I'm sorry that I don't have any specific appraisal related prompts to give you since mine are specifically tailored to my work process and what I am trying to accomplish. My advice is to understand the basics on how LLMs work. They are not magic so they can be led astray in their reasoning easily. That's when they hallucinate. Many models don't remember things to keep in context unless you specifically tell them to. Many AI models will get "dumber" if you use them too much. This is often how the AI companies "throttle" usage. If you want to dive deep, get a subscription to one or more of the AI services. I have paid subscriptions to both ChatGPT and Claude but I also have several LLMs running locally on my machine for privacy reasons. The higher tiered subscriptions will allow you to keep your data private, to a point, and that can keep you from violating USPAP or worse. Don't be like so many lawyers and rely on AI to supposedly research and pull data for you, and spit out the results in a nice format, to only to find that some of the data and citations were bogus. I first broke down my work flow into the steps I normally take to complete an appraisal. Break down everything into individual steps, no matter how small. See which steps are duplicated, or which steps can be automated. I use WinTotal for my residential work, so I can link a spreadsheet to any field in the report. I use my local AI model to read my appraisal orders from the engagement letter, pull names, file numbers, or other information that is required in the report or for billing and put that into an Excel spreadsheet which I then paste into WinTotal. This eliminates a considerable amount of data entry. I do this for a lot of the report. For actual analysis work, I don't really let online AI models handle that. I did and continue to let AI models help me design spreadsheets that analyze just about anything. Because of AI models losing context so easily, I just let Excel do the math. Because I can see and review any formulas, I can understand and most importantly explain how I calculated something. All I do is then load data into the analysis spreadsheet I created, review the results, and decide what to use in the report. I use AI to pull and geocode addresses using ArcGIS shapefiles so that I can determine distances between the subject, comps or other locations. Even though I use ArcGIS, I can upload several shapefiles into Claude and ask it to find the nearest grocery store, or nearest Target or Walmart. For my mass appraisal work, I can ask it to find me "all commercially zoned parcels, closest to a lighted intersection, with at least 5,000 vehicles per day traffic." It's great for stuff like this. Keep your questions brief but direct, with little or no ambiguity. You can tell AI agents to go to certain sites and look up data for you, but unfortunately most of the sites we use tend to employ log in tests to keep data scrapers out. None of my AI agents can look up deeds or plat maps for me in the counties where I work because to log in they first make me "Select every picture with a bicycle in it." to determine if I'm human. As a side note, ArcGIS shapefiles are great because they typically contain a wealth of data about the parcel you are working with. Most gis map files of county GIS data will have an attribute table attached with the current tax data from the Assessor's Office for each parcel. You can have your AI agent pull that out, calculate latitude and longitude of each parcel, and put it all into a spreadsheet. This is how AI will change what we do as appraisers. I hope this helped.