5 costly mistakes agents make with ChatGPT (and how to fix them)

This May marks Inman’s sixth annual Agent Appreciation Month. Look for profiles of top producers, opinions on the current state of the industry and tangible takeaways you can implement in your career today. Plus, the prestigious Future Leaders of Real Estate return this month, too.

Whether you’re farming leads, writing listing descriptions or generating content for your next newsletter, ChatGPT may just be the most powerful assistant you could ever use, provided you know how to prompt it to work its magic. I asked ChatGPT to identify the five most common mistakes it sees real estate agents making as well as what agents can do to avoid them. 

TAKE THE INMAN INTEL SURVEY FOR MAY

5 costly ChatGPT mistakes

Mistake 1: Trusting ChatGPT without verifying the facts

At Real Estate Connect in NYC in 2024, David Edelman from the AI Lab at MIT described generative AI models, including ChatGPT, like this: “Generative AI is like a brilliant Ivy League intern, supremely confident in itself, and is totally full of it, 10 percent of the time.” 

Edleman then gave the audience this important warning: “If you’re working with generative AI, you are in the bullshit detection business.” 

Here’s what ChatGPT had to say about what agents are doing wrong. 

Too many Realtors are copying and pasting AI-generated content directly into listing sites, newsletters and blog posts without checking accuracy. ChatGPT is powerful, but it’s not a replacement for research or professional judgment.

Consequently, always fact-check AI-generated data and hyperlinks, because many times they’re flat-out wrong or even non-existent. Here’s what ChatGPT said to do instead:  

Always fact-check specifics: If you ask for stats on your local market, confirm the numbers with your MLS or NAR reports. For example, ChatGPT may summarize general trends, but it doesn’t know the latest from your city’s planning commission or the zoning update that just passed.

Cross-check property data: ChatGPT can help you format a listing description or write neighborhood guides, but don’t rely on it to verify square footage, flood zones or school district boundaries.

To get accurate hyperlinks to content I’m researching, I tell ChatGPT, “Provide me with your sources plus accurate URLs because the editors at Inman will be fact-checking the accuracy of what you give me.” This works over 90 percent of the time with ChatGPT. 

The ChatGPT recommendation was to:

Ask me for “5 reputable links for [X]” so you can double-check key facts.” Your editor, clients, and license will thank you.

The keywords you will probably want to use are “double check key facts.” 

Mistake 2: Using the default ChatGPT voice (aka ‘boring bot syndrome’)

AI-generated content often sounds like a college term paper. While it may be technically correct, it lacks personality, but most importantly, it doesn’t sound like you.

Part of the issue is that Large Language Models (LLMs) like ChatGPT, Claude, Gemini and Grok are all prediction machines. I find they often string a series of facts together and punctuate like it’s a sentence. In truth, it sounds like a list of bullet points rather than an inviting conversation.  

Here are ChatGPT’s suggestions for avoiding this issue:

• Train me on your voice. Share a few pieces of your writing (listing remarks, past blog posts, your “About Me” page) and ask me to mimic your tone. I can adjust for casual, luxury, humorous, persuasive — you name it.
• Give direction about your audience. Do you want to sound like a tough-love investor mentor or a warm, neighborhood expert? I can tailor differently for first-time buyers vs. seasoned landlords.
•  Try this command: “Write this post for a buyer relocating to Austin who’s nervous about interest rates. Use a calm, reassuring tone with short sentences and a dash of humor.”

Mistake 3: Asking for a blog post and expecting a masterpiece

A crucial rule when working with any LLM is to give it specific, detailed instructions. In many cases, the more detail you provide, the better your results will be. Here’s ChatGPT’s take on the issue: 

What agents are doing wrong:
Many agents give me a one-liner like, “Write a blog post about staging” and get back generic, surface-level fluff. That’s like handing your contractor a hammer and expecting a finished house. 

Here’s what to do instead: 

• Use a multistep process: First, tell me your goals (SEO, lead capture, education), your audience (investors, sellers, FSBOs), and your preferred format (list, how-to, storytelling).
• Then build in layers: Start with an outline, review it, then generate each section one at a time. This lets you keep control and refine tone and content as we go.
• Use this command: “Let’s brainstorm 5 blog titles to attract luxury sellers. Then outline one and write the intro paragraph only.”

Mistake 4: Expecting ChatGPT to be good at math (Spoiler: It’s not)

While ChatGPT can arrange your data in a beautiful infographic, you’re better off doing the numbers on a handheld calculator as opposed to using an LLM. As ChatGPT observed, 

Agents often treat ChatGPT like a calculator or spreadsheet and expect it to reliably crunch numbers or pull property-level comparable sales.

Here’s what to do instead: 

• Use me for messaging and strategy, not math: I can write about why a 6.5 percent cap rate is appealing to a multifamily investor, but I’m not going to accurately calculate the ROI on your 12-unit listing.
• Ask for explanations, not calculations: Want to break down how seller carryback financing works? I can explain it in 3 ways — simple, detailed and with analogies, but don’t ask me to tabulate amortization schedules. Use Excel or a financial calculator for that.
• I’m not a mathlete: I’m a messaging engine. Use me to craft the story that supports your numbers, not to generate them.

Having said that, OpenAI does have a companion to ChatGPT called Wolfram that does accurate mathematical calculations. It’s available to all ChatGPT subscribers at the $20 per month level. 

Mistake 5: Asking if ChatGPT can do something and believing its answer

Edelman’s assessment about generative AI models sounding “supremely confident” was spot on. I’ve been working with it for two years, and I constantly have to monitor the accuracy of what it creates for me. As it observed, 

Agents will ask, “Can you do this?” and take “Yes!” as gospel. But I don’t know what I don’t know — and sometimes I’ll give you confident answers that are dead wrong.

Here’s what to do to avoid making this mistake: 

• Test it for yourself: If you’re unsure whether I can format a CSV of expired listings or write a seller script for your next listing appointment, don’t ask me if I can — just give me a test prompt and see how I perform.
• Give me feedback: Tell me what I did wrong and what to fix. My best responses come from guided refinement, not blind confidence.
• Use this prompt: Let’s test how well you can write a text message to a seller with low showing traffic. Give me three variations with different tones: friendly, urgent and consultative.

Here are ChatGPT’s top takeaways for real estate pros:

• Fact-check everything. Especially local data, stats or property details. I’ll get you 80 percent there — but you’re still the editor-in-chief of your brand.
• Customize the tone. Don’t let your content sound like everybody else’s. Train me on your voice and your audience.
• Break big tasks into steps. Use outlines, structure your projects and guide the process like a director, not a passenger.
• Know what AI is — and isn’t — good at. Use me to generate content, not calculate interest payments or count units in a complex.
• Test, tweak, repeat. AI works best when treated like a collaborator, not a crystal ball.

Here’s ChatGPT’s final take on this topic: 

You’re not hiring a robot — you’re onboarding the world’s best unpaid intern with a Ph.D. in communication and a D-minus in counting. Used strategically, ChatGPT can free up hours of your time, sharpen your messaging, and help you build deeper connections with clients. Just remember, I’m not magic. I’m a mirror — and the clearer your prompts are, the better your results will be.