Real estate investment AI tool fails crucial test, new research shows – realestate.com.au

Real estate investors who put their trust in unreliable AI systems for investment advice face major financial risks, a new report shows.

A national study by financial services group MCG Number Surveyors, which tasked ChatGPT with listing suburbs that met certain key investment criteria, found that AI remained woefully inept at simple real estate-related information gathering tasks, delivering recommendations that were statistically questionable.

The report put AI through its paces from beginner to advanced mode, with mixed results, and noticeably worse when run in ‘Deep Research’ mode.

The report commissioned ChatGPT to list – with supporting evidence – the three key suburbs in the wider Adelaide region that would fit a $1 million budget with rental yields of 4 per cent or higher and would have the best growth potential over the next five years.

AI proposed a two to three bedroom unit at Mawson Lakes, a two bedroom unit at Port Adelaide and a two to three bedroom unit at Oaklands Park.

While the MCG Quality Surveyors report praised the insights it provided into growth drivers, proximity to transport and the accuracy of realestate.com.au’s matching data, it criticized the listing of units only, despite plenty of homes being available.

Mike Mortlock of MCG Quality Surveyors.

---

Report author and MCG Quality Surveyors Mike Mortlock warned against buyers’ agents and other investment groups using AI models to choose investment locations.

“There are a lot of agencies saying they are using AI,” he says.

“When we saw the results, we realized that caution needed to be taken when making major financial decisions in this area.”

Mr Mortlock said the suburbs and housing AI was singled out because good investment was problematic, as it often involved high-rise markets where conditions were not in favor of investors – in some cases because there was an oversupply of certain types of apartments.

“AI selected units in areas that were not performing better and at prices that did not match guidance,” he said.

“It would be difficult to argue that these are good choices.”

AI for investment advice? Yes, we’re not quite there yet…

---

The information provided to ChatGPT in “Deep Research” mode was much more detailed and the requested output was more specific.

In this mode, AI recommended Brompton/Bowden units, Salisbury homes and Christies Beach/Noarlunga Downs homes – earning praise for the insights it provided into growth drivers and the fact it solicited homes in two of the three, but criticism for a failure to match the data provided and an unacceptable level of accuracy.

“When we gave it more context, 10-year trends and provided a lot of data points, it performed worse,” he said.

“It was quite a concern, it was hallucinatory. The accuracy of the data was not there.”

Cameron Galloway can’t imagine relying on AI for investment advice. Photo: Jonathan Ng

---

Property investor Cameron Galloway was 35 when he bought his first property in Modbury North through a buyer’s agency and said it would be a concern for him to bring in AI to help with property searches.

“I don’t think AI is at that stage yet and sometimes it gives you the answer you want rather than what you need to hear,” he said.

“It’s a little bit complacent, so it’s a little bit concerning.”

– with Aidan Devine