The ChatGPT Experiment

Ep 90 — Sharing How I'm Talking to PDFs and Getting Real Answers
Host: Cary Weston
Date: October 14, 2025

Episode Overview

In this episode, Cary Weston explores a hands-on experiment: using ChatGPT to “talk” to complex PDF documents—specifically municipal ordinances—to find reliable, verifiable answers faster than local officials could provide. Cary explains how he set up a custom ChatGPT project, shared two dense PDF ordinances as reference, and empowered the AI to act as his personal advocate. The episode's primary theme is about transforming curiosity into actionable capability, demystifying legal/technical documents using AI, and building trust in the process by demanding source citation.

Key Discussion Points & Insights

Context: Facing Technical Documents and Sparse Staff

• Cary is tackling the challenge of building on a property in a water-adjacent zone, where conflicting municipal ordinances apply.
• He notes the complexity: legal jargon, cross-references, and staff only available once a week.
  • “The code enforcement officer … is only available once a week. It's a contract-type basis.” (04:38)
• The difficulty isn’t about incompetence, but about part-time staff juggling multiple towns and ordinances.

Defining the Problem: Navigating Confusing Ordinances

• There are two main documents: a “land use ordinance” (overall zoning) and a “shoreland zoning ordinance” (for water-adjacent properties).
• They often reference or contradict each other:
  • “In each document they're saying that one precedes the other. But it's not, like I said, crystal clear.” (07:00)
• Cary is determined not just to wait for slow email responses; he wants to be prepared and informed himself.

The ChatGPT Project Approach

• Cary describes his method: making a dedicated ChatGPT “project” with clear instructions and context.
  • “I've loaded both documents into ChatGPT, and I've created a project … that is specifically built to help me navigate land use ordinances.” (08:36)
• The AI is told to act as his advocate—mirroring the code enforcement officer’s role—and always provide context-specific, cite-able answers.
• He also incorporates email communications with the municipality to keep the AI up-to-date and ensure continuity.

Methodology for Verifiable Answers

• Critical strategy: Restrict AI context strictly to the uploaded documents—no guessing, no extrapolation, no web searches.
  • “The only data set that I want you to use for context … are the documents that I’m giving you. And … when you give me an answer, I want you to cite the section inside the specific ordinance so I can verify.” (11:38)
• This approach lets Cary be confident in discussions with town officials, having citations ready:
  • “When I use it in a conversation with a real person, I will have a point of reference to point to.” (12:15)

Outcome: Gaining Confidence and Real Value

• Cary has found answers with ChatGPT that even the town official couldn’t locate quickly.
  • “I have been able to answer questions through the use of this particular project that the person that I'm working with at the municipality hasn't been able to.” (12:41)
• He emphasizes this is not about replacing experts, but empowering himself for informed conversations.

Notable Quotes & Memorable Moments

• On Overcoming Legal Jargon:
  “Ordinances are not written for common folks to understand. They're very legal and technical. … You can read it and kind of get into a circular state of confusion.” (06:09)

• On Setting Up ChatGPT Correctly:
  “I've told it in the instructions that the only data set that I want you to use … are the documents that I’m giving you. … And I want you to prove it. I want you to show me exactly where that's coming from. That's giving me the power to be confident.” (11:38 & 13:42)

• On the Importance of Verification:
  “Trust but verify. When I say when you answer a question from the document, I want you to prove it … Because I know one of the biggest things that I get from you and others is how do I trust? How do I know it's real?” (14:01)

• Encouragement to Listeners:
  “You can't break it. Try things, ask it questions, get feedback and see where it could fit into your … goals, projects, milestones, challenges.” (15:14)

Timestamps for Key Segments

• 00:00–03:00: Cary sets the stage—explains personal context and outlines the problem with local ordinances.
• 04:38–07:00: Details on the technical difficulties and how part-time staff and legal jargon complicate the process.
• 07:00–08:36: Introduction to “land use” and “shoreland zoning” documents; context for building project.
• 08:36–11:38: Walkthrough: how to set up a focused ChatGPT project, upload documents, and provide clear, goal-oriented instructions.
• 11:38–14:01: Cary’s core strategy—restricting ChatGPT to provided data, requiring citation, and the impact on his capability.
• 14:01–15:14: Reflections on verification, curiosity, and inviting listeners to experiment and apply similar approaches in their own lives.

Main Takeaways

• Practical strategy for using ChatGPT: Upload technical PDF documents, restrict the AI’s reference scope, demand citations, and use the AI as a trusted guide.
• Empowerment through preparation: With this system, anyone can interact confidently with complex documents or officials—even outside their expertise.
• Universal application: This method isn’t just for ordinances; it’s relevant to legal, health, customer service, or any field with dense documents.

Closing Words

Cary’s tone is direct, curious, and encouraging, constantly inviting listeners to try things for themselves:

“The most important thing moving forward in using ChatGPT is your own curiosity. … Try things, ask it questions, get feedback and see where it could fit into your personal or professional goals, projects, milestones, challenges.” (15:14)

Website: chatgptexperiment.com

Cary’s Mantra: Stay curious.