Adwoa’s family message. Its private context cannot be taken back.

Free access removes one barrier to trying AI in Ghana. Clear, usable answers about where a conversation goes, how long it stays there, and how to remove it are still necessary before people can speak freely.

At 8:40 on a humid evening in Kumasi, Adwoa is standing by her kitchen window with her phone in one hand and a half-folded school-uniform shirt in the other. Her younger brother has asked her to help explain a family money problem in Twi and English before he sends a message to their aunt abroad.

She starts to speak, then stops.

The message includes names, a mobile-money issue, and a detail her brother has not told the wider family. If she puts the whole thing into an AI conversation, will it stay on her phone? Will it reach the cloud? Can she remove it afterwards? The family message needs to go out that night, but the private context cannot be taken back once she has shared it.

That hesitation has nothing to do with the price of data.

Free access opens the door, but trust decides what people say

Google and Cassava Technologies are working to bring Gemini to millions across Africa. Free access can make a meaningful difference for someone who has avoided AI because every prompt felt like another cost to weigh.

That matters. A student can test an explanation before an assignment. A small business owner can draft product copy. A parent can ask for help making a message clearer. More people get a chance to find out whether AI is useful in their own daily lives.

But the first low-cost prompt is rarely the whole story. The conversations people most want help with are often personal: a voice note for a relative, a job application, a health question, a dispute, a financial worry, a message that moves between Twi and English because that is how the thought arrived.

“Free” answers the question, “Can I afford to try this?”

It does not answer, “What happens after I press send?”

A privacy policy is too far away from the moment of hesitation

Adwoa does not need a long document while her brother waits for a reply. She needs an answer at the point of speaking.

Can this conversation use the cloud? Does the app ask every time, or has she already agreed without noticing? Is her history saved on the device? If she turns history off, does the existing record disappear? Can she take her data with her, or delete her account without a difficult process?

These are practical questions, especially for voice. Speaking feels closer to thinking than typing. People often say more, include more context, and use names or family details they would edit out of a formal message.

A service can be free to access while still asking people to make a decision with incomplete information. That uncertainty encourages a safer but less useful behaviour: asking only generic questions, leaving out the detail that would make the answer helpful, or avoiding the tool entirely.

The question is not whether anyone should use cloud AI. Cloud use can be a conscious choice. The standard should be clear consent, in plain language, before a personal conversation goes anywhere.

Control has to work in the language people actually speak

For a Ghanaian conversation, privacy and language meet in the same moment. A person should be able to say, “Meka Twi ne English afra,” without first translating their life into a format an assistant can process.

That includes the small phrases that carry tone, caution, and family meaning. A mixed sentence may contain the exact word that tells a reader whether a message sounds respectful, worried, firm, or warm. What happens when a Sunday call switches between Twi and English? explores what can be lost when an assistant drops that thread.

Nkomo is built around that real conversation: natural Twi and Ghanaian English, by text or voice. It also makes the privacy choice visible. You can choose never, ask each time, or this session for cloud consent. You can turn off on-device history, which purges it immediately. You can export your data or delete your account in one tap.

Those controls do not ask people to trust a vague promise. They give them a decision they can make before they share the next detail.

The useful test is a personal one

When Adwoa returns to the draft message, the problem has changed shape. She can decide what to share, when to allow cloud use, and whether to keep the conversation on her device. Then she records the message in the mix of Twi and English her aunt will understand.

The shirt is still on the counter. Her brother is still waiting. But she is no longer choosing between help and silence.

As free AI access reaches more people, ask one practical question before using any assistant for a sensitive task: can you see, choose, and later reverse what happens to your words?