Can buyers tell it’s an AI?
Yes. Klaviyo’s 2026 consumer research tested exactly this: people flag replies that arrive too fast to be human, read as mechanical or overly formal, or simply feel unnatural. Worse, there is a guilty-until-proven-human bias — if something looks AI-made, most people assume it is. Hiding it is not a strategy. Disclosing it well is.
Do you have to tell them? In the EU, yes.
If your buyers sit in the EU, this stopped being a style question. Article 50 of the EU AI Act (Regulation (EU) 2024/1689) has applied since 2 August 2026: AI systems that interact directly with people — chatbots, voice assistants, agents — must make clear that the person is dealing with an AI, unless that is already obvious. The duty is not limited to “high-risk” systems. Penalties reach €15 million or 3% of worldwide annual turnover, whichever is higher, enforced by national market surveillance authorities (European Commission, 2026).
Scope is what matters for an exporter: the Act reaches providers and deployers that place AI on the EU market, or whose AI output is used inside the EU. Where your company and servers sit is not the test; where your buyer sits is.
One timing detail gets mixed up constantly. The machine-readable marking duty for generative systems already on the market has a transition period to 2 December 2026. The disclosure duty for interactive systems had no runway at all — it applied on 2 August 2026. The Commission adopted its guidelines on 20 July 2026.
China moved the same way. GB/T 47746–2026, effective 1 September 2026, requires a clearly visible route to a human, requires context to travel across the AI-to-human handoff so customers do not repeat themselves, and holds the company accountable for what its AI said. Two rulebooks, one direction: you may use AI, but not in the dark, and never without a way out.
Will disclosure scare buyers off?
Start with the discouraging numbers. AnswerConnect’s tracking study of 6,000 people found that between October 2025 and April 2026, preference for talking to a human rose from 83% to 85%, frustration with AI service rose from 54% to 59%, and the share who hang up when they reach an AI rose from 29% to 31%. In SurveyMonkey’s February 2026 research, 79% of Americans preferred a human and 84% believed humans are more accurate.
Now look one layer down, because the trigger is not the label. A study across 15 countries and 4,800 consumers found that 90% of companies believe their customers are satisfied with AI service while only 59% of consumers agree — a 31-point perception gap. In the same work, 63% want an easy route to a person at any moment and 78% have tried to bypass an AI, while only 47% of companies offer a convenient way through. What angers people is a wrong answer and a dead end, not the machine. Roughly a third of respondents said they will accept AI when it buys them something — a lower price or a faster answer.
Export inquiries land on the accepting side of that trade. Buyers want specs, MOQ, lead time, whether you can sample, and roughly where the price sits. Those are facts, not feelings. An AI that returns an accurate spec in ten minutes beats a person who replies the next working day: 52% of B2B leads arrive outside business hours (Salesforce), and replying within 5 minutes rather than 30 makes a lead about 21 times more likely to qualify (MIT/InsideSales).
How do you disclose without losing the buyer?
- Say it first, before they ask. One line, then straight into the answer they came for — the buyer cares whether you can answer, not about your identity statement.
- Keep a one-tap route to a person, and carry the context across. Zendesk CX Trends 2026 found 74% of consumers are frustrated by having to repeat themselves to a second agent, and a six-country study found 52% of Chinese consumers hit context loss when escalating from AI to a human. Pass the transcript, the confirmed facts and the reason for the handoff.
- When it cannot answer accurately, it should not answer. Parking an unverified spec, price or lead time for a human is far safer than a fluent guess — inaccurate or misleading answers are the single biggest complaint people have about AI conversations.
What should the AI never say on its own?
- Price exceptions and discounts — a range is fine, breaking the floor is not.
- Payment terms and credit — money risk belongs to a person.
- Exclusivity and territory protection.
- Contract wording and liability.
- Quality disputes and complaints — escalate on the first message.
Write that list down before launch. It is a configuration, not something to leave to the model’s judgement.
A disclosure line you can copy
“Hi, I’m the AI assistant for [company]. I can confirm specs, MOQ and lead times right away — ask for a colleague any time and I’ll bring one in.”
That does three jobs in one breath: states what it is, states what it can do for the buyer, and shows the exit. Keep the same three jobs in every language you sell in.
How DeepSales handles it
DeepSales introduces itself as an AI in the opening line, answers only from the product data you supplied, routes price exceptions, payment terms, exclusivity and complaints to your rep, and keeps every conversation logged and reviewable. More on the guardrails in our FAQ, or book a demo and hear how it opens on your own catalog.