What Should Your AI Disclosure Say? A Small Business Guide for Chatbots, Emails and Phone Agents

Picture a new hire on the front counter wearing a name tag that says "Trainee." Nobody reads that tag as an apology. It tells the customer what to expect and that a supervisor is nearby. An AI disclosure does the same job for software. When a chatbot answers your website, an AI agent drafts your emails or a voice agent picks up your phone, a short, honest line tells customers what is talking to them and who checks its work. This post is about writing that line, and the rules that now expect one.

Key Takeaways

• Short answer: a good AI disclosure tells customers, in plain words at first contact, that they are dealing with AI, what a person checks and how to reach that person. Then it stays true every day.
• Customers want it. In Twilio's November 9, 2026 report, 76% of consumers said an AI agent ought to say it is AI when a conversation begins, yet just 32% of businesses said AI transparency is an active priority for them (Twilio, November 9, 2026).
• The rules are arriving in pieces. The EU AI Act's transparency rules have applied since August 2, 2026 (European Commission, Quick Facts). California's AB 1609 adds chatbot duties for very large businesses from January 1, 2027 (California Legislature; Sheppard, October 27, 2026). Maine and Utah already have their own rules.
• A human option makes it work. Five9 found consumer trust in AI rose from 26% with no human option to 55% when one is reachable (Five9, October 14, 2026).
• My rule of thumb, the Name Tag Test: at first contact, can a customer tell that this is AI, what a person checks and how to reach that person, and is every word true today?

Disclosure: Agentic Trainers is paid to implement AI agents for businesses and to supply the human supervisors who check their work. Arguing that "a person checks this" should be true is an argument for work we sell. Weigh my opinion accordingly, and read the sources yourself.

Why Should You Listen to Me?

I started in 2006 as an outsourced call center agent in technical services, so customer conversations are where my career began. Virtually all of my career since then has been on the technical side of outsourcing, including scaling strategy and the human-capital side of scaling for tech firms, later AI firms, and some semiconductor firms. I joined SuperStaff as CEO in 2019. SuperStaff was founded in 2009 and is now in its 18th year of business. Today I am CEO of SuperStaff and Managing Director of Agentic Trainers, the AI division of SuperStaff, where we build AI agent workforces for businesses and staff the human-in-the-loop teams that supervise them. My graduate AI studies are underway. I am not a lawyer, and nothing here is legal advice. I do own an AI disclosure line: every post on this site carries one.

Why Is Everyone Suddenly Talking About AI Disclosure?

Because the gap between what customers expect and what businesses do is now measured, and regulators are writing it down. Three things landed in the last few weeks.

First, the customer data. Twilio's 2026 Customer Insights Series, released November 9, 2026, surveyed 7,652 consumers and 660 business leaders in 18 markets. Besides the 76% versus 32% gap, 63% of consumers want "a quick, easy way to escalate to a live person at any time" (Twilio, November 9, 2026). Five9's survey of 3,000 consumers in the US, UK and Germany found that "71% of consumers consider it very or extremely important to know when they are interacting with an AI agent" (Five9, October 14, 2026). Both companies sell customer service technology, so treat these as industry research.

Second, small business guidance. On October 12, 2026, KVK, the Netherlands Chamber of Commerce, published an explainer for entrepreneurs. For a chatbot or AI-powered telephone assistant, it says "you must inform customers during the first interaction that they are speaking to an AI assistant" (KVK, October 12, 2026).

Third, the US calendar. California's AB 1609 was approved and filed on September 28, 2026, and on October 27, 2026 the law firm Sheppard summarized it alongside a second California law on synthetic performers in ads, both taking effect January 1, 2027 (Sheppard, October 27, 2026). Washington's companion chatbot rules start the same day. Disclosure is turning from good manners into a line item.

What Is an AI Disclosure Line?

An AI disclosure line is a short statement, shown where the customer meets the AI, that says what the AI did and what a person did. It is not a privacy policy clause.

Here is ours, on every post on this site since November 2, 2026: "Written by Matt Narciso with AI-assisted research and drafting. Every fact and source was reviewed by a human editor." The first sentence names the author and the AI's role. The second names the human check. It does not say "AI-powered," it does not apologize and it does not hide in a footer. It also created an obligation: every post now gets a human read before it goes live, because otherwise the second sentence would be false.

That is the point most businesses miss. A disclosure line is a sentence your process has to keep.

Does My Small Business Legally Have to Disclose AI?

Sometimes, depending on where your customers are, what the AI does and how big you are. None of the federal sources I opened requires labeling everything made with AI, but several rules reach ordinary chat, calls and reviews. Here are the rules I opened for this post: a starting map, not a complete list or legal advice.

Rule	Who it reaches	What it asks	When
EU AI Act, Article 50	Companies that build AI systems people talk to, plus deployers of deepfakes and certain AI text	People are informed they are interacting with AI; deepfakes and certain public-interest text are labeled	Applies from August 2, 2026
California AB 1609	Businesses with national revenue above $500 million a year that serve California customers	No claiming a customer service chatbot is human; disclosure if a reasonable person could be misled; a way to reach a human agent	Approved and chaptered September 28, 2026; duties start January 1, 2027
Maine, 10 MRSA §1500-DD	Anyone using a chatbot or other computer technology in trade with a consumer	Clear and conspicuous notice if a reasonable consumer could think it is a human	Enacted 2025 (PL 2025, c. 294)
Utah Code §13-77-103	Suppliers using generative AI in consumer transactions; regulated occupations	Disclose when clearly asked; regulated occupations disclose prominently in high-risk interactions	Effective May 7, 2025
Washington RCW 19.440.030	Operators of AI companion chatbots	Clear and conspicuous "not human" disclosure at the start and at least every three hours	Effective January 1, 2027
FCC Declaratory Ruling FCC 24-17	Callers using AI-generated voices	Treated as "artificial or prerecorded voice" calls under the TCPA, so consent and caller identification rules apply	Released February 8, 2024
FTC rule on reviews and testimonials	Businesses that write, buy or post reviews	Fake reviews are banned, and the FTC says AI-generated reviews are covered	Effective October 21, 2024

Several of these rules turn on whether a customer could be misled, not on the mere use of AI, and nearly all reward the same habit: say it early, plainly and in the channel where the conversation happens.

What Must a Small Business Do Under the EU AI Act's Transparency Rules?

If you sell to people in the EU, Article 50 of the AI Act is the rule to know, and it has applied since August 2, 2026. The Commission lists four cases: providers must design AI systems so people know they are interacting with AI and must mark AI-generated content in a machine-readable way, while deployers must disclose emotion recognition, biometric categorization, deepfakes and text on matters of public interest "which did not undergo human review or editorial control" (European Commission, Quick Facts).

Which side you sit on matters. If you buy a chatbot, the design duty in the first case sits mainly with the vendor as provider. If you build your own, you may be the provider. KVK's guidance does not split hairs: if customers talk to your chatbot or AI phone assistant, tell them at the first interaction. My view: that is the safer habit either way.

Three more facts worth knowing:

• The Commission published guidelines on July 20, 2026 to assist providers and deployers in "meeting the AI Act's transparency obligations" (European Commission, July 20, 2026).
• Regulation (EU) 2026/1744, adopted July 8, 2026 and known as the Digital Omnibus on AI, gave providers of generative AI systems already on the market before August 2, 2026 until December 2, 2026 to "comply with Article 50(2)," the machine-readable marking duty (EUR-Lex, Regulation (EU) 2026/1744). That is a vendor deadline three weeks away, so expect your AI tools to start marking their output.
• The Commission lists fines for companies of up to €15 million, or up to 3% of their worldwide annual turnover, with "Proportionality for SMEs and SMCs" (European Commission, Quick Facts).

KVK also names exceptions small businesses will like: "The transparency obligations do not apply to printed materials such as leaflets and brochures," and you need not disclose using AI to help code your website or app.

Which US State Laws Cover Chatbots and AI Disclosure?

In my reading, Maine has the broadest rule for ordinary small businesses, Utah has an "if asked" rule plus a stricter one for licensed professions, and California's new law targets very large companies. Washington's new law is about companion chatbots.

Maine. A person may not use an AI chatbot or other computer technology in trade with a consumer "in a manner that may mislead or deceive a reasonable consumer into believing that the consumer is engaging with a human being" unless the consumer is clearly and conspicuously told otherwise. The definition covers "textual or aural communications," so voice counts, and breaking the rule counts as an unfair trade practice under Maine law (Maine Revised Statutes, Title 10, §1500-DD). The text sets no size threshold.

Utah. When generative AI is used while serving a consumer, the business has to say so if the customer asks, and the question must be "a clear and unambiguous request to determine whether the interaction is with a human or with artificial intelligence." Regulated occupations must disclose prominently in high-risk interactions, "verbally at the start of a verbal interaction" and in writing before a written one (Utah Code §13-77-103).

California. Under AB 1609, a large private business "shall not represent that a customer service chatbot is a human," must disclose if a reasonable person is likely to be misled, and must do so "in the medium in which the customer service chatbot interacts with the customer." "Large" means national revenue above $500 million a year. Customers must be able to request a human agent, with a good-faith effort to connect within 15 minutes or book a time within one business day. A first breach can draw a civil penalty as high as $5,000, and any later breach as high as $10,000 (California AB 1609, Chapter 733). Most small businesses are nowhere near that size, but large-company rules tend to become everyone's customer expectations.

Washington. From January 1, 2027, operators of AI companion chatbots must disclose that the bot is "artificially generated and not human" at the start and "At least every three hours during continued interaction" (RCW 19.440.030). A dental office booking bot is unlikely to count, but a product built for ongoing personal conversation should get a lawyer's look.

What Should a Chatbot Disclosure Say?

A chatbot disclosure should name the AI before the customer types, say what it can do and show the way to a person. Put it in the chat window, not behind a link.

These sample lines are illustrative starting points I wrote for this post, not legal templates.

Channel	Sample line	What it tells the customer
Website chat	"You are chatting with our AI assistant. It can answer questions about hours, pricing and bookings. Type 'person' at any time to reach our team."	It is AI, its scope, the way out
Support bot with review	"Our AI assistant drafts answers. A member of our team reviews any refund, cancellation or billing change before it happens."	It is AI, what a person checks
After hours	"Our AI assistant is answering while our office is closed. A person will follow up by 10 AM the next business day."	It is AI, when a person arrives
Asked "Are you human?"	"No, I am an AI assistant. I can connect you with a person now or schedule a call."	A direct answer, then the way out

Three design rules make these work. Use the plain words "AI assistant," because a first name alone can read as a person. Repeat the line in every channel, because a website label does not follow the conversation into text messages. And make "person" actually work: Twilio found that 78% of consumers say they have tried to bypass AI agents to reach a human. If you have not written the handoff triggers yet, my post on AI agent escalation rules covers them.

What About AI Phone Agents and Voice?

An AI phone agent should say it is AI in its first sentence, before it asks for anything. Voice is where people are most easily fooled.

For outbound US calls, the FCC ruled in February 2024 that the TCPA's restrictions on "artificial or prerecorded voice" calls "encompass current AI technologies that generate human voices," so, except in an emergency or where an exemption applies, the caller generally must have the person's advance express permission. FCC rules also require such messages to "state clearly the identity of the business" responsible at the beginning (FCC 24-17, February 8, 2024). That is about who is calling, not an "I am an AI" script, but the habit is the same.

For inbound calls, Maine's definition covers "aural" communication, Utah requires a verbal disclosure at the start for regulated occupations in high-risk interactions, and KVK names the AI-powered telephone assistant as an EU case. An illustrative opening: "Thanks for calling. I am the clinic's AI assistant. I can book or move appointments. Say 'staff' at any time to reach a person."

Do You Need to Disclose AI-Written Emails and Content?

Usually not for routine email a person has reviewed, but deepfakes need labels and fake reviews are off the table. The honest middle ground is a light, accurate line where customers would care how something was made.

The EU text duty is narrow: public-interest text without human review. KVK's example is broader and practical. An estate agent publishes an AI-written property listing without checking it, and KVK concludes, "Then you must state that the text was generated by AI and has not been reviewed by a human." The human review changes the answer, so make the review real.

In the US, the FTC's rule on reviews and testimonials, effective October 21, 2024, bans fake reviews, and the Commission wrote that "AI-generated reviews are covered by the final rule" (Federal Register, August 22, 2024). No disclosure line fixes a fake review.

AI faces and voices in ads are next. Sheppard reports that under California's SB 1050, beginning January 1, 2027, an ad published in California that prominently includes a synthetic performer "must include a clear and conspicuous disclosure that the performer is synthetic" (Sheppard, October 27, 2026). If your next promo uses an AI voice-over narrator, plan the label now.

For AI-drafted email, I suggest a line only when the AI sends without a person reading it first: "This reply was drafted and sent by our AI assistant. Reply 'person' and our team will pick it up." If a person reads and edits every email, the email is theirs.

What Can a Small Business Copy From Our Own Disclosure Line?

Copy the structure, not the words: name the AI's role and name the human check, specifically enough that a customer could hold you to it.

1. Be specific about the AI's job. "Research and drafting" is narrower than "created with AI." Customers can picture it.
2. Be specific about the human's job. "Every fact and source was reviewed" is checkable. If a reader finds a bad source, someone owns it.
3. Let it force a process. Once we published "reviewed by a human editor," a post without a human read became a broken promise.

A plumbing company might write: "Quotes are prepared with AI-assisted estimating. A licensed technician reviews every quote before we send it." An accounting office might write: "Our AI assistant drafts routine replies. A staff accountant reviews any answer about your taxes." Both are illustrative, and both only work if the review happens, which is the core idea in my post on human-in-the-loop agentic AI.

What Is the Name Tag Test?

The Name Tag Test is my rule of thumb, not an industry standard: an AI disclosure passes only if a customer, at first contact, can answer three questions from it, and every answer is true today. Is this AI? What does a person check? How do I reach that person?

Question	Pass	Fail
Is this AI?	"You are chatting with our AI assistant."	A first name and a stock photo
What does a person check?	"A team member approves every refund."	"Powered by advanced AI"
How do I reach a person?	"Type 'person' or call our office."	A contact form three clicks away
Is it true today?	The approval queue is staffed this week	The reviewer left in March

A name tag works because it is visible before the conversation starts, short and accurate. If your disclosure fails any row, fix the line or fix the process. Usually it is the process. Which actions need a human check at all is the subject of my post on AI agent approval rules. A disclosure line is only as honest as that list.

What Are the Most Common AI Disclosure Mistakes?

The most common mistakes are hiding the line, overstating the human role and letting the line go stale.

• Hiding it in the terms. Maine and California ask for "clear and conspicuous" notice, and California wants it in the medium of the conversation. A footer link is neither.
• A bot that claims to be human. A friendly persona is fine. One that insists it is human when asked is exactly what these laws address.
• "Reviewed by our team" when nobody reviews. It reads well and it is false.
• One channel only. The website says "AI assistant," and the text follow-up does not.
• A human who never arrives. Five9 found that 92% of consumers give AI less than 10 minutes before escalating to a human.

Why Do the Numbers on Consumer Trust Disagree?

They disagree because the surveys ask different questions of different people. Twilio's 76% measures whether an AI agent should identify itself at the start, across 18 markets. Five9's 71% measures how important it is to know, across three countries, and Five9 reported 76% in the UK, 69% in the US and 67% in Germany. Both come from vendors, but they agree on direction: most people want to know.

The legal sources differ too. The Commission's summary limits the EU text duty to public-interest text without human review, while KVK's guidance uses a property listing as its example. My view: write to the stricter version until a lawyer says otherwise. It costs one sentence.

What Is the Strongest Argument Against Disclosing AI?

The strongest argument is that disclosure can backfire: some customers trust a correct answer less once they know AI wrote it, and a label on everything becomes noise. Five9 found that 48% of consumers try to use AI as little as possible when seeking customer service. A small business could fairly ask why it should label a correct answer in a way that sends customers looking for the exit.

There is something to this. Labeling the spell-checker would be silly, and KVK's own exceptions show regulators agree.

My verdict, and it is only my view: disclose where a reasonable customer would care, which is when they are talking to the AI, reading something it sent without review or seeing a face or voice that looks real. Skip it for tools that only help a person work. And pair every disclosure with a real human path, because Five9's numbers suggest the human option, not the label alone, is what moves trust.

So What Is the Bottom Line on AI Disclosure?

Back to the trainee's name tag. The tag is not the problem. A trainee pretending to be the manager is. AI disclosure works the same way. Say what is AI, say what a person checks and say how to reach that person, in the channel where the customer is, before they have to ask. Then keep it true. Twilio's numbers suggest most businesses have not prioritized this yet, which means a clear line is still a small competitive advantage.

An AI disclosure is a sentence your process has to keep, so build the process before you write the sentence.

If you want help writing disclosure lines that match how your AI actually runs, or staffing the human checks that keep them true, book a discovery call with us. If you are still choosing where AI fits in your business, start with our AI agents for small business guide.

Frequently Asked Questions

Do I need to disclose that my chatbot is AI?

Often, and it is a good habit everywhere. Maine requires clear and conspicuous notice when a chatbot could make a reasonable consumer think it is human, Utah requires disclosure when a customer clearly asks, and the EU AI Act's transparency rules have applied since August 2, 2026.

What should an AI disclosure statement include?

Three things: that the customer is dealing with AI, what a person checks and how to reach that person. Put it at first contact, in plain words such as "AI assistant," and keep each part true.

Does California's AB 1609 apply to small businesses?

AB 1609 applies to large private businesses, meaning those with national revenue above $500 million a year, that serve California customers. Most small businesses fall below that.

Do AI phone agents have to say they are AI?

Rules vary. Maine's chatbot definition covers spoken communication, Utah requires regulated occupations to disclose verbally in high-risk interactions, and KVK says EU customers must be told during the first interaction. For outbound US calls, the FCC treats AI-generated voices as artificial voices under the TCPA.

Do I need to label emails or blog posts written with AI?

Not always. The EU labeling duty for text applies when AI writes about a matter of public interest and no person has reviewed the text or exercised editorial control. A short line is wise when AI sends content nobody read first. Never use AI for fake reviews; the FTC says AI-generated reviews are covered by its rule.

What is the Name Tag Test?

My rule of thumb, not an industry standard: an AI disclosure passes if a customer, at first contact, can tell that it is AI, what a person checks and how to reach that person, and every answer is true today.

Will disclosing AI make customers trust my business less?

It can for some, which is why the disclosure should come with a real path to a person. Five9 reported on October 14, 2026 that consumer trust in AI rose from 26% with no human option to 55% when one is reachable.

---

Disclaimer: This article is commentary and opinion based on public sources available as of November 10, 2026. It is not legal advice and not a complete list of the laws that may apply to your business. Laws, effective dates and official guidance change; consult a qualified attorney. The Twilio and Five9 figures come from surveys by companies that sell customer service technology. The Name Tag Test, sample lines and business examples are my own heuristics and illustrations, not legal templates, industry standards or client cases. Agentic Trainers is paid to implement and supervise AI agents, a conflict of interest I have tried to disclose.

Written by Matt Narciso with AI-assisted research and drafting. Every fact and source was reviewed by a human editor.

Sources

1. The AI Expectations Gap: Twilio's Report Finds Businesses Are Measuring AI Success on Metrics Customers Don't Care About, Twilio (November 9, 2026)
2. New Five9 Research Reveals Consumers Embrace AI in Customer Service, But Trust Still Depends on Having a Person Within Reach, Five9 (October 14, 2026)
3. Do you need to tell customers that you use AI in your business?, KVK Netherlands Chamber of Commerce (October 12, 2026)
4. Quick Facts: Transparency rules for AI systems, European Commission, Shaping Europe's digital future (last updated July 29, 2026)
5. Commission publishes guidelines on transparency obligations for providers and deployers of certain AI systems, European Commission (July 20, 2026)
6. Regulation (EU) 2026/1744 (Digital Omnibus on AI), EUR-Lex (July 8, 2026)
7. AB-1609 Customer service chatbots, California Legislative Information (chaptered September 28, 2026)
8. Title 10, §1500-DD: Required disclosure of use of artificial intelligence chatbot to engage in trade and commerce, Maine Legislature (PL 2025, c. 294; visited November 10, 2026)
9. Utah Code §13-77-103, Required disclosures, Utah State Legislature (effective May 7, 2025)
10. RCW 19.440.030, Disclosure requirements, Washington State Legislature (effective January 1, 2027)
11. Declaratory Ruling, FCC 24-17, Federal Communications Commission (February 8, 2024)
12. Trade Regulation Rule on the Use of Consumer Reviews and Testimonials, Federal Register (August 22, 2024)
13. California Enacts AI Spokespeople, Chatbots and Content Disclosures Laws, Liisa M. Thomas and Kathryn Smith, Sheppard (October 27, 2026)

---