- Role
- consultant, conversational design & safeguarding
- Client
- UNICEF, for Laaha, its platform for women and girls in humanitarian settings
- When
- July 2023 – April 2024
- Scope
- 74 content modules, one chatbot, English then Spanish, handed over for more countries
UNICEF asked for a chatbot to guide women and girls in humanitarian settings through 74 modules of content, often on a shared phone, in a second language, about things they could not ask anyone around them. My job was what Anya says, in what order, in what tone, and where she stops.
The voice
- Tone
- Friendly, non-judgmental, supportive. A warm voice that does not lecture.
- Personality
- A big sister. Conversational, positive, on the user’s side. Not a counsellor, not a form.
- Language
- Clear and simple. No trendy slang: it will not survive translation and it will not land in a refugee camp.
- Privacy and the exit
- The user can leave at any moment, take a break, come back. Anya says so before anything else.
- Emergencies
- Anya can give a directory of services and helplines. She cannot intervene, she has no call centre, and she says that plainly.
Three safeguarding decisions
01
The exit comes first
A girl on a shared phone needs to know how to leave before she knows what is here. So the exit is in the opening line, before the offer.
“I can help you explore cool videos, articles, and podcasts. If you ever feel unsure or unsafe on Laaha, you can easily exit the website. Ready to explore with me?”
02
The emergency button promises nothing it cannot do
The hardest line in the product. It has to stay warm, send her to people who can act, and not pretend the website is one of them. The first version was mine; UNICEF’s content team plain-languaged it with me, which is what the review pathway is for.
“You’re not alone. Here is a list of services available for your area. They can help with problems you face. Asking for help takes courage. You deserve it. We’re here for you!”
03
A check-in after heavy content, not a next button
After a module on violence or consent the flow does not push forward. A black node asks how she is and offers the break. A review question hung over every branch: what happens when a user reaches the end of one? In the build, she is asked how she is, offered a break, and offered a restart. She is not dropped.
“How do you feel about this content? Remember, it’s okay to take breaks if needed.”
One conversation stayed with me. A woman read, and heard, Anya explain financial control and the right to privacy, and then stopped. Not because the content was wrong. Because it was right, and knowing it did not change her room.
I still do not have an answer to the question that left: if the information cannot take someone out of her circumstances, do we keep delivering it?