How a small Nigerian newsroom used AI for a flooding investigation

In October 2024 floods devastated a farming community in southern Nigeria. Farmers in Arukwo, Rivers State, were displaced from their homes and forced to harvest crops early. The disaster killed more than 300 people across the country, affecting 1.2 million people.

This was a story that the independent, climate-focused investigative newsroom The Colonist Report had long been preparing for. Founded in late 2023, the outlet works with freelancers to report on local stories from developing economies, with a particular focus on Africa. 

“It’s stories that aren’t being told, especially in the Global North,” says founder Elfredah Kevin-Alerechi, 35, who works two-part time jobs while running the newsroom. “We don’t want to report something that audiences can get elsewhere. We try to give local people the voice they need.”

In March 2024, one of the newsroom’s freelance reporters visited the region around Arukwo after a community member raised concerns about the devastating impact that flooding could have on the area. The journalist collected interviews and continued to work on the story throughout the year, gathering data on flood risks and support schemes.

When flooding hit the area in October, Kevin-Alerechi expedited the investigation and asked why government programmes aimed at mitigating the impact of flooding had not helped Arukwo’s population.

AI tools helped with sourcing and analysing more than 3,000 pages of documents, fact-checking the analysis and visualising the results on a scale not previously possible for this small newsroom, which is currently funded by its founder. The result? An investigation into how marginalised communities are receiving inadequate support, published just a month after the October floods.

How The Colonist Report used AI

The Colonist Report was one of the smallest and newest newsrooms taking part in JournalismAI’s Sub-Saharan Africa Academy for Small Newsrooms in September 2024. During the programme, Kevin-Alerechi learned that the use of AI should focus on what her newsroom could not already do. On-the-ground reporting with sources without phones or internet is a task for her reporters. AI should be used to increase the volume of government documents the newsroom can handle and to improve the efficiency with which these can be analysed.

Kevin-Alerechi took on the role of prompt engineer and used paid-for versions of ChatGPT and Gemini to seek information on funding for flood victims in Rivers State. She ran the same prompts in each to compare the results. A free-to-use version of Copilot was used for basic spelling and grammar checks while drafting the report.

Initially, the information returned came from a diverse range of sources, including existing news reports. Altering the prompt to look for information from verified government websites produced more promising results, including links to lengthy budget documents from multiple government ministries. These documents, often downloaded or extracted with the help of Gemini, covered two to three years, and were then analysed using prompts to find evidence of funding allocated to specific communities or states.

Here’s an example: one document was identified by ChatGPT as containing details on flood-related issues in Rivers State. Kevin-Alerechi then prompted ChatGPT to dig into specific details, asking how much flood-support funding was allocated according to this particular document, and on what page this information was located.

“Shockingly, when I went to the page number, the page wasn’t there,” she says. “So I used the same prompt to find the right page with Gemini and then went back to ChatGPT to tell it that what it gave us was not correct. I asked it to check again and it came back with a new and correct page number.”

Without these tools, the newsroom would previously have gone through thousands of pages of information manually, using search functions within PDF readers, for example, to track down incidences of specific names or figures.

What they learnt

There were lessons about the best tools for the job along the way, says Kevin-Alerechi. One government agency had converted image files into PDFs – an issue that creates a lengthy delay for ChatGPT’s analysis. “As I'm speaking to you now, the file [analysis] has not ended,” she says, adding that Gemini did the job in under three minutes. In another investigation, she has had to correct the names of places and communities in the source data used by AI tools to create a visualisation, because the first images generated included misspelt place names.

Using AI has helped The Colonist Report expand its reach and resources. It previously used Canva and Excel to structure and visualise data for analysis and publication. For the flooding investigation it used ChatGPT and Gemini instead. 

Google Notebook LM was used to generate an automated podcast based on the published article as a way to further its distribution without spending any of their budget on audio.

Prior to publication, Kevin-Alerechi asked ChatGPT and Gemini to suggest any legal issues with the investigation that she should consider. “I told them to look at the story we’ve written and added that we were a small newsroom without money for legal issues, without any backup,” she explains, assigning these AI tools a role: “I asked them to check if we had used any word that may lead to litigation."

Whether it’s for legal checks, visualising data or fact-checking responses to prompts, there always needs to be “a human in the loop”, she says. “We are going to consider every step: the risk, the ethics, the legal aspects that may arise.”

Balancing challenges with opportunities

For fledgling newsrooms trying to experiment with AI, the costs of tools can be prohibitive. Subscriptions of around $20 per month for tools like ChatGPT are a problem, says Kevin-Alerechi.

She encourages smaller newsrooms to use limited subscription packages or test out alternative tools when a subscription cap is reached.

Kevin-Alerechi is excited by how AI might help The Colonist Report grow its audience through additional products, such as auto-generated podcasts. This technology could preserve precious resources for data analysis, visualisation and investigations. She’s exploring how it might help her write successful grant applications too.

“Newsrooms in the Global South should keep abreast of what is happening in the Global North to see what the new thinking is around AI technology,” says Kevin-Alerechi, who also runs JournoTECH, a training platform for journalists that plans to offer more training for journalists on AI. Her advice to other journalists? “Try to play with strategy, setting it up and understanding any legal aspects, and reach out to people using it.”