You can write listing copy with AI that doesn’t sound like AI, and it takes about two minutes of editing per listing. A raw ChatGPT draft reads as machine-written not because of the em dashes or a shortage of “human” adjectives, but because the model makes claims with no numbers behind them and arranges them in the same few predictable shapes every time. Give it your verified facts, name the buyer it is writing for, and run one short edit pass to hunt the tells — and the same tool produces a description that reads like a specialist wrote it. Here is how, with the tells named and a prompt you can reuse.
Why AI listing copy sounds like AI
A language model writes the most probable next word. Its training data is thick with residential marketing and thin with commercial deal facts, so when you ask for “a listing description” its instinct is the genre it has seen most: warm, fluent praise. That produces two problems a commercial audience notices instantly.
The first is empty adjectives. “Prime location,” “exceptional visibility,” “turnkey opportunity” — each is a claim with no number a buyer can check. An industrial tenant does not act on “excellent warehouse.” They act on 32-foot clear height and 60 dock doors. When every sentence asserts quality without evidence, the copy reads as generated because generation is exactly what it is.
The second is shape. AI prose falls into a handful of structures so reliably that the structures themselves are the giveaway: the tricolon, three items in a row over and over; the “not just a building, but an opportunity” reversal; paragraphs of near-identical rhythm; a closer that tells the reader not to miss out. None of these words are wrong on their own. The pattern of them is the tell, and it is stronger than any single vocabulary choice.
So “doesn’t sound like AI” is mostly a substance problem wearing a style costume. Fix the substance — real numbers, a real buyer, varied sentences — and the style follows.
The em dash is not the tell
Somewhere in the last two years the em dash became internet shorthand for “a robot wrote this.” It is a myth, and believing it will send you editing the wrong thing. Professional editors have used the em dash for a century; it is a normal mark of confident English, and stripping it out to dodge suspicion just makes your copy worse (see The Ringer’s defense of the em dash).
The real tell is not punctuation. It is the absence of anything checkable. A human broker who knows a property writes “220 feet of frontage on a road carrying 41,000 vehicles a day.” A model that knows nothing writes “outstanding exposure.” Readers cannot always name why the second one feels hollow, but they feel it, and no amount of punctuation surgery will fix a sentence that never had a fact in it. Hunt for missing numbers, not stray dashes.
Give the model facts, not a topic
The single biggest change you can make is to stop typing “write a listing for my office building” and start handing the model a short fact sheet first. A blank space where a fact should be is an invitation for the model to invent one, confidently, and an invented square footage or zoning is not a typo. It is a misrepresentation your firm may have to answer for.
Before you write a word of prompt, assemble the facts the copy is allowed to use:
- Hard specs — square footage, clear height, floor count, year built, parking ratio, power and HVAC, load rating — each one you could defend in a dispute.
- Deal facts — asking price or rate, cap rate, NOI, lease type, tenant credit, remaining term.
- Location facts — submarket, zoning, highway and transit access, drive times, notable neighbors.
- The one buyer — the single decision this listing has to move. An industrial tenant asks whether their operation fits and runs. A value-add investor asks where the upside is and what it costs to capture.
That sheet is the model’s entire universe of permissible claims; anything not on it is barred from the draft. Fifteen minutes gathering verified facts saves an hour catching invented ones, and it is the step every copy-paste prompt list skips. This facts-first discipline is the spine of the deeper five-stage method in our listing copy framework, where to go once you want to systematize it across a team.
Here is a prompt that puts the facts in charge:
You are writing a commercial listing description for [the one buyer,
e.g. an industrial tenant deciding whether their operation fits].
Facts (use ONLY these; do not add, estimate, or embellish any figure;
if a fact is missing, leave it out):
[paste your fact sheet]
Rules:
- Lead every claim with a specific number or named fact.
- No adjective survives unless a fact behind it appears in the same sentence.
- Ban: prime, stunning, turnkey, unbeatable, exceptional, must-see.
- Vary sentence length. No "not just X, but Y." No closing "don't miss."
- House voice: [two lines of your firm's style, see below].
- Length: [target] words.Expect the first draft to still slip in an empty adjective or a stray tricolon. That is normal. The edit pass is what catches them.
The tells to hunt, with a before and after
The edit pass is mechanical and anyone on your team can run it. Read the draft once for each tell in the list below, fixing as you go.
| Tell | What it looks like | The fix |
|---|---|---|
| Empty superlative | “prime,” “exceptional,” “turnkey” | Swap in the number that earned it, or cut it |
| The tricolon | three items in a row, repeatedly | Break one list into a sentence; vary the count |
| Parallel negation | “not just space, but an opportunity” | Delete the reversal; state the fact plainly |
| Uniform rhythm | every sentence the same length | Cut one sentence to five words; let another run long |
| Hedging | “offers what could be an ideal setup” | Say it or drop it — “offers” plus a number |
| Hollow closer | “don’t miss this rare opportunity” | End on the strongest fact, not a nudge |
A before and after makes the difference concrete. Here is a raw model draft for an industrial listing:
A truly exceptional opportunity awaits at this prime industrial property. Boasting outstanding features, ample space, and unbeatable access, this turnkey facility is not just a warehouse but a gateway to growth. Don’t miss your chance to secure this rare offering.
Now the same property after the fact sheet and the edit pass:
38,000 square feet with 32-foot clear height, 60 dock-high doors, and a 200-foot truck court. The building sits 0.4 miles from the I-90 interchange in the [named] submarket, with 4,000 amps of power and a fully sprinklered interior. A national-credit tenant occupies half the space on a NNN lease with eight years remaining. Available for occupancy in Q1.
The second version is not warmer. It is denser, and density is what persuades a commercial buyer, because every line is something they can check before they pick up the phone. It reads as human because a person supplied facts a model could not guess.
Make it sound like your firm
“Doesn’t sound like AI” has a second meaning for a small shop: it should sound like you, not like a generic broker and not like the four other people in your office each publishing in a different register. A model has no default voice, so if you do not give it one, it borrows the average of the internet.
Write a two-line house style note once and paste it into every prompt: the words your firm uses, the words it never uses, how you present price and disclosures, and roughly how long your descriptions run. That single input does more for consistency than any tool feature, and it is what lets five brokers produce copy that reads as one firm. When that note lives inside the platform where your listings already sit — the subject of our explainer on what a CRE CRM is — the model drafts from your records and your voice in one motion.
The review that is not optional
No listing publishes on the model’s word. This is not a quality nicety in commercial real estate; it is a liability control. Real estate advertising is regulated, and a commercial listing carries misrepresentation exposure a residential post does not. A model can assert a zoning it invented or a square footage it rounded, and once that number is in an ad, it is a fact your firm has published.
A named second person confirms three things before anything goes out: every number matches the fact sheet; required disclosures are present and no phrasing crosses fair-housing or misrepresentation lines; and the copy sounds like your firm. On a one-page checklist the review takes minutes, because you are confirming verified inputs rather than repairing a draft that invented its own. The same accountability runs through every AI-assisted message a firm sends, which is why our guide to AI email assistants for brokers keeps a human on the send button too.
A tool or a trained broker
You do not need a specialized product to do any of this. A general chat model — ChatGPT, Claude, or Gemini — takes your pasted fact sheet and a saved prompt and produces the constrained draft above, at the per-seat cost you likely already pay. A listing tool such as Buildout drafts from the record you filled in rather than from a paste, and a CRM-embedded assistant like HubSpot’s drafts from your contacts and properties. The category you choose mainly changes where the facts reach the model, not whether the tells still need hunting.
Two cautions decide whether a tool helps. A grounded assistant drafting from a messy record produces confidently wrong copy faster than a person could, so cleaner data matters more with automation, not less. And a broker who cannot write a specific prompt or spot an invented tenant will produce hype with any of these tools. Basic fluency comes first, which is what the hands-on prompting training we run covers — writing listing descriptions, location write-ups, and market notes — at the low-thousands a focused workshop costs, before any platform decision. Capability before software is the through-line of the small-firm operating manifesto, and the full picture of AI across the inbox, CRM, and listing marketing sits in the CRE communications playbook.
Frequently asked questions
How do you write listing copy with AI that doesn’t sound like AI?
Give the model your verified facts first, name the single buyer the listing is written for, then run a short edit pass. Raw AI copy reads as machine-written because it makes claims with no numbers and arranges them in the same few shapes every time. Replace each empty adjective with the number behind it, break the repetitive rhythm, and cut the generic closer. The same model then produces a description dense with checkable facts, which reads as expert rather than generated.
What makes AI-written listing copy sound like AI?
Two things: unbacked adjectives and predictable structure. Words like “prime,” “stunning,” and “turnkey” assert quality without evidence, and commercial buyers act on numbers, not praise. Structurally, models default to three-item lists, “not just X, but Y” reversals, uniform sentence length, and a “don’t miss out” closer. The pattern is a stronger tell than any single word. Fix the facts and vary the sentences, and the fingerprint disappears.
Does using an em dash mean copy was written by AI?
No. The em dash is a normal mark of good English that editors have used for a century, and treating it as an AI signature leads writers to edit the wrong thing. The real tell is the absence of anything a reader can verify. “Outstanding exposure” reads as machine-written with or without a dash; “220 feet of frontage at 41,000 vehicles per day” reads as human because a person supplied a fact a model could not guess. Hunt for missing numbers, not punctuation.
Which AI tool is best for writing listing descriptions?
There is no single best tool, because the discipline lives around the model, not inside it. A general chat model such as ChatGPT, Claude, or Gemini works well at low volume with a saved prompt. A CRE-native tool like Buildout or a CRM-embedded assistant like HubSpot’s earns its cost at higher volume by drafting from records you already keep. The choice depends on where your listing data lives and how many listings you run, not on which product writes the smoothest sentence.
Can AI write compliant commercial real estate listing copy?
AI can draft it, but cannot make it compliant on its own. Real estate advertising is regulated, and commercial listings carry misrepresentation exposure, so a model that invents a zoning or overstates square footage creates a liability, not a typo. A named human must confirm every number against the source facts and check that disclosures are present and no phrasing crosses fair-housing or misrepresentation lines. Verified numbers remove much of the risk earlier, but not the review step.
How do I make ChatGPT write in my firm’s voice?
Write a two-line house style note — the words your firm uses, the words it never uses, how you present price and disclosures, and your typical length — and paste it into every prompt. A model has no default voice, so without that input it borrows the average of the web and every broker drifts apart. The saved note is a fixed input to the draft and a checkpoint in review, which keeps a small team publishing in one recognizable voice.
How long should an AI-assisted listing description be?
Long enough to carry the facts that answer the buyer’s decision, and no longer. For most commercial listings that is a tight paragraph or two: the hard specs, the deal facts, the location facts that matter to this buyer, and availability. Length is not the goal — density is. A short description built entirely from checkable numbers beats a long one padded with adjectives, and it is faster to review.
Do I still need to review AI listing copy before publishing?
Yes, every time, and the review is a liability control, not a style preference. A model can assert a fact it invented or round a figure it should not, and once that number is in a published ad it is a claim your firm has made. A named second person confirms the numbers against the fact sheet, checks disclosures and prohibited language, and confirms the voice. On a one-page checklist this takes minutes, because you are confirming verified inputs, not repairing a draft.
Should I use a listing tool or a general chat model?
Start with a general chat model and a saved prompt if you run a handful of listings a quarter; it costs almost nothing beyond a seat you already have. Move to a listing tool or a CRM-embedded assistant when re-typing facts and voice drift across several brokers become the bottleneck. The method is identical either way, so the tool decision is about volume and where your data lives, not about writing quality.
Where to start
The reason AI listing copy reads as AI is not the tool and not the punctuation. It is the missing facts and the predictable shapes. Put the facts back, name the buyer, and spend two minutes hunting the tells, and the same model produces copy dense enough to persuade and clean enough to defend. Begin with the two steps that need no software: build a short fact sheet for your next listing, and write your firm’s voice in two lines.
A free AI-readiness assessment turns this into a plan for your firm. A short working session maps where your listing data lives, your listing volume, the brokers involved, and your team’s current fluency, then returns a plain recommendation on how to run this without sounding generic — and whether a month of prompting fundamentals should come first. Book a free AI-readiness assessment before you commit to a listing-marketing platform.
Arthur Wandzel
