Every business that puts an AI in front of customers reaches the same fork within about a week. The system works, the conversations are going well, and somebody asks the obvious question: do we have to tell people? The honest reason it gets asked is that hiding it feels like it performs better.
Two things have changed that calculation. One is that in a growing number of places the choice is no longer yours. The other is that the comparison most people are making in their heads is the wrong one.
The Instinct Is to Hide It
The reasoning behind not disclosing is not stupid. People still associate automated systems with the menu-driven chatbots that wasted their time for years, and announcing one at the top of a conversation invites that association before the customer has any evidence to the contrary. There is a real, measurable cost to the word "AI" in a first message.
The problem is what happens next. An undisclosed AI has to keep being undisclosed for the whole conversation, and it usually does not manage it. Somebody asks something slightly outside its range, or the replies come back a shade too fast and too tidy, or they ask directly. Then the customer is not evaluating an automated system, they are evaluating a business that tried to pass one off as a person, and that is a different and much more expensive judgement.
The choice is not between disclosed and unnoticed. It is between disclosed at the start and discovered in the middle.
Where Disclosure Is Already Required
This has moved from a question of taste to a question of compliance faster than most businesses have noticed, and it has moved unevenly, which is what makes it easy to get wrong.
Voice is where the rules arrived first, because automated calling has a long regulatory history to attach to. California now requires calls that use an artificial voice to disclose that fact at the start of the call. The EU AI Act places transparency obligations on AI systems that interact with people directly, with the general expectation that a person is told they are dealing with an AI unless that is already obvious from the context.
Neither of those is a worldwide rule, and plenty of jurisdictions have nothing specific at all. That is precisely the trap. A business calling or messaging across borders is subject to the rules where the recipient is, not where the business is, so the practical question is not "what does my country require" but "what applies everywhere I actually contact people". This is worth ten minutes with someone who knows the law in your markets rather than a guess, because the penalties attach per call.
What Disclosure Does to a Conversation
The effect is smaller than the fear of it, and it is not evenly distributed. A short disclosure at the start of a chat conversation is barely noticed, because people have been messaging businesses through automated systems for years and the bar for the interaction is set by whether it helps, not by what it is.
On a phone call it is more noticeable, because an unexpected call from a business is already a slightly adversarial situation and a person is deciding in the first three seconds whether to stay on the line. Some will hang up on hearing it. It is worth being clear-eyed that this is a genuine cost.
What it buys is that everything after those three seconds happens on honest terms. Nobody is going to feel deceived later, nobody is going to ask "wait, am I talking to a robot" halfway through giving you their availability, and the business is not exposed on a channel where the rules are tightening in one direction only.
An outbound follow-up call opens with roughly this: it is an AI assistant calling from the business, the call is recorded, and it is following up on the enquiry they sent on Tuesday about a boiler service. That is the disclosure, the recording notice and the reason for the call in one breath, and the last part is the one that keeps the person on the line. It works because it moves immediately to something specific to them, rather than pausing for a reaction.
How to Word It Without Killing the Exchange
Most disclosures fail not because they disclose but because they are written like a legal notice, which signals bureaucracy before the customer has heard anything useful.
- Put it first, then move on. Say it in the opening line and go straight to the specific reason for the contact. A disclosure that is followed by a pause invites a decision. One that is followed by relevant detail does not.
- Use ordinary words. "I am an AI assistant" is fine. "This communication is generated by an automated artificial intelligence system" is a sentence nobody has ever said out loud, and it makes the whole thing sound like a warning.
- Name the business in the same breath. The disclosure is only unsettling when it arrives without context. Attached to a business the person already contacted, it reads as organised.
- Offer the exit and mean it. A person who wants to speak to a human should be able to get one by saying so. A responder that acknowledges the request and then keeps going is worse than no disclosure at all.
- Do not overclaim what it is. Giving the AI a human name and a fake job title while also disclosing that it is an AI is a strange middle ground that satisfies nobody.
Common Questions
Am I legally required to disclose?
It depends on where the person you are contacting is, and on the channel. California requires disclosure for calls using an artificial voice, and the EU AI Act imposes transparency obligations on systems that interact with people. Neither is universal, so check what applies in the markets you actually contact rather than assuming.
Does disclosing hurt conversion?
A little on voice, very little in chat, and the comparison is wrong anyway. The alternative is not going unnoticed, it is being noticed halfway through, which costs more.
Where in the conversation does it belong?
The first message or the opening line of the call. A disclosure that arrives after the customer has already worked it out is not a disclosure, it is a confirmation.
Does chat need the same treatment as voice?
The requirements are generally lighter, and people are more used to automated messaging. Being clear in the first message is still the safer position and costs almost nothing when the responder is actually useful.
The Practical Position
Disclose, early, in plain words, and spend the effort on making the thing worth talking to rather than on concealing what it is. That position is defensible in every market, it does not need revisiting each time a rule changes, and it removes an entire category of risk for a cost that is real but small.
The businesses that get burned here are the ones treating disclosure as a marketing decision. It stopped being one. LeadOro builds response systems that identify themselves properly, hand over to a person when asked, and stay useful enough that being told does not cost you the conversation.
- Setup time: 7–14 days from the first call.
- Pricing: plans start at $500 setup + $150/month, month-to-month, no contract.
- Guarantee: 14-day satisfaction guarantee.
None of the above is legal advice, and the rules in this area are being written faster than most businesses are reading them. If you are calling people in more than one country, that is worth an hour of somebody's time who knows the answer properly.
Automation your customers can be told about.
Book a free 15-minute discovery call. We will go through what your responder should say about itself, and where a conversation should stop being automated at all.