The team at Raseef22, a Lebanese independent media organisation, had two goals in mind when they decided to build ‘Ask Aunty’ - an AI-based chatbot. One, was to build a tool where its readers would feel comfortable asking sexual and reproductive health rights (SRHR) questions. 

Trusted sources of information in Arabic about SRHR are not easy to access in the Middle-East, according to Rokaya Kamel, Product Lead at Raseef22. Information is fragmented and the ecosystem is broken when it comes to accessing accurate and judgement-free SRHR information. 

The second goal was to serve all their readers, but particularly, young women and LGBTQ+ readers in having a safe space to ask sensitive questions, added Kamel. 

The problem: Safe access to SRHR information

They wanted what they were building to create a stigma-free, anonymous space that would reduce misinformation around SRHR information generally. They also wanted to focus information-sharing on taboo or legally sensitive topics.

The problem that they were looking to solve was informed by data. Raseef22 found that a third of organic searches from their readers were for articles related to SRHR on their website. They took it as an educated signal that young people were searching for information that they couldn’t reach.

“We thought about ourselves and how would we solve a similar problem like that before the internet? We would ask our aunty. In the Arab world and the Middle East, it's very normal for you to ask your aunt before you ask your mother, because she's more culturally grounded,” explained Kamel. 

The thought led them to creating a chatbot for readers of Raseef22 using only their own and trusted partners’ articles and archives. In keeping with the tradition of asking an aunty, they aptly also named the app after it: Ask Aunty.

“It would be easier for people to ask the question and not search for individual articles. We decided to use AI because it would be much easier for people to just ask this question and not wait 24 hours for a response,” she added. 

Building the solution: A friendly AI chatbot

Interestingly, the team working on the Ask Aunty chatbot comprised four professionals from Raseef22’s marketing team: Line Itani, Diana Khanafer, Michel Abi Rashed. It’s the third AI project that the team is working on. They also hired an AI consultant, Fawzi Ammache, to support them in their journey.

“We have a little bit of background experience with AI. We were not very intimidated by it but also we were very proud because we know that thousands of people will actually benefit from this project because of how much of a taboo this project handles,” said Kamel.

They plan to internally test the chatbot for two months with hopes to release it to subscribers following that, and then to the public.

In terms of technology, the team initially wanted to use Delphi AI or Zapier. However, in the end they decided to build and use a custom GPT which Kamel said proved to be the “perfect” tool for their use case.

Choosing the right fit was not without its challenges, said Kamel. One of them was making the AI systems work in Arabic generally, and more specifically, reconciling the chatbots Egyptian Arabic with a knowledge base that consisted of Modern Standard Arabic content.

“It was very difficult and surprising how little the AI tool is adapted to Arabic. We had this problem that the language was completely butchered and in Ask Aunty. Arabic is in theory supported in these tools, but in practice the tools weren’t properly trained for it, so we faced many language-related problems. Thankfully we don't have this problem right now with custom GPT so it's way better dealing with the Arabic language,” explained Kamel.

They also faced issues early in the building phase when the data added to the custom GPT was ignored by the tool and it instead scoured the internet for answers.

“The second challenge was that we needed to ensure that chatbot answers were very accurate and very non-judgmental. They had to only use answers from Raseef22's articles and from trusted partners. We had this problem and now it's fixed. We have a closed dataset,” she said.

Focusing on editorial controls

Since all questions asked to the chatbot are anonymous, the organisation is also putting into place safeguards to make sure the system is not abused. While there is no such concern from a security standpoint, Kamel said, it can be from an editorial point of view.

“Since it's an SRHR chatbot, people cannot ask a question and turn this chatbot into something that it's not, for example - this is an outlet for me to ask sexual questions, so let's abuse that. We're trying to put editorial guidelines into the chatbot to be able to answer any kind of foul play,” she added.

Impact on the organisation

The Ask Aunty chatbot, apart from being a service to readers, is also positively impacting Raseef22’s operations. For example, it’s helping them identify what’s missing from their articles.

“Right now, we're expecting users to ask some questions that will open our eyes to some aspects in SRHR articles that we don't have. Now we can talk to the editors and tell them, ‘hey people are asking these questions. Let's make sure to add this to our article so we can answer them accordingly. Let's make our data bigger, let's enrich it so this will help us create more articles and better information,’” explained Kamel.

Future goals for the Ask Aunty chatbot include voice implementation, WhatsApp integration, and potential partnerships with SRHR organisations to use the API. They also plan to expand the database to include culture and history.

Lessons for newsrooms 

• Values over technology: One of the biggest lessons for Raseef22 that can apply to other newsrooms while building with AI is to always center organisational and editorial values in the technology. Kamel said that their newsroom prioritised these values over just implementing a technology solution. This made a difference to their considerations in building the chatbot whether it was maintaining privacy or inclusivity.

• Specify editorial guidelines for every project: The team created editorial guidelines specific to the Ask Aunty project that were different from the guidelines they used for other AI projects. This was to ensure accuracy and keeping in mind the sensitivity of the subject, Kamel said.