What Is AI Actually Useful For in a Kenyan NGO?

TL;DR:

• AI helps most with the writing, reading and translating work in an NGO: minutes, job adverts, policies, board papers, annual reports and community notices.
• Sort every task into one of three groups: use it now, use it with controls, or keep it human. Case work, safeguarding and decisions about who gets support stay with people.
• Your staff are probably using it already. In a January 2026 survey of humanitarian workers, 65% of respondents in Kenya used AI every day. Only 13% of respondents in local organisations said they had an AI policy.
• Free plans are enough to start learning, but they come with limits and privacy trade-offs. Read the settings before staff use them for work.
• A chatbot that talks to the people you serve is a much bigger decision than using AI in the office. It needs human backup and proper testing.

Who this guide is for

This guide is for directors, programme managers and administrators in Kenyan NGOs, community-based organisations and other public benefit organisations. You do not need any technical background.

It gives you a map of a whole NGO, department by department. Four companion guides go deeper on single topics:

• Writing donor reports faster with AI
• What funders allow in AI-assisted grant proposals
• AI and safeguarding
• Protecting beneficiary and donor data

What is AI best at inside an NGO?

AI is best at first drafts, summaries and changing the form of text you already have. Think of it as help with language work. It can turn rough notes into minutes, shorten a long report, or rewrite a technical paragraph for a community meeting.

The same pattern shows up in survey data. The Humanitarian Leadership Academy and Data Friendly Space ran a pulse survey in January 2026, with 1,729 responses from more than 120 countries. The most common uses were writing and editing, research and planning, pulling information together, and translation. Fewer than 10% of respondents used AI for coding, machine learning, forecasting or satellite imagery.

So most of the value today sits in ordinary office work. You do not need a data scientist to get it.

AI is weak in three places, and an NGO feels all three:

• Facts. It can state a wrong date, name or number with full confidence.
• Local knowledge. It does not know what happened at your site last quarter, or why a chief changed his mind.
• Judgement. It cannot weigh the risk to a child, a survivor or a community.

Keep those three weak points in mind as you read the table further down.

Are NGO staff in Kenya already using AI?

Very likely, yes. In the January 2026 survey, 95.3% of respondents had used AI tools for work, and 75% used them daily or weekly. Kenya stood out: 65% of respondents there used AI daily, among the highest country figures in the report.

The rules have not caught up:

One caution about these numbers. The respondents chose to take part and reported on themselves, so people interested in AI are likely to be over-represented. The briefing note does not say how many respondents were in Kenya. The survey covers humanitarian workers, so it is a guide to the wider Kenyan NGO sector and no more than that.

Even with that caution, the practical point holds. If you have no rule, your staff are deciding for themselves what goes into these tools. A short written rule protects them as well as the people you serve. Our guide to writing an AI usage policy shows what to include.

Why does this matter more for Kenyan NGOs now?

Two pressures have grown at the same time: less money from some donors, and more formal compliance.

In February 2025, Business Daily reported that the US government had terminated five contracts and grants in Kenya worth a combined Sh32.5 billion. Where a team has lost posts, the reports still have to be written.

Compliance has also changed. The Public Benefit Organizations Act, 2013 took effect on 14 May 2024 and replaced the NGO Coordination Act. We cover what that means for your paperwork in its own section below.

AI can reduce the hours spent on routine drafting, though it cannot replace lost funding or lost colleagues.

The task map: what to use now, what needs controls, what stays human

Use three groups:

• Use now: the input is public, invented or general. A mistake is cheap and easy to catch.
• Use with controls: the input is real organisational information with no personal details. Use an approved tool, and a named person checks the output.
• Keep human: the work involves identifiable people, safeguarding, or a decision that affects someone's life.

Kenya's Data Protection Act sits behind the "keep human" group. It treats health status, ethnic social origin, family details and more as "sensitive personal data". Section 35 gives people a right not to be subject to a decision based solely on automated processing that significantly affects them, with exceptions for contracts, laws with safeguards, and consent. NGO files are full of the first, and beneficiary selection can easily become the second.

Our data privacy guide explains what the Act requires of NGOs, and why removing a name is often not enough.

The compliance paperwork most AI articles miss

Annual reporting is a real, recurring writing job for Kenyan NGOs, and AI can help with the drafting.

The Public Benefits Organizations Regulations, 2026 (Legal Notice No. 43, gazetted on 18 March 2026) are specific. Regulation 26 says each organisation must keep up-to-date records of its audited accounts, annual financial statements, a detailed inventory of assets, and an annual report on its activities and programmes. Regulation 27 says it must submit an annual report to the regulator. The form in the Regulations (Form 13) asks for an audit report and filing within six months of the end of the financial year.

The position on moving across from the old law is less simple. The Star reported in June 2026 that existing NGOs have until May 2027 to complete the move, and that the extra time does not postpone compliance duties. Regulation 43 asks existing NGOs to give the regulator documents so that it can issue a certificate. But in June 2025 the High Court struck down the rule that made registered NGOs apply again, and ordered them to be moved across automatically. The Star reported in November 2025 that this part of the ruling still stood while an appeal went on. We could not confirm anything later, so ask the regulator or your lawyer what your organisation must do.

This has been a weak spot for the sector. The Star reported that, according to the 2022/23 sector report, only 2,828 of 12,000 registered NGOs filed annual reports. In 2026, the regulator extended the filing deadline for 2025 annual reports to 9 July, for organisations with a December year-end, after problems paying fees on eCitizen.

Here is a sensible division of labour:

• AI can turn your four approved quarterly reports into a single draft of activities for the year, in plain language.
• AI can check your draft against a list of required sections that you provide.
• People must confirm the legal requirement with the regulator or your lawyer, because an AI tool can be wrong or out of date on Kenyan regulations.
• Your auditor and finance lead own every financial figure.

Five prompts you can copy

We teach a simple method called the four Cs:

• Clarity: say exactly what you want.
• Context: give the background the tool cannot guess.
• Constraints: set the length, tone, format and what to leave out.
• Customisation: shape it for the reader and where it will be used.

Replace the square brackets with your own details. Keep names, ID numbers and case details out.

1. Minutes from rough notes

Turn my rough notes below into minutes for an NGO management meeting. Context: the minutes go to the Executive Director and are kept on file. Constraints: record decisions and action points only, do not add any decision or detail that is not in my notes, and write [CHECK] wherever my notes are unclear. Customisation: end with a table showing action, responsible role and deadline. Notes: [paste notes with staff performance matters and personal details removed]

The [CHECK] marker shows you the gaps. Without it, tools tend to fill them in.

2. A consultant's terms of reference

Draft terms of reference for a short consultancy to [task] for a Kenyan NGO working in [sector]. Context: the budget is limited and the consultant will work with a team of [number] for [number] days. Constraints: maximum two pages, with headings for background, purpose, tasks, deliverables, timeline, qualifications and how to apply. Do not invent facts about our organisation. Put [ADD] where you need information from me. Customisation: readers are independent consultants in Kenya, so be plain and specific about deliverables.

3. A year of activities from quarterly reports

Below are four approved quarterly summaries with all personal details removed. Combine them into one activities section for our annual report. Constraints: use only what is in the text, keep every number exactly as written, do not explain why results changed unless the text says so, and maximum 500 words. List any numbers that appear to conflict between quarters at the end, and do not fix them yourself. Customisation: readers are a regulator and our board, so use plain, formal English. Text: [paste]

Asking it to list conflicts, and not to fix them, keeps the decision with you.

4. A community notice in English and Kiswahili

Draft a notice for a community meeting about [public topic] at [public venue] on [date and time]. Context: it will be read aloud and shared on WhatsApp. Constraints: under 100 words, respectful and simple, with one contact role. Customisation: give an English version and a Kiswahili version that sounds natural for [area], and avoid word-for-word translation.

A fluent speaker from the community should read the Kiswahili before it goes out.

5. Finding gaps in a draft policy

I will paste our draft [travel, procurement or leave] policy. Compare it with the checklist below and list what is missing or unclear. Constraints: do not rewrite the policy, do not give legal advice, and quote the sentence you are commenting on. Checklist: [paste your own list]. Policy: [paste]

AI that faces your community is a different decision

Everything above is office work, where a staff member reads the output before anyone relies on it. A chatbot or SMS service that answers the public removes that check, so the risks rise sharply.

Kenya has two useful examples to learn from.

Jacaranda Health and PROMPTS. This non-profit, based in Nairobi, runs an SMS service for pregnant women and new mothers. Canada's International Development Research Centre reported in May 2025 that 650,000 mothers were registered on the platform in 2024, and that Jacaranda had built UlizaLlama, a large language model for health in Kiswahili. The same report gives figures any NGO leader should read closely. Jacaranda estimated at the time that the service still made errors in 30% of cases. Danger signs appeared in about 7% of messages and led to a call from a staff member. A team that included nurses and nutritionists stepped in behind the automated replies.

Digital Green and Farmer.Chat. Digital Green built an AI advisory chatbot for smallholder farmers. A randomised trial planned across 600 villages in Nakuru County was registered in October 2025, with an end date listed as 31 March 2027. The registry still showed it as ongoing on 10 November 2026, so the impact results for Kenya are not yet known.

Together, the two examples show what such a service takes: people who answer when the AI cannot, a way to catch danger signs, and serious measurement. Before you start one, ask:

1. What happens when it gives a wrong answer to someone in danger?
2. Who responds, at what hours, and who pays for them?
3. Who is left out: people without phones, or who speak neither English nor Kiswahili?
4. How will we measure whether it helps, and who will check that?
5. Have we done a data protection impact assessment? Section 31 of the Data Protection Act requires one where processing is likely to result in high risk.

If you cannot answer these yet, stay with office uses for now.

How to decide about any new task

Five questions will sort almost any task:

1. What goes in? Public or invented material is the easiest case. Anything about an identifiable person needs far more care.
2. What is the cost of a mistake? A clumsy newsletter costs little, while a wrong figure in an audit can cost a lot.
3. Can someone check it quickly? If checking takes longer than doing the work, skip AI for that task.
4. Who signs it? Name the person. If nobody will own it, it should not go out.
5. Would the people affected be comfortable? Picture the community member, the donor and your board reading how you did it.

What goes wrong

What does it cost? (as of November 2026)

You can begin without paying for software. Plans change often, so confirm them on each company's page before deciding.

Two details deserve attention.

Microsoft says prompts and responses in Copilot Chat are covered by its enterprise data protection commitments and are not used to train its foundation models. That is a stronger starting point for office work than a personal free account. The specific controls depend on your subscription.

Google's Kenyan eligibility page still names the NGO Coordination Board in its list of registration types, and verification runs through a partner called Goodstack. If you have already migrated under the new Act, ask them how they treat your registration.

Budget staff time as well: people need hours to learn the tools and to check what they produce.

How to begin: a four-week start

The HLA and NetHope quick-start guide, published in March 2026, gives advice that suits small organisations: set clear guardrails, recommend approved tools, test and share what works, and start small before measuring results. Here is one way to turn that into a month.

Week 1: set the rules. Bring the whole team together for an hour. Agree what must never go into a tool and which tools are approved. Everyone needs the rules, including people who will not use AI much. List the ten writing tasks that take the most time.

Week 2: pick one task. Choose something frequent and low risk, such as minutes. Note how long it takes now, including review.

Week 3: test it. A small group tries the task with a shared prompt on non-sensitive material. They keep a simple note of every error the AI made.

Week 4: review and share. Compare the time, including checking. Ask the reviewer whether quality improved. Then show the whole team what worked, what failed and how the prompt changed.

For the full training sequence, see how to train NGO staff on AI safely.

Where AI will not help

Be honest with yourself about these.

• If field data is missing or unreliable, AI will write a confident report on weak evidence.
• If a donor template is unclear, ask the donor. AI can only guess.
• If your team has no agreement on what the programme achieved, a tool cannot settle it.
• Trust with a community, a difficult conversation with a chief, and sitting with a survivor all remain human work.

Questions people ask

Is it legal for an NGO in Kenya to use ChatGPT or similar tools?

We have found no Kenyan law that bans it. The Data Protection Act still applies to any personal data you enter, and your donor agreements may add conditions. Use approved tools and keep personal data out.

Do we need to buy anything to start?

No. Free plans are enough for learning on public or invented material. For real organisational work, look at tools with stronger data terms, such as Copilot Chat through a work account. If you are eligible for Google Workspace for Nonprofits, first check that your edition is covered by Google's Workspace privacy commitments.

Can AI help us decide who receives support?

It can help organise general information, but it should not decide. Section 35 of the Data Protection Act gives people a right, with limited exceptions, not to be subject to a decision based solely on automated processing that significantly affects them. A person should make and own these decisions, and people should be able to ask for a review.

Should we train everyone, or only a few people?

Both, in order. Everyone should learn the rules and a shared method. A small group can then test one task first and share what they learn with the rest.

Is AI good enough in Kiswahili for community messages?

It often gives a usable first draft. Quality varies by tool and topic, so a fluent speaker should always review the text. Take even more care with Sheng and mother tongues.

Do we have to tell donors that we used AI?

That depends on the donor and the grant agreement. As of November 2026, we found few donors with published rules about reports. Our donor report guide sets out what has been published and when to disclose.

What should we never use AI for?

Do not put case details, safeguarding reports or identifiable beneficiary information into a public AI tool. Do not let AI make decisions about people. Do not create fake pictures of the communities you serve.

Sources

All links last checked 10 November 2026.

• Humanitarian Leadership Academy and Data Friendly Space, Humanitarian AI January 2026 pulse survey briefing note (March 2026)
• Humanitarian Leadership Academy and NetHope, Develop your organisational AI readiness: a quick start guide (March 2026)
• The Data Protection Act, No. 24 of 2019 (ODPC copy of the Gazette version)
• The Data Protection Act, No. 24 of 2019 (ICNL text copy, used to confirm section numbers)
• The Public Benefits Organizations Regulations, 2026, Legal Notice No. 43 (ICNL copy of the Gazette version)
• Business Daily, Sh33bn Kenya contracts, grants axed in USAID purge (17 February 2025)
• The Star, report on the Public Benefit Organisations Act taking effect (13 May 2024)
• The Star, Kenya lets NGOs run businesses in major overhaul of sector rules (23 June 2026)
• The Star, report on the 2022/23 NGO sector report (18 June 2024)
• The Star, Court strikes out key provisions in the PBO Act (21 June 2025)
• The Star, Court freezes PBO board's decision-making powers pending appeal (15 November 2025)
• Kenyans.co.ke, PBORA extends 2025 annual reports filing deadline for PBOs to July 9 (1 July 2026)
• International Development Research Centre, Technology in support of maternity (20 May 2025)
• AEA RCT Registry, Improving smallholder agriculture with generative AI: The impact of Farmer.Chat in Kenya
• OpenAI, ChatGPT pricing
• Google, Gemini subscriptions
• Google, Gemini Apps Privacy Hub
• Anthropic, Claude pricing
• Anthropic, consumer plans: Is my data used for model training?
• Microsoft Support, Get started with Microsoft Copilot
• Microsoft Learn, Enterprise data protection in Microsoft Copilot and Microsoft Copilot Chat
• Google, Generative AI in Google Workspace Privacy Hub
• Google for Nonprofits, Eligibility guidelines: Kenya
• Google for Nonprofits, Google AI in Workspace with nonprofit discounts

Where FuKazee fits in

FuKazee trains NGO teams to use AI on the work in front of them. We assume nobody in the room knows AI yet, we avoid jargon, and we show the free version of every tool. We do not promote one tool. We also say plainly that nobody becomes an expert in a few days.

Since 2022 we have trained more than 3,000 people. In September 2026 we trained 13 staff at Maisha Girls Safe House, a Nairobi NGO, on AI for communication and reporting. You can read their recommendation letter and how the training ran.

Our exercises use invented examples, so no real beneficiary information enters a tool during training. See what we cover on our page for AI training for NGOs in Kenya.

To talk through what would help your team, email lead@fukazee.com.