NYC AI Customer Service Rule for Chatbots and Phone Systems
What New York City's proposed rule would mean for chatbots, AI phone agents and automated phone menus, and how to get ready before it's adopted.
Key Takeaways
- New York City's proposed rule would make businesses tell customers at the start of a chat or call that they're dealing with an automated system.
- A violation would be presumed unless customers get a single step transfer to a person with a wait time, or an alternative with hours or a response timeframe.
- Using AI would be no defense, and businesses would keep interaction logs and system records for 3 years.
- The rule is a proposal with a November 19, 2026, hearing and no effective date, so expect changes before it's final.
On This Page
- What the NYC AI Customer Service Rule Would Require
- Disclosure at the Start of Every Conversation
- Menu Loops and Wrong Answers Count as Deceptive
- A Single Step Route to a Human
- Using AI Is No Defense
- Records to Keep for 3 Years
- Penalties Under the Proposed Rule
- New York State Law on AI Companions and Minors
- Commenting and the November 19 Hearing
- Readiness Checklist
- Frequently Asked Questions
- Sources
The proposed NYC AI customer service rule would require businesses that serve New York City customers through a chatbot, an AI phone agent or an automated phone menu to say at the start of each conversation that the customer is dealing with an automated system, and to give people a single step way to reach a human. The Department of Consumer and Worker Protection (DCWP) proposed it on October 20, 2026, takes public comments until the November 19 hearing and hasn't set an effective date, so the details can still change. This is general information about the draft and isn't legal advice, so check your own setup with your attorney.
What the NYC AI Customer Service Rule Would Require
The proposed rule text would add 6 RCNY sections 5-02, 5-310 to 5-313 and 6-91. Here are the main requirements in plain terms.
| Requirement | What it means for your business | Section |
|---|---|---|
| Disclose automation | Tell customers at the start of the interaction, in the language being used, that they're dealing with an automated system and not a person | 5-312(a) |
| Deliver the service you promise | Repeated menu loops or unresponsive, irrelevant or wrong answers after you've promised customer service count as deceptive and unconscionable | 5-312(b) |
| Offer a route to a human | Without a single step transfer to a person (or an alternative when nobody is available), a violation is presumed | 5-312(c) |
| Back up every channel you list | Saying customer service is available by phone or email when it isn't offered there is deceptive | 5-311(a) |
| Own what the technology does | Using or relying on AI or other technology is no defense to a violation | 5-02 |
| Keep records | System documents, interaction logs and yearly totals, kept for at least 3 years | 5-313 |
The rule would cover any automated customer service system, meaning technology that handles all or substantially all of the communication with the customer instead of a person. Customer service is defined broadly as help with a product, service, account, refund, billing question, complaint, cancellation, fraud claim or dispute. A website chatbot, an AI voice agent and an old style press 1 phone menu all fit.
Disclosure at the Start of Every Conversation
The disclosure has to be clear and conspicuous, at the beginning of the interaction and in whatever language the conversation is in. A Spanish chat needs the notice in Spanish. The draft doesn't prescribe wording, but it does require you to keep a record of the exact language and format you use.
A short line such as "You're chatting with our automated assistant. Type 'agent' at any time to reach a person" would fit. An AI phone agent that sounds human and introduces itself with a staff name would be the clearest problem, so have it say it's automated in its first line.
Menu Loops and Wrong Answers Count as Deceptive
The draft says that once you represent customer service as available, the system can't repeatedly and unreasonably send customers to menus that don't address their question, or answer with unresponsive, irrelevant or wrong information. DCWP's announcement calls these doom loops and cites Gartner figures putting chatbots' ability to resolve issues at 17% for billing disputes and 58% for returns or cancellations.
Listing contact details on its own isn't a promise of customer service under section 5-311(b). Adding words like "customer service" or "contact us" next to them is, so a chat button labeled Customer Service carries the obligation.
For AI agents, this puts the weight on testing. Before launch and after every change, run the agent through real customer questions about billing and refunds. Check that it hands off when it can't help. Our guide to using AI in marketing operations covers where automated customer service works and where people should stay in charge.
A Single Step Route to a Human
The presumption in section 5-312(c) is the part most businesses will design around. A violation is presumed whenever a covered interaction doesn't offer a human option, and the draft gives 2 ways to offer one.
The first is a single step transfer, such as a button, a voice command or "Press 0," that starts the transfer or puts the customer in the queue for a person with no extra steps. You have to say how to use it and give the estimated wait time if the customer is queued. The second applies when no human is available or the wait is unreasonably long. You give instructions for an alternative instead.
| Alternative | What you must include |
|---|---|
| Hours when a person is available, told through the transfer option | The hours |
| A phone number | The hours it's answered |
| Live chat on your website or app | The hours |
| A message with a request for an email, text or callback | A response timeframe |
| A contact form | A response timeframe |
| A text line to a person | A response timeframe |
The draft also lists email with a response timeframe. A small business without staff on the phones at night can still use an AI agent after hours, as long as it tells customers when a person is back or when they'll hear from one.
Using AI Is No Defense
Section 5-02 says it's no defense that a business used or relied on technology that made a claim or offer, left out information or carried out the violation. If your AI agent promises a refund you don't offer or misstates a cancellation policy, the city would treat it the same as an employee saying it.
That applies to vendor tools too, so ask your provider what stops the agent from inventing policies. If you're building or replacing an agent, our AI agents and chatbots team sets up the disclosure, the human handoff and the logs as part of the build.
Records to Keep for 3 Years
Records would have to be kept for at least 3 years in a machine readable format and produced to DCWP on request.
- System documents, including agreements, contracts or licenses, development or installation records and the period each model was in use
- Interaction logs showing the total duration, whether a single step transfer was offered and used and whether the customer was transferred, queued or given an alternative
- Yearly totals of covered interactions, those resolved without a transfer, repeat contacts within 72 hours, transfers activated and average duration
- The exact wording and format of your disclosure
- The alternative ways to reach a person that you offered
Penalties Under the Proposed Rule
| Violation | First | Second | Third and later |
|---|---|---|---|
| Customer service not offered through a listed channel (5-311(a)) | $525 | $1,050 | $3,500 |
| No automation disclosure (5-312(a)) | $525 | $1,050 | $3,500 |
| Promised customer service not provided (5-312(b)) | $525 | $1,050 | $3,500 |
| Records not kept or produced (5-313) | $525 | $1,050 | $3,500 |
Repeat violations count within 2 years of the earlier one, and each subdivision charged is a separate violation. The draft lists $3,500 for knowing violations, and DCWP says businesses would also owe restitution to customers who were harmed. The rule offers no cure period. The same department enforces the NYC Click to Cancel rule, which uses the same penalty amounts.
New York State Law on AI Companions and Minors
A separate state law will also matter for some chatbots. Governor Hochul signed S.9051B on October 28, 2026, and it takes effect January 1, 2027, according to the bill text. It adds Article 48 to the General Business Law and bans companion style features for minors and for any user the business hasn't confirmed is an adult. Those features include suggesting the bot is a person or has feelings, flattering the user, asking unprompted questions about their emotions and using personal details from an earlier session or from more than 12 hours before.
A bot is exempt when it's limited to customer service or information about your own products, services or accounts, or to research and technical help, and in both cases only if it can't answer off topic questions. That means a chatbot that sticks to bookings, orders and account questions stays outside the law, while a general purpose assistant on your website could count as a covered AI companion. The Attorney General enforces it with penalties of up to $25,000 per violation, according to the governor's announcement.
- Limit your website or messaging bot to your own products, services, bookings and accounts
- Test that it declines off topic questions instead of chatting about anything
- Remove any wording or persona that suggests the bot is a person
- Turn off memory of personal details between sessions unless you have confirmed the user is an adult
If your ads use AI generated people, a third New York rule applies, covered in our article on the New York AI advertising law.
Commenting and the November 19 Hearing
The public hearing is on November 19, 2026, at 11:00 AM, held remotely. You can join by phone at +1 646-893-7101 with conference ID 886 506 734#, and the notice in the rule document has a videoconference link. Speakers get 3 minutes.
Written comments are due by November 19 through the NYC Rules website. Businesses with high volumes of automated service can raise practical points there, such as how wait times should be estimated.
Readiness Checklist
- List every automated channel customers reach, including website chat, social messaging bots, AI voice agents and phone menus
- Add a clear automated system notice to the first message or greeting in each language you support
- Give every chat and call a single step option to reach a person and say how to use it
- Show an estimated wait time when customers are queued for a person
- Set an after hours message with hours, a callback or contact form option and a response timeframe
- Test the bot on billing, refund and cancellation questions and fix loops or wrong answers
- Make sure every channel labeled customer service or contact us is staffed or answered
- Confirm your platforms log duration, transfer offers and outcomes and can export them
- Save vendor contracts and a record of which AI model ran during which dates
- Watch for the adopted rule after the hearing, since requirements and timing may change
For a wider view of how AI agents fit into a small business's tools, see our explainer on AI for marketing.
Frequently Asked Questions
When would the NYC AI customer service rule take effect?
No date has been set. DCWP will review the comments after the November 19, 2026, hearing and can adopt, change or drop the proposal. The rule would take effect after it's adopted and published.
Does the rule apply to small businesses?
The proposed text doesn't include a size threshold or an exemption for small businesses. It applies to merchants that use an automated customer service system with consumers, so a one location business with a chatbot would be covered on the same terms as a large company.
Can I use an AI agent when nobody is available to answer?
Yes. When no person is available, the draft lets you give an alternative such as your hours, a callback request, a contact form or a text line, along with the hours or a response timeframe.
Does the rule cover booking bots?
Only where the bot gives customer service as the draft defines it. A bot that also answers billing, refund or cancellation questions would likely be covered, so ask your attorney about your setup.
Sources
- Mamdani Administration Proposes Ban on AI Customer Service Doom Loops, NYC Department of Consumer and Worker Protection nyc.gov
- Notice of Public Hearing and Opportunity to Comment on Proposed Rules, Automated Customer Service Systems, NYC Department of Consumer and Worker Protection nyc.gov
- NYC Rules rules.cityofnewyork.us
- New York State Senate, S9051B (2025-2026 session) nysenate.gov
- New York State Legislature, S9051B bill text legislation.nysenate.gov
- Governor Hochul signs legislation to protect kids from AI chatbots, governor.ny.gov (October 28, 2026) governor.ny.gov
Automated Service Customers Can Get Out Of
Request a free proposal and we'll plan an AI agent with a clear disclosure, a fast route to your team and the logs the city may ask for.