Commercial agencies are being offered AI in a handful of shapes, and each shape is good at one thing and weak at another. Here is a plain map by category, what the Australian Consumer Law says about AI-written listing copy, and where we think the real gain is. It is memory and timing, more than the writing.
Five categories, by what they do
| Category | Good at | Weak at | Check before you use it |
|---|---|---|---|
| General chat tools | Drafting letters, first-pass copy, and summaries of a document you paste in. | They don't know your leases or clients unless you paste them in each time, and they can be confidently wrong. | The OAIC recommends not entering personal information, and particularly sensitive information, into publicly available generative AI tools. |
| Add-ins to your CRM | Summaries of contact history, follow-up drafts, tidier notes. | Only as good as the records underneath. Stale contacts go in, stale summaries come out. | Where the data is processed, and whether the vendor's terms let it train on your inputs. |
| Listing copy generators | Quick first drafts of descriptions for a listing or a marketing pack. | They invent or overstate details: the permitted use, the size, the zoning, the parking. | Every figure and every planning claim, checked against its source before it goes out. |
| Valuation and data products | Starting points: comparable sales, yield data, market trends. | Weak on the specifics of the building in front of you. | Any number you publish needs a source and a date. |
| Private systems built around one agency | Connected to your mail, diary and records, and built around your rules. | Costs more to build, and only as good as its connections and the rules it was given. | What it reads, what it sends, who approves it, and what it cannot see. |
The categories overlap, and vendors don't always stick to one. Ask what a tool reads and writes. Its name tells you little.
What the law says about AI-written listings
In our reading, the Consumer Law attaches to what goes out, whoever or whatever drafted it. Section 18(1) is the general rule: "A person must not, in trade or commerce, engage in conduct that is misleading or deceptive or is likely to mislead or deceive."
For land, section 30(1) is more specific. A person must not make a false or misleading representation concerning "the location of the land", "the characteristics of the land", or "the use to which the land is capable of being put or may lawfully be put" (s 30(1)(d) to (f)).
For example, a draft from a tool describes a Dandenong South warehouse as suitable for a food manufacturing use, but the planning permit only allows storage. The words came from a tool, but the representation is yours, and so is the problem. Check the permit before you check the adjectives.
How we would test one
Before you pay for any of these, test it on a deal that has already closed. Take the file as it stood at the time, see what the tool would have flagged, and compare that with what happened. Keep a record of the misses as well as the hits. A tool that cannot show you its misses is asking you to take its hits on trust.
Ask the vendor for the same test. A useful answer is a worked example from a real file, with the names removed. If the answer is a headline percentage, ask what it measured and over which files.
What AI still can't do
These are our view:
- Judge a tenant's business. A tool can read a set of accounts. It cannot tell you whether the owner is about to walk away from a second site.
- Build the relationship. The call that gets the deal is still a person's, and the trust behind it is built over time.
- Negotiate in the room. It doesn't read the other side, or know when to stop.
- Know what the client didn't say. The most useful fact is often the one nobody wrote down.
Where the real gain is: memory and timing
For example, a landlord called Ruth Hadley mentions in March that her warehouse tenant is "looking at a second site". Her lease option falls due in November. The March remark sits in an email nobody reopens, and the option date sits in a file nobody reads until the notice window is closing.
A system that kept the March note beside the option date would flag it early enough to act. That is the gain. It is not the first draft of the letter. The writing is the part people already do well, and the easiest part to find elsewhere.
That is also why we would start from dates. Our piece on commercial lease option dates in Victoria sets out the calendar we would keep on each file. Before you buy any of these tools, ask what it reads, where the data goes, who approves what it does, and what it costs when it is used a lot. Our piece on questions to ask any AI firm has the detail.
Command is a private system that sets itself up around one business. It watches dates like these and the news that touches your clients, and tells you early. Give us one past deal and we'll show you the day you could have known. → First look
This is general information, not legal advice. Get advice for your own situation before you rely on any AI output in a listing or a contract.
Sources
- Competition and Consumer Act 2010 (Cth), Schedule 2 (Australian Consumer Law), ss 18(1) and 30(1)(d) to (f), compilation dated 16 September 2026: legislation.gov.au.
- OAIC, Privacy and the use of commercially available AI products (guidance), on publicly available generative AI tools and personal information: oaic.gov.au.