The daily publication · October 28, 2026 · Small business

New York just told chatbots: let the customer reach a person

By Tabron Merryhill · The Workflow

New York City just put a price on the chatbot loop. If your automated service traps a customer, the city wants a human on the other end of the exit.

The idea

On October 20, 2026, the city's Department of Consumer and Worker Protection proposed a rule for automated customer service. It covers AI chatbots and phone voice systems, not just one app. The rule would do four plain things.

First, say it up front. A business must disclose at the start that the customer is talking to an automated system, not a person.

Second, offer a real exit. A customer must get a simple, one-step route to a human, such as pressing 0 or saying "agent." If there is a queue, the business must give an estimated wait time. If no human is available, it must give another route to one, such as published hours, a callback, an email, a text, or a form with a stated response time.

Third, stop advertising help you do not give. It would be a deceptive practice to promise customer service and then repeatedly send people through menus that do not answer their questions, or answer with information that is unresponsive, irrelevant, or wrong.

Fourth, no hiding behind the tool. A business cannot dodge responsibility by saying the technology made the claim, left out the information, or carried out the act. The business answers for what its system does.

The city cites Gartner research that chatbot resolution rates run from just 17 percent for billing disputes to 58 percent for returns or cancellations. That gap is the whole point. A bot that handles a return fine can still fail the customer with a billing error, and that customer is the one who needs a person.

Fines would start at $525 for a first violation, rise to $1,050 for a second, and reach $3,500 for a third or later violation, plus restitution to harmed customers. This is a proposal, not a final rule. The city scheduled a public hearing for November 19, 2026, and a 30-day comment period, and it says businesses would get a transition period if the rule is adopted.

The example

Picture a shop owner who adds a chatbot to answer questions and take refund requests. It works well for hours, directions, and simple orders. Then a regular customer gets charged twice.

The bot offers the refund page again. The customer types "refund" again. The bot offers the refund page again. There is no button for a person, no callback, and no email with a response time. The customer gives up, disputes the charge with the bank, and does not come back.

Under the proposed rule, that loop is the violation, not just bad service. The fix is not to scrap the bot. The fix is to build the exit first: one step to a person, a wait time if there is a line, and a callback or form with a response time when no one is on duty. The business must also keep machine-readable records of these interactions for three years, including whether a transfer was offered and used and whether the customer reached a person.

Here is the fair counterpoint, and it matters for owners. Georgetown law professor Adam Levitin argues that a human transfer does not guarantee a fix, because many call center workers follow scripts and lack the authority to solve the problem. He also warns that the cost of human staffing may show up in higher prices for every customer. Steven Fulop, CEO of the Partnership for New York City, adds that the rule leaves open how to judge whether a chatbot actually solved a problem, which could raise litigation risk. A transfer to a person who can only read the bot's script back is not the help the rule promises, and it is not the help that keeps a customer either.

The rule would also reach businesses that serve New York customers even if they are not based in New York. That is why it matters beyond one city. A standard written for the country's largest consumer market tends to travel, and the owners who build the exit now will not have to rebuild their support system city by city.

The next step

Test your own front door this week, as a customer would.

Open your chatbot or phone menu and try to reach a person in one step. Time it. If you cannot find the exit, your customers cannot either. Add a clear "talk to a person" option, a wait time when there is a queue, and a callback or form with a response time when no one is available.

Then give that person real authority. A human who can fix a refund, correct a bill, or cancel an order is the exit. A human who can only take a message is another loop with a voice attached.

Finally, count the repeats. If the same customer comes back within three days about the same problem, the first conversation did not fix it, no matter what your bot dashboard says. That repeat count is the number to watch, and it is the number this rule is really about.

Want help finding the one bottleneck worth handing to AI first? A free 10-minute workflow conversation will map where the hours go and who signs off before anything final. No pitch, just the map.

Sources: City of New York, Department of Consumer and Worker Protection, https://www.nyc.gov/site/dca/news/072-26/mamdani-administration-proposes-ban-ai-customer-service-doom-loops; LexBlog, https://www.lexblog.com/2026/10/26/new-york-city-proposes-sweeping-rules-for-automated-customer-service-systems-including-mandatory-ai-disclosure-and-access-to-human-representatives/; The Cool Down, https://www.thecooldown.com/green-tech/new-york-ai-chatbot-regulations-businesses/. Opposing view: Adam Levitin, Credit Slips, https://creditslips.org/2026/10/20/how-does-mandating-human-customer-service-help-consumers/; Steven Fulop, Partnership for New York City, quoted by Bloomberg Law, https://news.bloomberglaw.com/banking-law/mamdani-targets-companies-with-ai-customer-service-doom-loops.