I’ve read thousands of property descriptions over 25 years. Most of them are forgettable. “Sun-drenched living areas,” “moments to shops and transport,” “inspect to appreciate”—the same tired phrases recycled across every listing on realestate.com.au.

AI tools can genuinely help agents write better copy. But if you don’t direct them properly, you’ll end up with descriptions that are polished yet soulless. Here’s what I’ve learned from a few months of testing.

Start With the Right Input, Not the Right Tool

The biggest mistake agents make is opening ChatGPT and typing “write a property description for a 3-bedroom house in Marrickville.” That’s how you get generic rubbish.

AI is only as good as what you feed it. Before you even open the tool, gather these specifics:

• The actual selling points. Not “great location.” What makes the location great for the likely buyer? A 4-minute walk to the station? The café strip around the corner? The school catchment?
• Buyer profile. Young family upgrading? Downsizer? Investor chasing yield? The likely buyer changes everything about tone and emphasis.
• Recent comparable sales data. Clearance rates, average days on market, price movements—this context helps you frame the property honestly.
• The vendor’s story. Why did they love living here? Often the most compelling details come from the people who actually lived in the property.

The Prompt Structure That Works

Here’s a template I use regularly:

Write a property listing description for a [property type] in [suburb]. The target buyer is [profile]. Key features include [list specifics]. The tone should be warm and confident but not over-the-top. Avoid clichés like “sun-drenched,” “entertainer’s kitchen,” and “inspect to appreciate.” Emphasise [top 2-3 selling points]. Keep it under 200 words.

That last instruction matters. Portal descriptions that run 400+ words don’t get read. According to REA Group, buyer attention on listing pages drops significantly after the first few sentences. You want every word earning its place.

What AI Gets Wrong (and How to Fix It)

It invents features. I’ve seen AI add a swimming pool that didn’t exist and describe harbour views from a ground-floor apartment in Ashfield. Always fact-check every detail against the actual property.

It defaults to hyperbole. AI loves “stunning,” “breathtaking,” and “unparalleled.” Buyers are sceptical of that language. If the kitchen is nice, say it’s well-designed with quality finishes. You don’t need “a masterpiece of culinary design.”

It misses local context. AI doesn’t know that the north-facing backyard is the real selling point in Sydney, or that strata levies of $800 per quarter are actually low for the building’s size and amenities. You need to add the local knowledge that makes a listing feel authentic.

It ignores the transaction reality. Vendor paid advertising budgets, VPA expectations, whether a property suits auction or private treaty—none of this context reaches the AI unless you provide it. The description should match the sales strategy.

When to Add the Human Touch Back

Run the AI output through this checklist before publishing:

1. Does it sound like something you’d actually say? Read it aloud. If it sounds like a brochure written by committee, rewrite the awkward bits in your own voice.
2. Is there at least one specific detail only you’d know? “The morning light fills the living room by 7am” or “you can hear the bellbirds from the back deck.” These details can’t be generated—they come from inspecting the property.
3. Does it answer the buyer’s first question? For apartments, that’s usually strata costs and car parking. For houses, it’s land size and outdoor space. Don’t bury the essentials.
4. Have you removed every cliché? Search the text for “boasts,” “nestled,” “prestigious,” and anything that could describe literally any property in the suburb.

A Real Example

Here’s what AI gave me for a 2-bedroom apartment in Surry Hills:

“This stylish apartment boasts an enviable position in the heart of Surry Hills. Featuring two generous bedrooms, a contemporary kitchen, and a sun-soaked balcony, this residence offers the ultimate urban lifestyle.”

After editing:

“Two bedrooms, one bathroom, and a north-facing balcony big enough for breakfast—right on Crown Street. Updated kitchen with gas cooking. Common rooftop terrace with city views. Strata $1,100/quarter including water. Walk to Central, walk to everywhere.”

The second version gives buyers actual information and sounds like it was written by someone who’s stood in the apartment.

Getting Help With AI Integration

If you’re running an agency and want to build AI into your listing workflow properly—beyond just copy-pasting from ChatGPT—it’s worth talking to specialists who understand both the technology and the property industry. Firms like AI consultants Sydney can help you set up custom prompts, integrate AI with your CRM, and train your team without the usual tech headaches.

The Honest Summary

AI won’t make a bad agent good at writing. But it’ll make a busy agent faster, and it’ll help a decent writer produce consistent quality across 15 listings a month instead of three.

The agents getting the best results treat AI as a first draft—a starting point that still needs local knowledge, genuine voice, and the kind of specifics that reduce days on market because buyers can actually picture themselves living there.

Use the tool. Just don’t let it do the thinking for you. For now, Domain’s listing data and your own experience remain the best guides for what buyers actually want to read.