How Can a Small NGO Use AI for Communications?

TL;DR:

• A small NGO can keep its communications going without a comms hire if one named person owns the work and AI does the first drafts.
• Start with one real update a week. Turn it into a social post, a newsletter item and a WhatsApp message, then check every fact.
• Give the tool a short voice sheet and real details. Vague requests produce the bland NGO language nobody reads.
• Stories and photos need the informed consent of the people in them. Think hard before using an AI image of a person, and label any you do use.
• Have a fluent speaker check Kiswahili, and give every mailing list an easy way to opt out.

Who this guide is for

This guide is for programme officers, admin leads and directors at small NGOs in Kenya and East Africa who handle communications on top of their main job. You may post when you find a spare hour. You may also feel that your organisation does good work and that too few people hear about it.

You need no design or technical skills. All the prompts below work on the free plans of common AI tools.

This guide covers everyday public communication: social media, newsletters, WhatsApp, event notices and thank-you messages. For formal reporting, read how to write donor reports faster with AI.

Can AI replace a communications hire?

AI can take over much of the drafting, which is the part that eats time. It cannot visit the field, build trust with a journalist or decide what your organisation should say on a hard day.

Many nonprofits start here. A Google survey of 6,461 nonprofits, shared in July 2026, gives 80% for organisations piloting AI in marketing and content, and describes this as the main starting point. Those respondents use Google's nonprofit programme and work around the world, so treat the figure as a guide for Kenya.

One thing AI will not fix is ownership. If communications belong to nobody, they will still slip when programme work gets busy. Name one person, give them protected hours each week and name a deputy for when they are in the field.

Where should a small NGO focus its communications?

Pick two channels and an email list, and do them well. Choose by asking who needs to hear from you, and where they already are.

DataReportal's Digital 2026 report for Kenya gives a rough picture of platform reach in late 2025. These are advertising reach estimates, which can differ from the number of active users.

Big numbers matter less than your own audience. Our suggestions:

Before you add a channel, ask who will post on it in the busiest month of your year. If nobody can, leave it out.

Which communications tasks suit AI?

AI suits tasks where a person supplies the facts and can check the result quickly.

Before anyone starts, agree your organisation's rules on AI. Our guide to an AI usage policy for Kenyan organisations can help you write them, and our data privacy guide for NGOs explains what must stay out of public tools.

How do you make AI sound like your organisation?

Write a one-page voice sheet and paste it in at the start of each drafting session. You only need to make it once. Update it when your work changes.

We teach a simple method called the four Cs: Clarity (what you want), Context (the background), Constraints (limits on length and tone) and Customisation (who it is for and where it will appear). All the prompts in this guide follow it.

Prompt 1: Build your voice sheet

I will paste three pieces of writing from our NGO that we are happy with. Study them and write a one-page voice sheet. Context: we work on [sector] in [county or area] and speak to [audiences]. Include: our tone in three plain phrases, typical sentence length, words we use for the people we work with, words we should avoid, and how we describe our results. Constraints: base it only on the samples and do not invent facts about us. Customisation: a new staff member with no communications training should be able to follow it. Samples: [paste public writing only]

Then add these by hand:

• Words you never use. Your team's list might include "empower", "beneficiaries" (if your community prefers another word), "transform" or "vulnerable".
• How you name people and places, following what the community prefers.
• Your rule on claims. For example: "Only say what we have counted or seen."

Our sister guide on donor reports has more on keeping your voice in formal writing.

How do you get material without chasing colleagues?

Ask every programme lead the same five questions each week, and accept short answers by WhatsApp or email.

1. What happened this week?
2. How many took part, and how do you know?
3. What surprised you, or what went wrong?
4. What is coming next?
5. Is there a photo with consent on file?

Remind colleagues to leave out names, phone numbers and case details. A voice note can feel harmless, but AI transcription tools raise their own risks. Our safeguarding guide explains when to avoid them.

How do you turn one update into several pieces of content?

Give the tool the raw facts once, and ask for each format you need. This is the habit that saves the most effort, because the thinking only happens once.

A worked example

Everything in this example is invented for practice.

Raw update: "In October we repaired two shallow wells in [ward]. Fourteen water committee members were trained on basic upkeep. Pump parts took three weeks to arrive, which delayed the second repair. The committee now meets monthly."

Prompt 2: One update, three formats

Below is our voice sheet and a programme update. Write three versions. (1) A Facebook post under 70 words that opens with what changed for the community. (2) A WhatsApp message to supporters under 50 words with one clear ask: [ask]. (3) A newsletter item of about 100 words that mentions the delay honestly. Constraints: use only the facts in the update. Do not add numbers, quotes, names or outcomes. Mark anything unclear with [CHECK]. Avoid words on our banned list. Voice sheet: [paste] Update: [paste]

A possible Facebook draft, after a person has edited it:

Two wells in [ward] are working again. Fourteen members of the local water committee have learned how to keep them running, and they now meet every month. The repairs took longer than we hoped, because pump parts took three weeks to arrive. Thank you to everyone who stayed with us.

Look at what the editor must check. The update does not say the wells serve a particular number of people, so the post should not say so. A tool will sometimes add that kind of detail, which is why the prompt forbids it and why a person reads the result.

More prompts to copy

Use these on any common AI tool. Replace the brackets and do not include personal details.

Prompt 3: A personal thank-you note

Draft a short thank-you message from our director to a supporter who [what they did]. Context: [what it made possible, using facts we have confirmed]. Constraints: under 90 words, warm and simple, no exaggeration, no claims about lives changed unless I give them. Leave a [PERSONAL DETAIL] marker where I will add something only we would know.

Prompt 4: An event invitation

Write an invitation to [event] for [audience]. Give the date, time, place, what will happen, why it matters and how to confirm. Constraints: under 100 words, friendly, with one clear action at the end. Customisation: it will be read on a phone in WhatsApp, so use short lines. Details: [paste]

Prompt 5: Make long text simple

Rewrite the text below so a busy reader with no background in development work can understand it at a glance. Constraints: keep every fact and number exactly as written. Remove jargon such as "stakeholders", "capacity building" and "leveraging". List any sentence where you had to guess the meaning. Text: [paste public text]

Prompt 6: Reply to a routine public comment

Draft a polite public reply to this comment on our page. Context: [the correct information]. Constraints: under 40 words, no promises and no personal information. If the comment is angry, suggests harm or raises a safeguarding concern, do not draft a reply; tell me to pass it to [role]. Comment: [paste, with the name removed]

Prompt 7: Describe an image for accessibility

Write alt text under 125 characters for a public photo on our website, based on my description. Describe only what is visible. Do not guess anyone's age, ethnicity, health or status. Description: [your own description]

A monthly routine that needs little time

A small system that you actually keep beats an ambitious calendar. The one below is a suggestion. Adjust it to your team.

You can plan and schedule Facebook and Instagram posts in Meta Business Suite. Scheduling means a post can still go out when the owner is travelling.

Time this routine for a month, including the checking. If it is taking far longer than the table suggests, cut a channel before you add a tool.

How should you handle consent and people's stories?

Get informed consent before you use anyone's image or story, and keep proof of it. This matters more than any tool choice.

Bond, the UK network for international development organisations, updated its guidelines on ethical content, Putting the people in the pictures first, in November 2024. They are written around UK data law, and they are a useful standard for any NGO. Key points:

• Consent is a conversation as well as a form. The person should understand how and where their story will appear.
• Tell people how long you will keep their image and information, and how they can withdraw consent.
• Keep evidence, such as a signed form or a recording of spoken consent.
• Check they know the story may be seen by people they know, and may stay online for a long time.

Kenyan law adds its own duties. Under section 33 of the Data Protection Act, processing a child's data generally needs parent or guardian consent and must protect the child's best interests. Our data privacy guide explains how this applies to photos.

A practical rule for AI: do not paste a person's story, name or photo into a public AI tool. If you want help with a story, write a version with no identifying details first, and have the person who collected the story review the draft. Editing can change meaning, and the person whose story it is should not be misrepresented by a tool that never met them.

Should an NGO use AI-generated images of people?

Our advice is to avoid AI images of the people you serve, and to use them only in rare, clearly labelled cases.

Bond's guidelines say AI images are no way to solve problems of representation or to avoid consent issues, since the tools are built on existing photography and carry documented biases. They recommend using AI images for only two reasons: budget limits for smaller NGOs, or sensitive contexts where real images could put people at risk. When you use one, they advise saying so clearly in the caption, for example: "this image is AI-generated".

The sector has seen why this matters:

• Dr Arsenii Alenichev of the Institute of Tropical Medicine in Antwerp has found large collections of AI-made images of poor and vulnerable people for sale, as the International Broadcasting Trust reported in November 2025.
• In the same piece, the Trust's deputy executive director urged charities to stop buying AI stock images of this kind.
• Bond published a May 2026 piece on a study of 171 AI-generated images used by 17 major organisations. Of more than 400 public comments, only 80 engaged with the humanitarian issue. Most of the rest questioned the image.

So even a well-meant AI image can pull attention away from your cause and towards the picture.

Meta's transparency page, last updated in February 2025, says Facebook and Instagram may show an "AI Info" label on content when Meta detects AI signals or when the person posting discloses it.

Can AI write in Kiswahili and local languages?

AI can draft Kiswahili, and a fluent speaker should always review it before it goes out. Other local languages need more caution.

The AfroBench study tested large language models on African languages. Kiswahili scored highest among the 14 African languages in one of its test sets, and the researchers linked performance to how much text exists online. African languages still trailed English on average, by more than 25 points for GPT-4o. Models change quickly, so test the current tool yourself.

If your community language has little text online, expect weaker results. Ask a speaker from the community to write or check the message.

Prompt 8: A natural Kiswahili version

Rewrite this English message in natural Kiswahili for [audience] in [area]. Context: it will be sent on WhatsApp. Constraints: do not translate word for word. Keep every date, number and place exactly as written. List any word where you were unsure of the best choice, so that a fluent speaker can check it. Message: [paste]

What are the rules for mailing lists and WhatsApp?

Give people a free, simple way to opt out of any promotional message. Treat your supporter list as personal data, because it is.

Kenya's Data Protection (General) Regulations, 2021 set rules for direct marketing:

• Regulation 14 defines commercial use broadly. It covers using personal data to advance commercial or economic interests, including persuading someone to join, subscribe to or provide something as well as to buy.
• Regulation 15 says people must be able to ask you to stop without paying for it, and that the sender must not hide who they are.
• Regulations 16 and 17 say the opt-out must be clear, take little effort and be mentioned in your messages.

We found no ruling on whether an NGO's fundraising newsletter counts as direct marketing. The safe course is to follow these rules anyway, and to ask your legal adviser if you are unsure. This guide is general information and is no substitute for legal advice.

A simple closing line for a newsletter:

You are getting this because you signed up at [event or page]. To stop these emails, reply "STOP" or click here.

On WhatsApp, a broadcast list only reaches people who have saved your number, according to WhatsApp's help pages. Ask people to save your number when they join, and only add those who agreed. Never paste your contact list into a public AI tool to "tidy" it.

Should you tell people you used AI?

For ordinary drafting that a person has checked, you can decide for yourself. For anything that could mislead someone, tell them.

The Fundraising Regulator, which covers England, Wales and Northern Ireland, has guidance on this, last updated in December 2025, and it is a sensible model even outside those countries. It asks charities to have a person check AI output for accuracy, fairness and legality. It says labelling is not always needed, and that the risk of misleading donors should decide how open you are.

Our suggestion:

• Say so whenever an image, video or voice is AI-made.
• Say so if a donor or funder asks, or if their rules require it.
• Write one sentence in your AI policy that describes how you use AI, so that any staff member can answer the question the same way.

What goes wrong, and how to avoid it

Check before you publish

• Every fact, number, date and name matches the source.
• Nothing was added that the update did not say.
• Nobody can be identified without their consent, including through location or context.
• We have consent on file for every person shown.
• Partners and funders are named and credited correctly.
• Any AI-made image is labelled.
• A fluent speaker has read any Kiswahili or local language text.
• The message has one clear action and works on a phone.
• Mailing messages include a simple way to opt out.
• A second person has read anything sensitive.

Questions people ask

Can one person run an NGO's communications with AI?

Yes, for a small organisation, if they have protected time and a deputy. AI shortens the drafting, while collecting real stories, getting consent and checking the facts still take time.

Which AI tool is best for NGO social media?

There is no single best tool, in our view. Start with one your team already has access to, such as ChatGPT, Claude, Gemini or Copilot, and test them on the same update. The quality of your details and your checking will matter more.

Can we paste beneficiary stories into ChatGPT to polish them?

Do not paste anything that could identify a person into a public AI tool. Remove the details first, and check that the story cannot be traced through its location or circumstances. Our safeguarding guide has a clear rule you can adapt.

Can we use AI images in fundraising?

We advise against AI images of people you serve. Bond's guidelines recommend them only for budget reasons or in sensitive contexts where real images could cause harm, and always with a clear label.

Do we need to label AI-written posts?

As of November 2026, we found no Kenyan rule that requires a label on checked, AI-assisted text. The Fundraising Regulator for England, Wales and Northern Ireland links the need to the risk of misleading donors. Always say so for AI-made images, video and voices.

How often should a small NGO post?

Post only as often as you can keep up with accuracy. One good post a week and a monthly newsletter is a solid starting point. Regular and accurate work matters more than volume.

Will AI make our communications less authentic?

It can, if you let it write without your real details. Give it true updates and a voice sheet, write the most personal lines yourself, and cut any sentence you would never say aloud.

Sources

All sources were last checked on 10 November 2026.

• Google for Nonprofits, "Putting AI to Work for Your Mission" survey report
• DataReportal, Digital 2026: Kenya (8 November 2025)
• Bond, Putting the people in the pictures first (2024 version, PDF)
• Bond, Updating our ethical storytelling guidelines (25 November 2024)
• Bond, Beyond the label: values live in the image edit (6 May 2026)
• International Broadcasting Trust, Why AI-generated "poverty porn" fake images need to be stopped (19 November 2025)
• Fundraising Regulator, New guidance to support charities using AI in fundraising (8 December 2025)
• Meta, Labeling AI content
• Meta, Meta Business Suite
• WhatsApp Help Center, How to use broadcast lists on the WhatsApp Business app
• Ojo and others, AfroBench: How Good are Large Language Models on African Languages?
• The Data Protection Act, No. 24 of 2019 (ODPC)
• The Data Protection (General) Regulations, 2021 (ODPC)

Where FuKazee fits in

FuKazee trains teams to use AI on the work in front of them, and that includes the writing that small NGOs do every week. We assume nobody in the room knows AI yet, we avoid jargon, and we say plainly that nobody becomes an expert in a few days. We show the free version of each feature, teach the four Cs and never push one tool.

Practice uses invented cases and sample documents, so no beneficiary data enters the training. For Maisha Girls Safe House, 13 staff trained together, including communications, programme and leadership staff, and you can read their recommendation. Since 2022 we have trained more than 3,000 people.

Training will not give your organisation something to say, or the time to say it. If those are missing, start there. If your team has the stories and needs a simple system, read how to train NGO staff on AI and see our page on AI training for NGOs in Kenya.

To talk through training for your team, email lead@fukazee.com