promise from you. A price, an appointment time, a warranty, a "we'll call you back in ten minutes." If it can't back that up with real data, it shouldn't say it.

Most of the damage I see comes from agents that guess. They quote a price that changed last month, confirm a time slot that's already taken, or promise a callback nobody makes. In our own calendar data, a booking made in the wrong time zone turned into a no-show. That's a small broken promise, and it still cost a sale.

Here's the setup I'd check this week:

  1. Write down what your agent is allowed to promise: prices, service area, hours, booking windows. Anything outside that list goes to a person.
  2. Connect booking to your live calendar so the agent only offers open slots, in the caller's time zone.
  3. Have the agent say it's an AI assistant at the start of the conversation. It costs you nothing and matches where the rules are going.
  4. Give every caller and texter a one-step path to a human, and a real callback time when nobody is free.
  5. Read 20 transcripts a week. Flag any answer the agent made up and fix the source it should have used.

How fast do customers expect an answer?

Fast. 44% of shoppers in the Forethought survey will wait less than 30 minutes for support before they cancel or buy somewhere else. For a local service business, the window to win a new lead is usually shorter than that.

That's the trade every owner is making. A human-only team misses calls after hours and at lunch. A badly built bot answers instantly and then says something you have to pay for. The fix is an agent that answers right away, sticks to what you've approved, and hands off cleanly. If you want to see what your unanswered calls are costing you first, run your numbers through our free missed call calculator.

AI that only promises what you can deliver

KenjiAI's Voice AI agents and Conversation AI answer calls and texts in seconds, book into your real calendar, and pass anything off-script to your team with the full conversation attached. Every reply lands in one inbox, so you can read exactly what the agent told each customer. We've been running ads and answering leads for businesses since 2013, and 500+ businesses have used our systems.

What should you do before the holiday rush?

Audit what your AI is allowed to say before your busiest weeks, not after a customer screenshots a bad answer. Start with the five checks above, then decide in advance how you'll handle a broken promise: refund, credit, or a call from you.

If your agent calls or texts leads, also make sure your consent and calling rules are in order. We cover that in our guide to the FCC rules for AI calls.

Next step: pull last week's AI conversations, highlight every promise the agent made, and check each one against what your business can actually deliver.

Frequently asked questions

Do customers expect a refund when a company's AI chatbot makes a mistake?

Most do. In Forethought's November 2026 holiday survey of 1,039 US adults, 82% said they expect a refund or credit when a brand's AI makes a promise the brand fails to keep. Plan for it before it happens.

Is a business responsible for what its chatbot tells customers?

Courts and regulators are treating it that way. A 2024 British Columbia tribunal made Air Canada pay a customer after its chatbot described a refund policy that didn't exist, and New York City's October 2026 proposed rule would stop businesses from blaming unlawful conduct on their technology.

What does New York City's proposed AI customer service rule require?

Businesses would have to disclose at the start that a customer is dealing with an automated system, offer an easy way to reach a human, and keep records of those interactions. It covers voice menus and rules-based chatbots as well as AI. A public hearing is scheduled for November 19, 2026.

How do I stop my AI agent from making promises I can't keep?

Give it a written list of what it may quote, connect it to your live calendar and price list, and route everything else to a person. Then review a sample of real conversations every week and fix any answer it made up.