Case Study 2: The Journalist's AI Toolkit

The Setting

Marta Reyes is a reporter at a mid-sized metropolitan newspaper. She covers local government — city council meetings, zoning disputes, school board budgets, the kinds of stories that do not go viral but shape the lives of 400,000 residents. Her newsroom has shrunk by 40 percent over the past decade. She now covers a beat that used to be handled by three reporters.

When her editor announced that the paper was providing staff with access to AI tools, Marta was cautiously interested. She was not a technophobe — she used spreadsheets to analyze public spending data and had taught herself basic data visualization. But she was a journalist first, and she understood that her profession lives or dies on accuracy, attribution, and trust. If AI could help her work faster without compromising those values, she was on board. If it could not, she wanted nothing to do with it.

The Experiment

Marta decided to systematically test AI tools across different aspects of her work, keeping a detailed log of what worked, what failed, and what worried her. Here is what she found over three months of experimentation.

Task 1: Summarizing Public Documents

The challenge: The city released a 247-page budget proposal. Marta needed to identify the most newsworthy changes within a few hours.

What she tried: She uploaded the budget to an AI tool and asked it to "identify the ten largest year-over-year changes in departmental spending, expressed as both dollar amounts and percentages."

The result: The AI correctly identified eight of the ten largest changes. It missed two because they were spread across multiple line items in different sections of the document. It also flagged one change that turned out to be a formatting artifact — the same line item appeared twice under different labels, and the AI counted it as a new expenditure.

Marta's assessment: "Useful as a first pass. Saved me about two hours. But I still had to read the whole budget myself to catch what the AI missed. The dangerous part is that if I had not read it myself, I never would have known the AI's list was incomplete."

Task 2: Background Research

The challenge: A local developer was seeking a zoning variance. Marta needed background on the developer's previous projects and any legal issues.

What she tried: She asked a chatbot to provide a summary of the developer's history, including past projects, lawsuits, and regulatory actions.

The result: The chatbot produced a detailed, confident summary — which was approximately 60 percent fabricated. It attributed projects to the developer that were actually completed by a different company with a similar name. It mentioned a lawsuit that did not exist. It did correctly identify two real projects, but the details (dates, costs, sizes) were wrong.

Marta's assessment: "This was the moment I understood that AI is fundamentally different from a search engine. A search engine shows you sources. A chatbot shows you assertions. If I had published any of this without checking, I would have destroyed my credibility and potentially faced a defamation lawsuit. I will never use AI for biographical or legal research without independent verification."

Task 3: Generating Interview Questions

The challenge: Marta was preparing to interview the city's new police chief about a controversial change in use-of-force policy.

What she tried: She gave the chatbot context about the policy change and asked it to generate 15 interview questions, "including five that the police chief would find difficult to answer."

The result: The questions were good — surprisingly good. Several pushed beyond the obvious talking points. One question, about the statistical methodology used to evaluate the previous policy, was something Marta had not thought to ask. She added it to her list.

Marta's assessment: "This is probably the best use case I have found. AI is excellent at brainstorming and stress-testing your preparation. It does not replace reporting — it does not go to the city council meeting, it does not build relationships with sources, it does not sit across from someone and read their body language. But it can help you prepare more thoroughly."

Task 4: Drafting Stories

The challenge: Marta wondered whether AI could help her write routine stories faster — meeting recaps, brief announcements, calendar items.

What she tried: She gave the chatbot her notes from a zoning board meeting and asked it to draft a 400-word news story.

The result: The draft was structurally competent — it had a lead, a nut graf, quotes (which Marta had provided), and context. But it was lifeless. It lacked the telling details that make local journalism valuable: the resident who stood up and trembled as she spoke about losing her neighborhood, the board member who voted yes but looked physically pained doing so. The AI produced information without insight.

Marta's assessment: "If all you need is 'the board voted 4-1 to approve the variance,' AI can do that. But that is not really journalism. Journalism is: here is what it felt like to be in that room, here is what this decision means for real people, here is the context that explains why this matters. AI cannot do that. At least not yet."

The Policy

After three months, Marta wrote a set of personal guidelines that her editor later adapted as an informal newsroom policy:

1. AI may assist with research. AI does not do the reporting. AI can help identify patterns, generate questions, and summarize documents. It cannot replace interviews, source-building, or firsthand observation.

2. Every AI-generated fact must be independently verified. No exceptions. If it cannot be verified, it does not get published.

3. AI does not write stories that carry a byline. If a reporter's name is on the story, the reporter wrote it. AI-assisted elements (data analysis, document summaries) are disclosed in an editor's note.

4. Confidential materials are never entered into AI tools. Sources, unpublished documents, and investigation details stay out of any tool that might store or train on the data.

5. AI is disclosed to readers when relevant. If AI tools played a significant role in the research or analysis behind a story, readers are told.

The Broader Debate

Marta's experience reflects a broader conversation happening across journalism. The Associated Press, The New York Times, and other major outlets have published AI use policies. Some news organizations have experimented with AI-generated articles for routine content (sports scores, earnings reports), with mixed results — including incidents where AI-generated stories contained factual errors that were published without human review.

The fundamental tension is this: newsrooms are understaffed. AI can help reporters cover more ground. But the value of journalism depends entirely on accuracy and trust — and AI tools, as Marta discovered, are confident but unreliable. The efficiency gains are real, but so are the risks.

Discussion Questions

1. Evaluating the Tools: Of Marta's four use cases (document summarization, background research, interview prep, story drafting), which do you think is the most appropriate use of AI in journalism? Which is the most risky? Why?

2. The Fabrication Problem: Marta's background research produced content that was 60 percent fabricated. Why is this particularly dangerous in a journalistic context compared to, say, a student writing a term paper? What are the professional and legal consequences?

3. The Efficiency Argument: Marta's newsroom has been cut by 40 percent. If AI allows one reporter to do the work that used to require two or three, is that a net positive? What gets lost when journalism becomes more efficient but less human?

4. Reader Trust: Marta's policy says AI use should be disclosed to readers "when relevant." What does "relevant" mean? If AI was used to analyze data but a human wrote every word of the story, should that be disclosed? Where would you draw the line?

5. Comparing Contexts: Compare Marta's professional AI policy to Priya's academic AI policy from Case Study 1. What principles do they share? Where do they differ, and why?

6. Your Profession: Think about the career you are pursuing or a field you are interested in. What would a responsible AI use policy look like for that profession? What tasks would be appropriate for AI assistance, and which should remain exclusively human?