Category Archives: AI

Managing a mass FOI project? Here’s an AI-assisted methodology for that

Leave a reply

Sending FOIs to multiple bodies across the country to get the big picture on an issue sounds like a great idea — until the responses start to trickle in. Differences between responses often make mass FOI projects extremely time-consuming as you try to get everything into a format that allows you to ask journalistic questions and compare different authorities. Can AI help?

On one recent project I decided to put together a methodology that made the process less stressful, faster and more accurate. Here’s how it works.

Data structure Extract & reshape Check & verify Combine Audit & prioritise Audit responses to identify the level of detail in each response and identify edge cases. Include a caveats column. Augment manual audit with NotebookLM audit. Identify a priority order for data, e.g. totals by outcome, hospital, category or year where these are provided separately Design a data structure that can accommodate all responses Structure should follow ‘tidy’ data principles, i.e. one row per combination of features (force, category, hospital, outcome, year) Structure should include source details, e.g. filename, sheet name, name of person entering data PDFs: use Tabula or vibe coding (design a prompt template to generate code to attempt to extract data). Multi-sheet XLS files: use Open Refine to import and combine sheets Design a prompt template for generating code to reshape CSV responses Manual checks (e.g. compare entries, check page-ending rows) Analysis-based checks (e.g pivots, totals) AI-based checks using a prompt template (e.g. compare files) Use OpenRefine or: Design a prompt template for generating code to combine the resulting CSV files.
Continue reading →

“Many newsrooms are not optimised for what humans do best” (but we have an opportunity to change that)

Leave a reply
Amazon worker with horse head
Image by Cory Doctorow, from Revenge of the Chickenized Reverse-Centaurs

Some essential reading by Agnes Stenbom Swedling explores how news organisations integrate AI into their workflows and the idea of the “human in the loop“. Many newsrooms, she points out, “are not optimised for what humans do best”, and so far the introduction of AI hasn’t involved a critical consideration of whether we want to embed those features in new systems, or rethink them:

“What is being built – incrementally, often unintentionally – is a form of machine-centric hybridisation. Workflows are optimised for what machines do well: speed, scale, pattern recognition, cost efficiency. Humans are then positioned around those systems, adapting their tasks, roles, and decision-making to fit the logics of machines. 

“The consequence is a subtle but significant inversion: rather than engaging in uniquely human activities, work is reorganised to fit machine-driven processes. And once that inversion is embedded at the infrastructural level, it becomes increasingly difficult to reverse.”

Continue reading →

How to: generate hundreds of maps by combining QGIS with Python (code included!)

Leave a reply

At this year’s Dataharvest I delivered a workshop on using Python in QGIS to automate the process of exporting maps for multiple locations. Here’s how to do it (you can find a GitHub repository with materials and links here).

Making a map for a story is cool — but what if you could make a map for every reader? Or if you’re working on a project involving teams in different regions or countries, what if you could give each one of those teams a map centred on their own patch?

Normally you would have to manually move the map to centre it on a key city, and then export an image. Then do it again and again and again for every area.

Luckily, QGIS has the ability to run code. And this is a great excuse to start using it.

By organising the layers on the left you can put shapes such as flood defences over a base OpenStreetMap layer. You can also change the scale in the box underneath the map
Continue reading →

Parallel prompting: another way to avoiding deskilling with AI

Leave a reply
Train tracks
Photo by Markus Winkler

Too often discussion around using AI is “either/or” — an assumption that you either use AI for a task, or do it yourself. But there’s another option: do both.

“Parallel prompting“* is the term I use for this: while you perform a task manually, you also get the AI to perform the same task algorithmically.

For example, you might brainstorm ideas for a story while asking ChatGPT to do the same. Or you might look for potential leads in a company report — and upload it to NotebookLM to perform the same task. You might draft an FOI request but get Claude to draft one too, or get Copilot to rewrite the intro to a story while you attempt the same thing.

Then you compare the results.

Continue reading →

“I don’t want it to be easy” and other objections to using AI

Leave a reply

In September I took part in a panel at the African Journalism Education Network conference. The most interesting moment came when members of the audience were asked if they didn’t use AI — and why.

Continue reading →

How to stop AI making you stupid: hybrid destination-journey prompting

1 Reply
A local map-style illustration where a pinned "answer" destination is visible, but the route is overlaid with checkpoints labelled “confidence”, “sources”, “counter-arguments”, “verify”, “edit” (image generated by ChatGPT).

Last month I wrote about destination and journey prompts, and the strategy of designing AI prompts to avoid deskilling. In some situations a third, hybrid approach can also be useful. In this post I explain how such hybrid destination-journey prompting works in practice, and where it might be most appropriate.

Continue reading →

FAQ: How has journalism been transformed?

1 Reply

In the latest FAQ, I’m publishing here answers to some questions from a Turkish PR company (published on LinkedIn here)…

Q: In your view, what has been the most significant transformation in digital journalism in recent years? 

There have been so many major transformations in the last 15 years. Mobile phones in particular have radically transformed both production and consumption — but having been through all those changes, AI feels like a biggest transformation than all the changes that we’ve already been through. 

It’s not just playing a role in transforming the way we produce stories, it’s also involved in major changes around what happens with those stories in terms of how they are distributed, consumed, and even how they are perceived: the rise of AI slop and AI-facilitated misinformation is going to radically accelerate the lack of trust in information (not just the media specifically). I’m being careful to say ‘playing a role’ because of course the technology itself doesn’t do anything: it’s how that technology is designed by people and used by people. 

Continue reading →

7 técnicas de design de prompts para IA generativa que todo jornalista deveria conhecer

1 Reply

Ferramentas como o ChatGPT podem parecer falar a sua língua, mas, na verdade, falam uma linguagem de probabilidade e suposições fundamentadas. Você pode fazer-se entender melhor — e obter resultados mais profissionais — com algumas técnicas simples de prompting. Aqui estão as principais para adicionar ao seu kit de ferramentas (Este post foi traduzido do inglês original usando o Claude Sonnet 4.5 como parte de uma experiência. Por favor, avise-me se encontrar algum erro ou traduções incorretas).

Técnicas de design de prompts para IA generativa Prompting de função Prompting de exemplo único Prompting recursivo Geração aumentada por recuperação Cadeia de pensamento Meta prompting Prompting negativo

Prompting de função

O prompting de função envolve atribuir um função específico à sua IA. Por exemplo, você pode dizer “Você é um correspondente experiente de educação” ou “Você é o editor de um jornal nacional britânico” antes de delinear o que está a pedir que façam. Quanto mais detalhes, melhor.

Há pesquisas contraditórias sobre a eficácia do prompting de função, mas no nível mais básico, fornecer um papel é uma boa maneira de garantir que você fornece contexto, o que faz uma grande diferença na relevância das respostas.

Continue reading →

At its best, AI can help us to reflect on our humanity. At its worst, it can lead us to forget it

Leave a reply
Frankenstein's monster and Maria

Almost all conversations around AI come down to these hopes and fears: that at its best AI can help us to reflect on our humanity. At its worst, it can lead us to forget it — or subjugate it.

When AI is dismissed as flawed, it is often through a concern that it will make us less human — or redundant.

The problem with this approach is that it can overlook the very real problems, and risks, in being human.

When people talk about the opportunities in using AI, it is often because they hope it will address the very human qualities of ignorance, bias, human error — or simply lack of time.

The problem with this approach is that it overlooks the very real problems, and risks, in removing tasks from a human workflow, including deskilling and job satisfaction.

So every debate on the technology should come back to this question: are we applying it (or dismissing it) in a way that leads us to ignore our humanity — or in a way that forces us to address our very human strengths and weaknesses?

4 ways you can ‘role play’ with AI

2 Replies
4 roleplay design techniques for genAI Rubber ducking Using AI for ‘self explanation’ to work through a problem. Critical friend/mentor Using AI for feedback or guidance while avoiding deskilling. Red teaming/ devil’s advocate Using AI to identify potential lines of attack by an adversary, or potential flaws/gaps in a story. Audience personas Using AI to review content from the position of the target audience.

One of the most productive ways of using generative AI tools is role playing: asking Copilot or ChatGPT etc. to adopt a persona in order to work through a scenario or problem. In this post I work through four of the most useful role playing techniques for journalists: “rubber ducking”, mentoring, “red teaming” and audience personas, and identify key techniques for each.

Role playing sits in a particularly good position when it comes to AI’s strengths and weaknesses. It plays to the strengths of AI around counter-balancing human cognitive biases and ‘holding up a mirror’ to workflows and content — and scores low on most measures of risk in using AI, being neither audience-facing nor requiring high accuracy.

Continue reading →