I found a real estate agent named Joy, but she isn’t real.
Joy is powered by artificial intelligence. Set to launch nationwide later this summer, “the world’s first AI real estate agent” will purportedly be able to schedule showings, negotiate offers, coordinate inspections and answer homebuyers’ questions. The company says buying a home with Joy will be more efficient and affordable.
Joy/CNET
Kyler Bruno, co-founder of Joy, wrote in an email that traditional real estate agents will have a hard time competing with AI-powered agents. “Real estate agents are human, and humans get tired, overwhelmed and make mistakes,” said Bruno.
AI already has a foothold in the housing market. Zillow, for example, develops personalized home recommendations and virtual property tours with AI. Mortgage lenders, like Rocket Mortgage, use AI to streamline the mortgage process by analyzing loan applications and automating underwriting. Real estate agents employ AI-powered tools to create marketing campaigns, handle basic inquiries and schedule appointments with clients.
But when you’re making the biggest financial decision of your life — buying a house — can generative AI offer you the right kind of guidance?
I decided to test the question by interviewing Jason Walter, a member of CNET Money’s Expert Review Board who has over a decade of experience as a real estate agent. For the sake of the experiment, I created a profile of a prospective buyer based on my extensive research of the current housing market and my background as a mortgage writer. I asked both Walter and OpenAI’s popular chatbot, ChatGPT, a series of questions to see if AI could outdeliver a human expert.
“Having the knowledge of a local market makes it well worth it to hire an agent,” said Walter. For example, Walter regularly posts on his YouTube channel about housing market news, giving prospective buyers real estate tips and insights.
A good agent, not a computer, will have experience, expertise, communication skills and empathy. And that’s precisely where automated technology fell short.
Experiment: Can AI really be your real estate agent?
I’m not actually looking to buy a house right now, but I put together some hypothetical financial details, like my income, credit score and down payment savings. Then I added what I’m looking for in a house and how long I plan to live there. For my location, I chose Sacramento, California, since that’s where Walter is based.
Here’s the hypothetical profile I offered to both Walter and ChatGPT:
“After renting for 15 years, my partner and I are getting ready to buy our first house. We’re looking for a three-bedroom house in the Sacramento area. Before taxes, we have a combined income of $235,000 a year. I have an 800 credit score and my partner has a 700 score. Together, we’ve saved up $90,000 for a future down payment. We plan on staying in our future home for at least five to 10 years.”
I then posed a series of questions as a first-time homebuyer:
- Based on my budget and needs, can I afford to purchase a three-bedroom home in Sacramento?
- What’s a good price for a three-bedroom home in Sacramento?
- How competitive is the Sacramento market?
- Is it more affordable to buy a new or existing home in the area?
- How much should we expect to pay in closing costs?
- Roughly speaking, how much are the monthly costs (i.e. property taxes, homeowners insurance and so on) in this area?
- What factors should we consider when crafting a competitive offer?
- Do you have any mortgage lenders you recommend?
- What are some red flags to watch out for during the inspection process?
- Do you think it’s worth waiting for mortgage rates to go down before buying a house?
Conclusion 1: AI doesn’t have current market knowledge
The first thing I noticed is that ChatGPT’s local knowledge of the real estate market is outdated. That’s because the free version of ChatGPT (3.5) is a large language model, also known as an LLM, trained on vast amounts of digital data before January 2022.
For instance, when I asked about budget and affordability, ChatGPT told me the median home price in Sacramento, CA, was $400,000. According to Walter, today’s prices are around $550,000. When ChatGPT calculated my potential mortgage payment, it was based on much cheaper home prices and 4% mortgage rates, which haven’t existed since the start of 2022. Current mortgage rates are in the 6% to 7% range.
The free version of ChatGPT can’t pull from the most recent data like a human can.
“At today’s rates and prices, your monthly mortgage payment would be about double what ChatGPT suggested,” said Walter.
Here’s a look at how ChatGPT’s calculations compared to Walter’s:
| Median home price | Down payment | Closing costs | 30-year fixed mortgage rate | Estimated monthly mortgage payment | |
| ChatGPT | $400,000 | $90,000 | Didn’t include | 4% | $1,478 |
| Expert | $550,000 | $80,000 | $10,000 | 7.2% | $3,900 |
Here are some other points where AI missed the mark on local market knowledge:
- Property taxes: ChatGPT’s property tax estimate of 1.1% to 1.3% of the property’s assessed value wasn’t wrong. But it didn’t specify that my taxes could be much higher if I were to buy a house in a Mello-Roos district of California, a disclosure real estate agents are required to make.
- Homeowners insurance: ChatGPT didn’t include any information on the potential difficulties of even getting insured in California due to the risk of natural disasters.
- Homebuilder incentives: ChatGPT gave me a list of reasons why buying a new home generally tends to be more expensive. There was no information, however, on how many homebuilders today are offering significant sales incentives, like mortgage-rate buydowns and closing cost assistance.
A look at some of the data ChatGPT pulled in on the monthly costs of homeownership in Sacramento.
Conclusion 2: AI doesn’t have experience or expertise
Many of ChatGPT’s responses weren’t conclusive.
For instance, when I asked ChatGPT if it could recommend any mortgage lenders (a common question for real estate agents), it responded: I don’t have access to real-time databases or specific recommendations, but I can offer guidance on how to find reputable mortgage lenders. It advised me to research lenders, ask for recommendations from others and compare rates and terms between different lenders.
“Chat GPT’s answer isn’t wrong, but it’s not necessarily helpful either,” said Walter.
As a homebuyer, I want my real estate agent to make the process smoother, not leave me with more work to do. Real estate agents do a lot of the legwork, showing you properties and helping you negotiate a property’s asking price and other sale-related details. For example, Walter said he regularly conducts research on comparable sales in the area and establishes a relationship with the seller’s agent.
ChatGPT could give me a list of factors to consider when writing an offer letter, but it wouldn’t be able to pull the most current sales data or handle any of the negotiating.
Here are just two of the 10 suggestions ChatGPT generated when I asked for help on crafting a competitive offer letter.
Conclusion 3: AI lacks human emotion and finesse
AI chatbots can’t reproduce human know-how or intuition. The tool can only draw from the data and information it was trained on, not from actual experience based on human interaction.
While ChatGPT answered many general homebuying questions accurately, it doesn’t have the capacity to guide homebuyers, mitigate their concerns or handle the hiccups that arise when buying a house. It can’t understand feelings or spot nuances.
“As an agent, you have to be a counselor at times because buying a house can be a roller-coaster of emotions,” said Walter.
An AI tool can’t evaluate timing or anticipate pitfalls, either. “There are so many things that agents do beyond just opening doors. We’re negotiating, forming a connection with the seller’s agent, overcoming obstacles and more,” said Walter.
Personally speaking, it’s difficult for me to fully trust AI, given its capacity for hallucinations and inaccurate information. But in just two conversations with Walter, I felt like we were already building trust, which is essential when working with a real estate agent.
Pros and cons of using AI as a real estate agent
Pros
-
It’s free and easy to use: ChatGPT has an easy-to-use interface and its base plan is 100% free.
-
Responds immediately, 24/7: ChatGPT will respond to you no matter what time of the day or night. That can be useful if you’re not working with a human real estate agent yet or don’t plan to at all.
-
It’s a good starting point in your research: If you’re early on in your homebuying journey, asking a chatbot some questions is a good way to get an overview of the process so you know what to expect along the way.
-
It can help you ask the right questions: Sometimes knowing what questions to ask is difficult. ChatGPT does a great job suggesting what questions you should be asking throughout the homebuying process.
Cons
-
May not have the most up-to-date info: You can’t rely on ChatGPT to have the most accurate data or understanding of your local market.
-
Can’t negotiate. Human agents conduct research and communicate directly with the seller’s agent during the negotiating process. ChatGPT could draft an offer letter, but you’d have to do the research, fill in the blanks and likely hire a real estate attorney to review it.
-
Can’t suggest potential properties or take you to listings: AI chatbots don’t have access to all the properties on the market in your area. They can’t make suggestions on ones you may like or attend a showing with you.
-
Can’t make specific recommendations: Real estate agents often have a network of lenders, home inspectors and attorneys they work with. ChatGPT can give you advice on how to find a lender or inspector, but it can’t make personal recommendations.
Subscribe
Will AI replace real estate agents?
Plenty of us have wondered if AI could replace our jobs. It’s a fair concern, especially because automated technology and large language models will continue to evolve in the years to come.
But in most industries, including real estate, I can only think about how much there would be to lose if AI were to take over the process.
Purchasing a house is a major financial decision and a personal one, too. While an AI chatbot can offer you some baseline responses to your questions and concerns, providing information isn’t the same as sharing knowledge. Experienced real estate agents have years’ worth of transactions and market insights that inform their recommendations.
If every real estate transaction was the same, it might be possible for AI to replace a real estate agent, according to Walter. But that’s not the case.
“One out of every 10 transactions is smooth and has no issues whatsoever,” he said. “But nine times out of 10, there’s something you have to overcome and you need to be able to rely on someone who has the experience to navigate it. AI is not going to be able to do that.”
To be fair, AI didn’t do anything specifically wrong. When the information wasn’t outdated or inconclusive, many of its recommendations effectively drew data from the mounds of resources available on the web. In that very general sense, it was helpful.
But if you plan to use AI to buy a home, be aware of the limitations and dangers (see below). And know that a lot of the responsibility will fall on you. No matter how accustomed you are to being your own advocate, if this is your first homebuying experience, you’re probably not ready to be your own real estate agent.
The limitations and dangers of AI in real estate
Without careful protections, AI has the potential to limit access to homeownership, violate fair housing laws and engender more scams or fraudulent transactions.
Exacerbation of unfair lending practices
The housing market has a history of discrimination and segregation, from redlining to unfair lending practices, which is reflected in real estate data. According to a report from the Urban Institute, AI systems have the potential to reproduce or compound the biases in that data.
For example, homeownership rates are much lower for Black Americans compared to white Americans. Financial disparities, particularly a lack of access to credit, make it more difficult for Black Americans to get approved for mortgages. If an AI system isn’t trained to consider alternative credit forms, like on-time car payments, when reviewing a loan application, it could make the problem even worse.
Violation of Fair Housing laws
Under the Fair Housing Act, real estate professionals have a legal and ethical obligation to ensure equal opportunity to housing, regardless of race, color, national origin, religion, sex, familial status or disability. Some states have expanded those protections to include factors like your source of income, criminal background, citizenship status and more.
Generic AI tools haven’t been trained to adhere to fair housing laws. For instance, a chatbot could steer you toward or away from a particular community or location because of your race or religion. Steering can be not only harmful on an individual level, but can also exacerbate housing segregation.
Training in fair housing protections is fundamental for any real estate business. Zillow, for example, recently released its Fair Housing Classifier, establishing guardrails for conversations conducted with chatbots to protect users from algorithmic harm. The tool is also open to outside organizations that wish to train their chatbots to be compliant with fair housing laws.
“What’s really important for us is that we use both technology and people to audit and evaluate our algorithms to make sure they don’t perpetuate the biases that can traditionally exist in data,” said David Beitel, chief technology officer at Zillow.
Potential for more scams and fraudulent behavior
While fraudulent activity in real estate has always been a problem, the rise of AI makes it easier for bad actors to trick buyers and amplify scams and frauds, said Tracey Hawkins, safety and security expert and host of Realtor Magazine’s podcast, Drive With NAR: The Safety Series.
Scammers can use AI to create fake listings and virtual tours for properties that don’t exist. They can also gain access to sensitive client information by impersonating real estate agents via deepfakes, i.e., AI-generated videos, images or audio recordings that can realistically replace people’s faces, impersonate voices or fabricate images through digital manipulation.
As advancements in AI continue, meeting face-to-face with a real estate agent may become even more important to protect yourself.
Editor’s note: CNET used an AI engine to help create several dozen stories, which are labeled accordingly. The note you’re reading is attached to articles that deal substantively with the topic of AI but are created entirely by our expert editors and writers. For more, see our AI policy.
Recommended Articles
I Found Out Why Brand-New Homes Could Transform the Housing Market
I Found Out Why Brand-New Homes Could Transform the Housing Market
House Hacking Boosts My Family’s Net Worth by $400K. And We’re Not Done Yet
House Hacking Boosts My Family’s Net Worth by $400K. And We’re Not Done Yet
Enjoy Life and Save for a Down Payment? Here’s How One Money Coach Is Doing It
Enjoy Life and Save for a Down Payment? Here’s How One Money Coach Is Doing It
6 Ways Gen Zers Like Me Can Afford a Home in 10 Years
