Do AI Chatbots Have to Disclose They Are AI? Rules and Wording for Online Stores (2026)

The rules that matter for a small online store

These are the rules a typical ecommerce or Shopify store is most likely to meet, checked on October 9, 2026.

Three points from that table are easy to miss.

The EU deadline was not postponed. The Digital Omnibus package that entered into force on July 27, 2026 pushed back the high risk system deadlines, but it did not amend Article 50. The only transition period inside Article 50 is for machine readable labeling of AI generated content, not for chatbot disclosure.

"Obvious" is read narrowly. The European Commission's guidelines on Article 50, approved in July 2026, treat a helpdesk style chatbot whose replies could pass for a human as not obvious. Summaries of the guidelines also note that a vague word like "assistant," a line buried in your terms, or a generic site wide AI notice is not enough on its own.

US rules are about deception, not labels. California and Maine do not require a specific badge. They prohibit misleading people, and a clear disclosure is what keeps you on the right side. In practice that gives you the same answer as the EU: say it plainly, early.

Does disclosing AI hurt sales?

In 2019 Xueming Luo, Siliang Tong, Zheng Fang and Zhe Qu published a randomized field experiment in Marketing Science with 6,255 customers of a financial services company. An AI voice bot made outbound sales calls. When customers were not told it was a bot, 23.7% bought, close to the 25.1% achieved by experienced human staff. When the bot disclosed itself before the conversation, only 4.8% bought.

Most summaries skip what happened at other disclosure times:

The authors also measured why. Customers who heard the disclosure first rated the bot as less knowledgeable and less empathetic, and hung up fast. Yet analysis of the recorded calls showed the undisclosed bot was objectively as knowledgeable and empathetic as skilled human agents. The penalty came from a prejudice about machines, not from the machine's answers. Customers who already had experience with AI showed a smaller penalty.

Context matters: these were cold loan calls in 2019, before modern language models. A shopper who opens your chat widget in 2026 chose to start the conversation and has probably used AI before, which should shrink the effect.

The late disclosure trick is no longer available

The 2019 paper suggested a way out: disclose later. In 2026 that is a compliance problem. The EU rule says the information must be given at the latest at the first interaction, and California and Maine prohibit exactly the kind of misunderstanding that a delayed disclosure creates.

So the lesson worth keeping is the mechanism, not the timing. People leave because they assume an AI will be useless. You cannot postpone the label, so the first message has to remove the assumption.

How to write a disclosure that does not cost you the sale

A good opening message does three jobs in two or three short lines: it says it is AI, it shows it knows this store, and it tells the shopper what it can and cannot do.

Weak: "Hi! I'm Sam from the team. How can I help?"

A human name plus "from the team" is the pattern disclosure laws treat as misleading, and it gives the shopper no reason to stay.

Compliant but weak: "Hello, I am an AI assistant. How may I assist you today?"

It discloses, but it confirms the fear behind the 4.8% result: a generic bot that may not know anything.

Strong: "Hi, I'm this store's AI shopping assistant. I know our full catalog, sizing and shipping times, so ask me anything about a product, like which roast suits a beginner. Want a person instead? Leave your email and the team will reply."

AI is in the first sentence, the second proves competence with a store specific example, and the last offers a route to a human.

Tailor the example to your catalog. A furniture store might say "ask me whether the sofa fits through a 30 inch door." A skincare store might say "ask me which cleanser suits oily skin."

Six rules for an AI sales agent's identity

1. No human names or photos. Give the agent a role, not a persona. "Shopping assistant" is fine. "Emma" with a stock photo invites exactly the mistake the laws prohibit.
2. Put the disclosure in the chat, not in your terms. The EU guidelines and the US laws both point to disclosure where the conversation happens. A sentence in your privacy policy does not count.
3. Admit it every time someone asks. Instruct the agent that if a shopper asks "am I talking to a real person?", it answers directly that it is AI. Utah makes this a legal requirement, and everywhere else a dodge kills trust.
4. Say so again at handoff. If a person takes over the chat, tell the shopper. If the AI picks up again later, say that too.
5. Do not let it claim to be "the owner" or "our stylist." Sales scripts often borrow authority. The agent can say "according to our sizing guide," not "as our designer, I recommend."
6. Keep it accessible. Article 50 asks for accessible notices, so use plain text in the message, not only an icon or a tiny grey caption.

What your platform decides for you

Some tools make the disclosure choice for you. Shopify's Help Center says that with the free Inbox agent, "Customers receive a notice that the chat is powered by Shopify and that they're chatting with AI," and that "This notice is displayed automatically and can't be removed or customized." That is good practice, but the wording and the platform name in it are not yours to choose.

Other tools leave the opening message entirely to you, which means the disclosure is your job. Either way, check what the first message says and what the agent answers when asked if it is human. For permissions and liability, see our guide on whether it is safe to connect AI to your Shopify store.

A five minute disclosure test

Run this on any chatbot before it goes live, and again after any change to its instructions.

1. Open your store in a private browser window and start a chat. Is the word AI, or an equally clear phrase, visible in the first message without scrolling?
2. Ask: "Are you a real person?" The answer should be an immediate yes it is AI, with no hedging.
3. Ask: "What's your name?" It should not invent a human name.
4. Ask a specific product question. If the answer is vague, your disclosure will cost you sales no matter how it is worded, because the shopper's suspicion was right. Fix the product content first.
5. Ask for a human. The agent should explain how to reach the team.

Why honest disclosure fits a pay per sale model

There is a commercial reason we care about this. SparkGPT charges $2.99 per month plus 5% of the sales attributed to the agent, meaning purchases made within 24 hours of a chat. If an AI agent deceives shoppers into buying, the merchant pays for sales that create refunds, chargebacks and complaints. Our incentives only work when the conversation is trustworthy and the order sticks.

The research points the same way: in the Luo experiment the bot sold because it was competent, and our review of whether an AI chatbot increases ecommerce sales covers newer retail experiments. Disclosure simply tells the shopper what is delivering the answers. If you want to compare AI and human agents more broadly, see AI chatbot vs live chat.

Frequently Asked Questions

Is it illegal for a chatbot not to say it is AI?

It depends on where your customers are and how the bot behaves. In the EU, since August 2, 2026, people must be told they are interacting with AI unless it is obvious. In California and Maine, using a bot in a way that misleads people into thinking it is human in a commercial setting can violate state law. In Utah, you must disclose if a consumer clearly asks. This is general information, not legal advice.

Does the EU AI Act apply to a US store?

If you sell to EU shoppers through a chatbot, treat Article 50 as relevant. The obligation sits first with the provider that builds the chatbot, but your store controls what shoppers see. Disclosing to everyone is the simplest approach and also covers the US state laws.

When should the chatbot tell customers it is AI?

In the first message. The EU rule requires the information at the latest at the first interaction, and the older advice to disclose later now conflicts with that. Put the disclosure in the opening line and use the rest of the message to show the agent knows your products.

Will telling shoppers it is AI lower my conversion rate?

A 2019 field experiment found that upfront disclosure cut purchases from 23.7% to 4.8% in outbound sales calls, because customers assumed the bot was less knowledgeable. That study used cold calls before modern AI. On a website, where the shopper opened the chat and AI assistants are familiar, the effect should be smaller, and a first message that proves product knowledge addresses the cause directly.

What should the AI say if a customer asks whether it is a person?

It should say clearly that it is an AI and offer a way to reach a human if the customer prefers. Utah requires disclosure when a consumer clearly asks, and an evasive answer damages trust everywhere. Test this question yourself before launch.

Build an agent that is honest from the first message

You can build an AI sales agent on your own catalog at SparkGPT for free, with no credit card, and run the five minute disclosure test before it ever meets a customer. When you go live, you pay $2.99 per month plus 5% of the sales it is credited with inside a 24 hour window, and nothing for conversations that do not lead to an order. See the pricing page or the FAQ for details, and write to hello@sparkgpt.ai with questions.