An onboarding assistant should identify itself as AI before it asks for information or gives guidance. Adding that disclosure may delay a launch, but it gives people a clear basis for deciding what they are consenting to.
On Monday morning, with the EU release nearly ready, I would treat that screen as a product decision, not a line for legal review at the end.
The moment the release stopped feeling finished
Consider Marta, a composite founder in Berlin with a thermos on her desk and a release checklist open beside her laptop. Her small team had built an onboarding assistant to guide new customers through setup. The assistant could explain the first steps, ask for missing details and suggest what to do next.
The demo felt good. The copy had been reviewed. The team had planned to release that week.
Then someone raised a plain question: would a new customer know they were speaking with AI before the assistant began asking questions?
That question changed the work.
The upcoming transparency rules make the direction clear for certain interactive AI systems: people need to know when they are dealing with AI rather than a human. Marta could have added a quiet sentence below the input box and kept the launch date. But the assistant was also collecting context that shaped its answers. If the disclosure appeared after the first exchange, the person had already started revealing information before they understood what was responding.
The bad ending was no longer a late release. It was a customer discovering the nature of the interaction halfway through and wondering what else had been left unclear. That kind of doubt stays with a product longer than a missed launch week.
Disclosure changes the conversation design
A disclosure screen is not a warning label pasted onto a finished flow. It defines the first agreement between the product and the person using it.
Marta’s team pulled apart the opening exchange. They asked what the assistant actually needed to say before a customer typed anything. The answer was short: this is an AI assistant; it can help with setup; its responses may be wrong; and the customer can choose another path when the issue requires a person.
That last part mattered. A customer who needs a nuanced answer, has concerns about sensitive information or simply prefers human support should not have to fight the interface to find it.
The first version of the onboarding flow had treated the assistant as the default because it reduced the number of screens. The revised version made the choice visible. The delay came from rebuilding the first few moments, testing the language, and making sure the handoff path worked. None of that was glamorous. It was still product work.
This is close to the problem in AI product demos: What Nia’s real workload taught us about customer trust. A convincing interaction can create trust quickly. The product has to earn the trust it receives.
Timing is part of consent
Teams often frame a disclosure as a compliance question: what exact words must appear? The harder question is when a person needs those words to make a meaningful choice.
If an AI assistant greets someone, asks about their business and adapts its next prompt, the interaction has already begun to shape the person’s behaviour. Put the disclosure before that moment. Do not rely on a settings page, a help-centre article or a small icon that only a careful reader will notice.
The same logic applies to generated or altered content. Labels and machine-readable marks are meant to reduce deception and manipulation. Product teams should decide early where generated content appears, how a person can tell what changed, and what they can do when they disagree with the result.
There is a commercial temptation to keep the AI aspect quiet because a human-sounding experience can appear more polished in a demo. That approach creates a fragile product. It asks people to trust an interaction they have not been given a fair chance to understand.
The week after the delay
Marta’s release did not arrive on the original day. The team used the extra time to watch how people reacted to the disclosure, where they hesitated and which questions belonged with a human.
A few days later, the assistant opened with a direct explanation and two clear options: continue with AI guidance or contact support. The screen had more words than the original. It also left less room for confusion.
That is the trade worth making. A launch date is visible inside a company. The first minute of trust is visible to the person who decides whether to continue.
Before shipping an AI assistant, map every point where it speaks, asks, generates or changes content. Then put the explanation where the person needs it, before the system has earned information they might have given differently.
Comments
No comments yet.