Permission requested after someone begins speaking cannot protect the sensitive detail already disclosed. For a voice assistant, the cloud choice needs to appear before the first sentence that may need to leave the device.
At 9:18 p.m., Abena sat on the edge of her bed in Kumasi with her phone in one hand and her younger brother’s voice note open in the other. Their auntie had sent money for a family matter, then gone quiet. Abena pressed hold to speak and began in Twi: “Auntie no, ɔkaa sɛ…” She was trying to explain why the payment had become tense, who had seen the message, and why her brother was frightened of making the wrong move.
Halfway through, the app stopped her. It said the next step might involve the cloud and asked for permission.
The warning arrived too late. Abena had already said the part she had hoped to keep close: a family name, the payment problem, and the worry underneath it. She could choose what happened next, but she could not take back the first sentence.
That is the practical problem with late consent. A prompt can look careful while leaving the person with no meaningful control over the disclosure that came before it.
A private conversation begins before the sensitive detail feels sensitive
People rarely open an assistant with a neat, labelled request. They start where the thought starts.
“Please help me reply to my auntie.”
Then the context comes out. A name. A balance. A voice note that changed the mood at home. A small detail that makes the answer possible, but that the speaker may not have planned to share beyond the device.
This is especially easy in Twi-English conversation. The important instruction may arrive in Twi after an English opening, or the emotional part may emerge when someone switches languages because that is how the story makes sense. What Happens When the Important Instruction Arrives in Twi? explores why that detail cannot be treated as an optional extra.
A consent prompt after the first disclosure asks people to make a privacy decision with the most private part already behind them. It turns permission into a cleanup task.
The cloud choice belongs before the conversation moves forward
A useful consent control gives a clear choice at the moment it matters: before content that may be sent to the cloud. The wording should be plain enough to understand while someone is holding a phone, tired, worried, or midway through explaining a family issue.
Nkomo makes that choice explicit. A person can choose never, ask each time, or allow cloud use for this session. The point is not to make privacy settings feel impressive. The point is to let someone decide before they begin sharing context.
That clarity matters because “yes” can mean different things depending on the moment. Someone may be comfortable asking for help with a greeting in Twi. They may want a separate decision before describing a family payment, a health concern, or a message that names another person.
Consent also needs to stay understandable when the app cannot continue as expected. Silent failure creates its own privacy and trust problem: the speaker does not know what happened to their request, or whether the app heard it at all. Nkomo shows errors rather than swallowing them, so the person has something concrete to respond to.
Privacy control has to continue after the reply
The first prompt is only one part of the decision. Conversations leave traces, and people deserve to know where those traces are.
Nkomo keeps conversation history on the device when history is enabled. Turn it off, and that history is purged immediately. There is also one-tap export and account deletion. These controls do not erase the need to think before speaking, but they give people a direct way to manage what remains under their control.
This is where vague privacy language causes harm. “Your privacy matters” does not tell someone whether a conversation stays on their phone, whether it may reach the cloud, or how to remove it. Specific controls do.
Ghana’s Data Protection Commission guidance identifies AI as a technology that can affect privacy and data security and deserves impact-assessment attention. That is a reminder to treat consent as part of the product experience, not a screen people rush past before the product begins.
Make the first pause useful
Back in Kumasi, Abena starts again, this time after seeing the choice first. She decides what she is comfortable sending before she explains the family situation. Then she speaks in the mix of Twi and English that comes naturally to her, pauses to correct herself, and continues.
The stakes have changed. She is still dealing with an awkward family matter, and the right reply still matters. But she is no longer discovering the privacy decision after her voice has already carried the detail she meant to protect.
A good voice companion should leave room for that pause. Before the story gets personal, show the person where their words may go. Let them choose. Then let them talk.
Comments
No comments yet.