Which AI Pitfalls Does an Investigative Journalist Need to Watch Out for Most?

AI models tend to “hallucinate.” The biggest threat is invented quotes, words attributed to the wrong people, and links to articles or sources that do not exist. Using data like this without checking it creates an enormous risk to the journalist’s reputation and to the good name of the people in the story.

Is It Safe to Enter Unpublished Data Into AI?

No. For ethical reasons and for the safety of sources, you should not upload confidential materials, sensitive data, or unpublished facts to AI models. The data processing policies of AI developers are often unclear, and a leak of such information could put third parties at risk.

Which AI Model Handles an Analysis of Hundreds of Pages of Court Files Best?

Claude (which understands context and hidden nuance in a single long file very well) and Gemini (which has an enormous context window, ideal for large sets) are best suited to working with very long documents. Tools such as Perplexity handle in-depth analysis of long files less well, because their main purpose is scanning the web in real time.

AI in Investigative Journalism: How to Search for Information, Check Facts, and Choose the Right Model

In this article you will learn:

How to use AI in journalistic research.
Which model errors and hallucinations to watch out for.
Which AI model to choose for a specific investigative task.
Which AI tools are useful in investigative work.

How Can AI Help in Investigative Journalism?

AI can be an enormous help to a journalist at the research stage. The keyword here is “help”. Artificial intelligence should be used as a supporting tool and a starting point for further investigation, not as the final answer.

AI models can work with extensive data sets, both those pulled from the internet and those uploaded by the user. They can analyze them at lightning speed and pick out the most important information.

Deep Research: A Useful AI Tool for Journalists

Deep Research, also called DeepSearch, is a particularly useful AI tool for an investigative journalist. Unlike an ordinary conversation, it works like an autonomous agent that takes a topic, searches the available sources on its own, cross-references specific pieces of information, and comes back with a finished report after several minutes or a dozen or so. The file is structured and includes links to sources, which makes it easier for the journalist to verify the information.

Where to find the Deep Search feature in practice?

ChatGPT: Deep Research mode is available in the app, on the website, and within the Plus, Pro, Team, and Enterprise plans.
Gemini: in the Gemini app and in Google Workspace, within the Google AI Pro/Ultra plan.
Perplexity: Research mode is available on perplexity.ai and in the mobile apps on the Pro and Max plans.
Grok: as DeepSearch in the X app and on grok.com, within the X Premium+ subscription or the SuperGrok plan.

These mechanisms work on the same basis, but they differ in details such as the number of sources searched or how long they run, which is why it is worth testing several of these tools. Pick one topic, assign it to different models, and then compare their results to see which tool hallucinates more often.

Does AI Check Facts Reliably? Opportunities and Pitfalls

Artificial intelligence makes mistakes, which is why it should not replace research entirely but only streamline it. AI can be as dangerous as it is helpful, so the information, advice, and facts it provides need to be verified on your own.

AI models are very good at cross-referencing dates and catching errors in documents that are hard to spot when you read them linearly. Serious problems appear with quotes, however, because AI often invents them, attributes them to the wrong person, or alters their wording.

This is especially dangerous in the work of an investigative journalist, where the reputation and good name of the people in the story are at stake, because an error like this can carry serious consequences.

Most artificial intelligence models cite the sources behind their answers, but it may turn out that the article they point to does not exist or does not contain the information they presented.

Learn how to find information and check its reliability so that you do not fall into the trap of AI hallucinations.

Does AI Check Facts Reliably? Opportunities and Pitfalls

Not every model will suit a given task. When choosing a tool, pay attention to a few criteria:

the ability to search the internet in real time: when you work on current affairs, make sure you are using an AI that has access to the latest information,
the capacity to analyze very long documents: some models may struggle with case files running to hundreds of pages, for example,
source citation: an AI that states where it took a given piece of information from is the safest tool in the work of an investigative journalist.

What Is the Best AI Model for Investigative Work?

There is no single universal AI model that can be called the best, because the choice depends on the specific goal the journalist wants to achieve. The table below shows how four popular tools perform against typical investigative tasks.

Criterion	Claude	ChatGPT	Gemini	Perplexity
Data recency	Good with web search enabled	Good with search enabled. Strong integration with the internet	Very good. Integrated with Google Search	Very good. Works like an AI-based search engine
Working with long case files	Very good	Good, but dependent on the selected model and plan	Very good, one of the largest context windows on the market	Limited, the tool is geared more toward synthesizing many shorter sources
Source citation	Cites sources when searching the web	Cites sources when searching the web	Cites sources, integrated with search results	Every answer comes with links to sources by default
Working with files (PDFs, spreadsheets, images)	Very good, especially when analyzing documentation, reports, and tables	Very good, broad format support	Good, integrated with Google Workspace	Considerably more limited than the other models
In-depth analysis of a single document	Very good at understanding context and nuance in a text	Good	Good, especially with very long materials	Weaker, the model is geared toward broad scanning of the web
What to use it for in an investigation	Analyzing case files, contracts, and reports; drawing hidden conclusions from long materials	All-round research work and text analysis	Working with very large document sets and within the Google ecosystem	Quickly checking current facts and preliminary research with ready-made sources

Remember that the capabilities of these tools change with every update and that many options depend on the specific subscription plan, so treat this table only as a point of reference when choosing the right AI model for a given task.

Journalistic Ethics and AI Models

Data privacy, sensitive materials, and previously unpublished facts are fundamental to a journalist’s work. Confidential information should not end up in AI models, especially when their data processing policy is unclear. Ideally, the content you hand to artificial intelligence for analysis should not contain information that could endanger the safety of third parties once it is made public.

Accuracy is one of the most important qualities of a good journalist, which is why every piece of information obtained from AI models requires double verification.

AI in Investigative Journalism: What Rules to Follow?

Artificial intelligence can meaningfully speed up the work of investigative journalists by analyzing huge data sets quickly, finding the sources they need, and spotting specific patterns that a human may miss. AI should not replace a journalist’s work, though; it should strengthen and streamline it. The key is to understand and recognize the models’ limitations and to verify the content they deliver.

AI is a powerful supporting tool in more than just journalism. Do you want to implement AI safely in your company but worry about making costly mistakes? Do not put your brand image at risk. Trust Delante! We know how to use artificial intelligence safely.