All Things AI · Season 2 · The Models · Episode 5
ChatGPT is the one to reach for when your hands are full and you have four minutes. The voice conversation is genuinely good, the phone app is the best of the category, and thinking out loud in a car park is a real use rather than a novelty.
What it is
The hosted service from OpenAI that most people met first, and the reason most people have an opinion about AI at all. Capable across writing, analysis, and reasoning, available in a browser and in an app.
Because it was first, it carries most of the industry’s assumptions about what this technology is, and a lot of those assumptions are about the weakest thing it does rather than the strongest.
What it is comparatively good at
The honest answer is not what it is marketed on.
Voice conversation
You talk, it talks back, and the exchange holds together well enough to be genuinely useful while driving or walking. It interrupts naturally and handles you changing direction mid-sentence. This is the single most underused capability in the whole category.
The phone experience
The app is better than its competitors in the specific circumstances an agent is usually in: standing outside something, between appointments, with one hand. Convenience is not a small feature when the alternative is doing it later, which means not doing it.
Being widely understood
Everything anyone has written about using AI applies to it, which makes it easier to learn from other people. That is a real advantage for a first tool even though it says nothing about the technology.
The use that actually matters
Thinking out loud, on the way somewhere, about something you are working through.
Talk for two minutes about a situation you are dealing with and ask what you might be missing. Talk through how you want to approach a conversation. Dictate everything you remember from an appointment while it is fresh and have it come back structured.
That last one is worth adopting immediately, because the alternative is the twenty minutes you spend reconstructing it that evening, or more commonly the version where you do not and the detail is gone.
Where it is weaker
It is a hosted service. Anything you tell it leaves your machine, which makes it the wrong place for client documents no matter how convenient it is.
It is not the tool for working across your own files. Attaching things one at a time is the ceiling, and that ceiling is low compared to what a model with folder access does.
And it is not the cheapest option for volume. For running something across fifty items, the fast inexpensive services and the free local ones both make more sense.
What to try in the first week
Two exercises, both using the capability people ignore.
Have an actual spoken conversation while driving to something. Not a query. A conversation, five minutes, about a real situation you are thinking about, where you talk in incomplete sentences and change your mind. The point is to find out that it holds up, because almost everyone assumes voice is a gimmick until they use it properly once.
Then, immediately after your next appointment, sit in the car and dictate everything you remember for three minutes without organising it. Ask for it back as structured notes. That single habit is worth more than most of what is written about prompting, and it takes less time than the version where you type it later.
Where this goes wrong
Three failures.
Putting client material into it because it is the thing that is open. Convenience is exactly why this happens and it is not a good enough reason. Voice-dictating your own impressions is fine. Uploading somebody’s documents is a different act.
Treating it as the only tool. Because it was first, a lot of agents evaluate the entire category through it, conclude the ceiling is what they experienced, and never find out what folder access or a local model or an adapter trained on their own writing does.
Never using voice. Most people use the best thing about it once and then go back to typing, which is using the least differentiated capability of the most convenient tool.
How it sits next to the others
Against the fast inexpensive services it trades cost for polish and convenience. Against local models it trades privacy and permanence for capability and an app that works properly on a phone.
The sensible arrangement for most agents is this one on the phone for thinking, dictating, and anything while moving, a local model for anything with a client in it, and something fast and cheap for volume.
That is three tools, and it sounds like more complexity than it is, because each has an obvious trigger. Hands full, use this. Client file, use local. Fifty items, use the cheap fast one.
Making dictation a habit rather than an experiment
The dictation use is the one that changes a week, and it only works if it happens every time rather than when you remember.
Attach it to something that already happens. The moment you sit back down in the car is a reliable trigger and it exists after every appointment without you having to build anything. Three minutes, unstructured, before you drive off.
Do not organise while you talk. Organising while speaking is what makes people avoid dictation, because it turns a two-minute brain dump into a ten-minute composition exercise. Say everything in whatever order it arrives, including the parts that seem irrelevant, and let the structure happen afterwards.
Then put the result somewhere consistent. Notes that live in a chat history are notes you will not find. Notes that land in the same folder every time become the material every other capability in this season runs on.
What it is not for
Worth stating directly, because the convenience makes it tempting for everything.
It is not where your long-form drafting should happen, because you are paying for a hosted service to do something a free local model does well and privately.
It is not where volume work belongs, because per item it is among the more expensive options and the cost is invisible until it is not.
It is not a research tool you can rely on without checking. It will produce a confident answer about a specific place, a specific rule, or a specific number, and being widely used does not make it more accurate about your particular situation than any other model would be.
A note on being the default
There is a real cost to a tool being everyone’s first experience, and it is worth naming because it shapes how this industry talks about AI.
Most agents formed their view of what this technology can do from typing into this specific box, getting a competent generic paragraph, and stopping. That experience is representative of one tool used in one way, and it is not representative of the category.
None of that is a criticism of the tool. It is a caution about generalising from it, and it is the reason Season 2 exists at all.
What it costs
A free tier that is enough to evaluate it and a subscription in the usual monthly range for the better models and higher limits. For most agents this is one of the one or two paid services worth having, and the voice capability alone justifies it more than the writing does.
Questions agents actually type
What is ChatGPT best for as a real estate agent?
Voice conversation while your hands are full, and dictating what happened straight after an appointment so it comes back structured. The phone experience is the best in the category, and convenience matters when the alternative is doing it later and therefore not at all.
Should I put client documents into ChatGPT?
No. It is a hosted service and anything you send leaves your machine. Dictating your own impressions is fine. For documents with somebody’s details in them, an open model running on your own computer handles that work well and keeps the file where it is.
Is ChatGPT the best AI for real estate agents?
It is the most convenient on a phone and the best at voice, which is a real and underused advantage. It is not the cheapest for volume, not the right tool for working across your own files, and not appropriate for client material.
Why is voice worth using instead of typing?
Because it works in the circumstances agents are actually in, which is standing outside something or driving between appointments. The three minutes of dictation after an appointment replaces twenty minutes of reconstruction later, or more often replaces nothing being recorded at all.
Season 1 is Claude Code for Real Estate Agents, filmed and on its own hub. All three seasons: AI for real estate agents.