AI as a research assistant

Let AI help you with your research in a verifiable way.

‍From brainstorming ideas and critical questions to tracking down sources and checking a draft story: LLMs now deliver astonishing results. In this training you will learn to use various AI models without compromising yourself or company secrets. You remain firmly in control and decide what to do with AI’s suggestions. So we do not use AI as a writing automat.

Target audience

‍This training is for journalists, researchers, podcast and documentary makers, communication specialists, and NGO staff who work independently or in a team on in-depth investigations. The approach suits both professionals with little AI experience and advanced users who want to improve and structure their AI use.

The training is about applying AI, not about assessing its societal, moral, and ecological aspects (see the note at the bottom of this page). We can certainly discuss that, and I can point you to honest and worthwhile publications, but that is not the goal of this training. We assume you are genuinely curious about how the technology can make your research work lighter and more enjoyable.

What you will learn

• ‍‍Working with Perplexity
• Using Sonar web search, different models, and the ‘Projects’ and ‘Computer’ options
• Writing prompts that return verifiable answers
• Having AI create or improve prompts itself
• Finding strong prompts made and tested by others
• Stress-testing AI’s answers
• Having AI carry out tasks
• Connecting with other software (e.g. Google Drive and Notion)
• Detecting wrongdoing with AI
• Improving hypotheses, timelines, and source maps with AI
• Tracking down documents and actors with AI
• Fact-checking with AI
• Analyzing topics and people with AI
• Supplementing interview questions with AI
• Translating, summarizing, analyzing, and comparing documents with AI
• Tracking down and analyzing datasets with AI
• Final editing with AI (structure, style, content)
• Quality control with AI (red-team review of the draft story)

Structure

‍The structure follows the steps of the Story-Based Inquiry method: we use AI to find topics, improve our hypothesis, timeline, and source map, track down sources, supplement our questionnaire, analyze documents and datasets, and assess our draft story. We mainly use AI to search, structure, and question, but not to write. It is a dedicated assistant in a verifiable investigation chain.

Teaching method

The training combines compact explanations with examples and exercises. Participants apply the prompts to the components and documents of their own research. The results are then discussed in the group.

Duration

‍One day. The exact duration, schedule, and format — for example in-company, online, or on location — can be adapted to the organization and available time.

Outcome

‍After the training you will be able to:

• Use AI to turn a broad research question into a testable, journalistic hypothesis.
• Thoroughly and systematically track down stakeholders, documents and datasets.
• Force AI answers into verifiable output: source citation, literal quote, uncertainty label, and distinction between fact, interpretation, and unknown.
• Analyze long documents without treating AI as a summarizing authority.
• Create interview questions based on the compiled dossier.
• Use AI as a critical second reader and red team before publication.

Note

‍This is how I personally think about AI, after two years of intensive testing:

It is the best thing that has ever happened to us journalists, it is the worst thing that has ever happened to us.

AI is currently in the hands of people we cannot trust. It guzzles energy. It steals from media and creatives. AI has the potential to completely hollow out our profession. These threats must be averted quickly. Journalists must stay on top of that.

At the same time, AI is the most beautiful innovation I have experienced in more than thirty years in the profession; on par with the internet and the laptop. It is a fascinating piece of technology that lets us do our work faster and better. Who wouldn’t want a dedicated assistant that usually offers useful suggestions and takes annoying or time-consuming chores off your hands?

In their groundbreaking book “Journalism in the Age of AI”, authors Lewis, Zamith, and Dodds argue that every technology in history has been partly shaped by its users. The ‘Houses of AI’, as they call them, (OpenAI, Anthropic, Google, Meta, etc.) may have their own devious plans, but they still have to sell the technology. So in the end they will also have to take our wishes into account.

This is the moment, the authors argue, for journalists to put their demands on the table: this is how we want to use AI. Instead of ignoring the technology (which will be difficult), we can better try to shape it as much as possible to our own purposes — by devising safe and useful applications while at the same time combating the amoral sides of the AI houses in our columns and airtime.