New York City wants a human behind every chatbot. What it means for your customer service

What happened

On Tuesday, Oct. 20, New York City's Department of Consumer and Worker Protection proposed a rule aimed at what it calls AI customer service "doom loops." That's the experience of asking a chatbot or phone menu for help and getting sent in circles. The full text of the proposed rule does four main things:

1. Tell people it's a bot. A business must clearly disclose, at the start of the conversation and in the language being used, that the customer is talking to an automated system and not a person.
2. Give them one step to a human. If a business uses only automation, the rule presumes it is breaking the law unless it offers a way out. The main option is a single step, like "Press 0" or "say 'agent'," with the estimated wait time. No extra questions and no reasons required. If no one is available, the business can instead give hours, a phone number, live chat, a callback, a contact form, text or email, with a stated response time.
3. Don't promise help you don't give. It would be deceptive to advertise customer service and then repeatedly send people to menus that don't answer them, or reply with "unresponsive, irrelevant, or erroneous information."
4. Keep records for three years. That includes the contracts for the bot, a log of each conversation showing how long it ran and whether a transfer was offered and used, and yearly totals. One required total stands out: how many conversations came from someone who had already contacted the bot in the past 72 hours.

The rule also says a business can't use "the technology did it" as a defense for breaking any law the department enforces.

The fines and the timeline

Those amounts come from the penalty schedule in the proposal. Each separate item charged counts as its own violation, and businesses could also owe restitution to customers. The public hearing is at 11 a.m. on Nov. 19, 2026, by phone and video. There is no start date yet. Gothamist reports that Commissioner Samuel Levine said timing depends on how many comments come in, and that companies would get a transition period.

Does this reach a company in Dallas?

Maybe, but less often than the headlines suggest. As reported by Gothamist, Levine said the rule would apply to any company doing business in New York City, not just companies based there. So a Texas company that sells to New York shoppers online could be in scope.

But this is a consumer protection rule. Its text talks about "consumers" and "merchants," and the city's notice ties it to consumer goods and services. If you are a distributor or manufacturer that sells only to other businesses, this rule probably doesn't touch you directly. That's our reading, not legal advice. Ask your lawyer if you sell to the public in New York.

New York also isn't alone. The city's notice says disclosure rules like this already exist in California, Maine, New Jersey and Utah. And Levine wants it to spread. As Gothamist quoted him: "We think every jurisdiction should do this."

A bot that can't hand off to a person isn't saving you money. It's moving the cost onto your customer.

Why the city thinks bots fail

The notice leans on survey data. The best known is from Gartner. In a 2023 survey of customers, Gartner found that only 8% had used a chatbot in their most recent service contact. How often the issue got solved depended a lot on the job.

Chatbots handle simple jobs better than hard ones

Share of issues resolved by customers who used a chatbot, by type of issue (Gartner customer survey, 2023).

1. Billing disputesJudgment, account history, money at stake17%
2. Returns or cancellationsA clear rule and a set process58%

Source: Gartner (June 15, 2023), as cited in New York City's proposed rule (Oct. 20, 2026).

Customers also say they don't want to be walled off. In a separate Gartner survey of 5,728 customers, 64% said they would prefer companies didn't use AI for customer service, and 53% said they would consider switching to a competitor over it, as reported by Insurance Canada. Their top worry was that AI would make it harder to reach a person. Both are surveys of how people feel, not measures of what bots actually do. Treat them as direction, not as a scorecard for your own tools.

The other side has a point

Not everyone thinks this helps. Adam Levitin argued on the consumer-finance blog Credit Slips that the rule will push companies back toward human call centers, raise costs, and pass those costs on to every customer, including people who never call. He also notes that reaching a person doesn't guarantee a fix, because human agents may lack the authority or the reason to help. And since companies may not be able to treat New York callers differently, one city's rule could set the standard for the whole country.

Those are fair points. Bots are good at the easy, high-volume questions: order status, hours, tracking numbers, resetting a password. Forcing a human into every one of those would be waste. But the rule doesn't do that. It requires an exit, not a person on every call. And the Gartner numbers say the hard cases, the ones with money and judgment involved, are exactly where bots struggle most.

What changes, and what doesn't

For most mid-size companies, nothing changes legally this month. The rule is a proposal. It is one city. And it is aimed at consumers.

What does change is the bar. A regulator has now written down, in plain terms, what a fair customer-service bot looks like. Customers will start to expect it everywhere, including your B2B buyers, who are consumers at home. The same design is what we'd tell any company to build anyway. In our view, AI that answers customers does best when it handles the routine question, knows when it is stuck, and hands the case to a person with the full history attached, so the customer never has to repeat themselves.

The record-keeping list is also worth stealing. Logging how long each conversation took, whether a hand-off was offered, and how many people came back within 72 hours is a cheap way to find out if your bot is solving problems or just delaying them.

What to do this week

1. Make a list. Write down every place a customer can hit automation: website chat, phone menu, after-hours voicemail, order-status email replies, an AI inbox assistant.
2. Test the exit yourself. On each one, try to reach a person. Count the steps. If it's more than one, fix that first.
3. Say it's a bot. Add one clear line at the start. "I'm an automated assistant. Type 'agent' anytime to reach our team."
4. Route the hard cases to people. Billing disputes, credits, damaged shipments and angry customers should go to a person quickly, with the conversation attached.
5. Start the log. Track conversation length, hand-offs offered and taken, and repeat contacts within 72 hours. Review it monthly.
6. If you sell to New York consumers, have your lawyer read the proposal. Comments are open until the Nov. 19 hearing.

Our guide Human-in-the-loop automation covers how to design the hand-off.