One unique type of document in UC Berkeley’s Institute of Governmental Studies Library’s collection is foldouts—brochures, tri-folds, and other non-traditional publications. Unlike thick reports, these materials are thinner, contain much less text, and fold together in various ways. However, they still offer valuable insights that can enhance research.

So, how do you efficiently analyze a foldout using ChatGPT? Here’s my baseline approach, which you can refine and expand based on your own needs. (Using a foldout pamphlet from the Crime Prevention Committee of Contra Costa County)

Step 1: Upload & Set Context

When uploading a foldout to ChatGPT, I start by framing the conversation so ChatGPT understands what I’m working with. I might say:

👉 "I have a foldout. Can I ask some questions about it?"

ChatGPT will typically respond with something like, "Sure! What questions do you have?"

At this point, I begin guiding ChatGPT in how to process the document.

Step 2: Extract the Text in Order

Next, I provide a simple instruction to organize the content properly:

👉 "Write out the text exactly as it appears on each page of the foldout, and mark the pages for me."

ChatGPT will then extract and structure the content based on how the information is laid out across the pages.

Step 3: Ask for Initial Insights

Once ChatGPT has transcribed the content, I ask for insights:

👉 "Do you see any key themes or insights worth exploring?"

If I already have a framework in place, ChatGPT will proactively generate insights based on that structure. But even without a framework, over time ChatGPT learns how I think, refining its responses based on what I typically consider significant.

Step 4: Verify & Challenge ChatGPT’s Assumptions

One of my core principles is ensuring AI is not left unchecked. I periodically challenge ChatGPT’s logic to keep it from relying solely on its internal reasoning.

For example, ChatGPT categorized insights under labels for each page of the foldout, including one titled "Geographic and Societal Implications”. However, it didn’t specify that this label and corresponding insights were from page seven—raising the question: Did ChatGPT infer this label on its own, or did it come directly from the document? In other words, what text in the foldout led ChatGPT to use the words “Geographic and Societal Implications”?

I asked ChatGPT…

👉 "How did you categorize this? What did you read in the foldout that led you to label it this way?"

This forces ChatGPT to justify its choices, helping me evaluate whether the categorization is accurate or biased. Here is the full text of ChatGPT’s response to page seven of the foldout…

👉 Why I Categorized It Under Geographic & Societal Implications

Location-Specific Focus
- The initiative is tied directly to Contra Costa County, CA, suggesting it was a local movement rather than a state or national effort.
- This raises questions about how crime prevention efforts were approached in different California counties. Was this committee’s approach unique, or did other counties have similar programs?
Regional Crime Patterns & Local Policy Impacts
- Understanding the crime rates in Contra Costa County at the time could reveal whether this initiative had a measurable impact on crime reduction.
- How did this county’s crime prevention strategies compare to others in California or nationwide? Were there similar committees elsewhere?
Community vs. Law Enforcement Dynamics in Contra Costa County
- The presence of a formal Crime Prevention Committee suggests that the local government and police chiefs saw crime as a shared responsibility between law enforcement and residents.
- This makes me wonder whether there were tensions between police and communities at the time—was this committee formed as a proactive measure to build trust, or was it a response to rising crime rates and public pressure?
Neighborhood-Level Involvement
- The Neighborhood Safety Project (Page 5) mentions community organization efforts, which might indicate that different areas within Contra Costa County had distinct crime challenges.
- Did urban areas vs. suburban areas experience crime differently? Was there a particular crime wave that led to this initiative?

Potential Next Steps

Research historical crime trends in Contra Costa County during the period this was published.
Compare this initiative to other California county-led crime prevention efforts.
Look into whether this committee still exists today or if it evolved into another program.

Would you like me to dig into any of these areas further?

Why Approaching Documents This Way Matters

You might think, "Why get so granular? Why question even bolded text or categories?" But that’s exactly what improves research accuracy. Every letter, every label, and every summary that ChatGPT produces—whenever possible—should be fact-checked and verified.

This human-in-the-loop approach elevates the quality and reliability of AI-assisted research. When researchers have structured strategies for verifying AI-generated outputs, they can trust their insights and ensure that important findings aren’t lost or misrepresented.

This method isn’t just for foldouts—it applies to any document type you analyze using ChatGPT. The key is to train the machine, set parameters, and keep it accountable for every output.

My name is Nick, and I enjoy teaching and speaking about the intersection of research, ChatGPT, and prompt engineering. My work focuses on developing easy-to-use frameworks and strategies that ensure AI doesn’t just generate answers, but also verifies and checks itself—helping researchers use ChatGPT more effectively and responsibly. If you have questions, need help setting up, or want to improve your prompts, feel free to reach out—I’d love to help!

Analyzing Foldouts with ChatGPT: A Step-by-Step Guide