Choose an AI companion that clearly tells you when a conversation will reach the cloud and lets you decide before it happens. Look for controls that separate cloud consent from local chat history, so a private exchange does not remain on your device longer than you want.
At 7:40 on a Thursday evening, Kwesi is at his kitchen table in Brixton with a mug gone cold beside his laptop. He is flying to Kumasi soon to see family, and he has been practising the phrases he keeps losing halfway through a sentence.
He holds his phone and tries again: “Mɛpɛ sɛ meka sɛ, I’m happy to be home, but I’m shy small.”
The reply comes back in natural Twi and Ghanaian English. Kwesi repeats it, interrupts himself, laughs, and tries a less formal version. For twenty minutes, it feels like the kind of practice he needs before Auntie corrects him across the dinner table.
Then his phone buzzes.
A message from his sister changes the conversation. She says their uncle has been unwell, and asks Kwesi not to mention it around the younger cousins until the family agrees on what to say. Kwesi starts composing a question for the AI: how would he explain this gently in Twi, without sounding cold or dramatic?
He stops with his thumb over the send button.
The family visit is still coming. The conversation he wants help with is personal now. If he sends it without understanding where it goes, he could share a private family matter more widely than he intended. That is the bad ending on the table, and it matters before a single carefully chosen Twi phrase appears.
Consent should be a decision, not a hidden setting
Cloud processing can be useful for an AI conversation, especially when you are speaking naturally, switching between Twi and Ghanaian English, or asking for help with a complicated thought. The important question is whether the app lets you make that choice clearly.
A good consent control gives you meaningful options. You should be able to say never, ask each time, or allow cloud use for the current session. Each option should be understandable without reading a privacy policy like it is a contract dispute.
“Ask each time” can suit a conversation that starts light and turns serious. Kwesi might allow cloud use while practising greetings, then pause when the subject becomes his uncle’s health. A session-based choice can fit when you know what kind of conversation you are having and want the setting to end when the session ends.
Watch for vague labels such as “improve your experience” or “personalise responses.” They do not tell you what leaves your device, when, or why. Clear language should tell you what the setting changes before you speak or type.
For a closer look at how a small voice action can carry bigger privacy consequences, read The Three Jobs Behind Hold, and What Cloud Consent Can Cost You.
Local history needs its own controls
Cloud consent and chat history answer different questions.
Consent asks: may this conversation be sent to the cloud?
History asks: what remains on this phone after the conversation ends?
An app can handle one well and still leave you with uncertainty about the other. If you practise a private phrase, ask about a family disagreement, or dictate something while walking home, you should be able to see whether that exchange stays in local history and how to remove it.
Look for a history setting you can turn off directly. Check what happens when you do. “History off” should mean more than hiding a list from the screen. In Nkomo, turning off on-device history purges it immediately. That gives the setting a real outcome: the old conversation is removed from the device rather than simply becoming harder to find.
Also check whether you can export your data and delete your account in one tap. Those controls matter when you want a copy for yourself, or when you want a clean exit without guessing which support address to contact.
Privacy controls should still work in a real conversation
Kwesi chooses to keep the sensitive exchange out of history and pauses before granting cloud consent. Then he rewrites his question more generally: he wants a respectful Twi phrase for asking an elder how they are feeling, without naming anyone or explaining the family situation.
The reply helps him practise. He says it aloud twice, changes one word, and hears the difference. Later, when he closes the conversation, he knows the practice does not sit in his phone’s history.
That is the practical test. A privacy control should fit the moment you are actually in, including the moment when an ordinary language exercise becomes personal.
A quick check before you share something personal
Before sending a voice note or typed message to an AI companion, take ten seconds to check four things:
- Can you choose cloud consent at the point where you are about to speak or send?
- Does the app explain what “never,” “ask each time,” and “this session” mean?
- Can you turn off local history, and does the app say whether existing history is removed?
- Can you export your data or delete your account without a long support process?
Kwesi still wants to arrive in Kumasi with better Twi. He also wants the private part of the practice to stay private. Both are reasonable expectations from the same conversation.
Comments
No comments yet.