Use AI to turn raw CMA data into a presentation sellers actually understand

You walk into a listing appointment. The seller has interviewed two other agents already. They each handed her a 22-page CMA packet she didn’t read. You can win this listing in two ways — by being more honest, and by making your CMA genuinely understandable. AI can help with the second one without compromising the first.

What sellers actually want from a CMA

Sellers don’t want a spreadsheet. They want answers to four questions, in this order:

What’s it actually worth right now? (a defensible range, not a single suspicious number)
Why? (the reasoning, not just the comps)
What can we do to maximize it? (your specific recommendations for THIS home)
What’s the realistic timeline? (and what could change it)

A CMA structured around these four questions wins listings. A CMA dumped as 40 pages of comps loses them.

The pre-AI work (you still have to do this)

Pull your comps the way you always do — MLS, your tool of choice. You need:

3–5 sold comps (last 90 days, ideally)
2–3 active comps (the competition)
1–2 expired or withdrawn (proof of what didn’t work)
The honest pros and cons of the subject property
2–3 specific recommendations for prep work, with dollar impact estimates

AI doesn’t pull comps. It turns them into something humans can read.

The CMA narrative prompt

📋 Prompt — copy & paste v1

Turn raw CMA data into a 2-page narrative summary for a seller.

SUBJECT PROPERTY:
- Address: [address]
- Beds / baths / sq ft / lot: [N / N / N / N]
- Year built: [N]
- 3 honest pros: [1, 2, 3]
- 2 honest cons: [1, 2]
- Condition reality (be honest, you'll edit): [dated kitchen, original windows, etc.]

COMPS DATA (paste table or describe):
- Sold comp 1: [address, price, sq ft, key differences from subject, days on market]
- Sold comp 2: [...]
- Sold comp 3: [...]
- Active comp 1: [address, list price, days on market, what they're competing on]
- Active comp 2: [...]
- Expired/withdrawn: [address, original list, why it didn't sell — your read]

LOCAL MARKET CONTEXT:
- Inventory trend last 90 days: [up/down/flat]
- Median DOM in this segment: [N]
- Pricing vs list price ratio: [N%]

MY RECOMMENDATION (you'll preserve in output):
- Suggested list price range: $["[$low – $high]"]
- Reasoning in one sentence: [your call]
- 2–3 prep recommendations with dollar impact: [paint exterior +$8–12k, refinish
floors +$10–15k, etc.]
- Realistic timeline at recommended price: [N weeks]

OUTPUT — produce this structure:

PAGE 1 — THE NARRATIVE
- Opening: 2 sentences setting the market context plainly
- "What your home is worth right now" — defended price range with the
reasoning in plain language. Include the 3 strongest comps by name and
ONE-line each on why they're relevant.
- "What works in your favor" — 2–3 sentences on the pros, specific
- "What's working against the price" — 2–3 sentences acknowledging cons
honestly without alarming
- "What I'd recommend" — the prep work list with dollar impacts and a sentence
on the why

PAGE 2 — THE SUPPORTING DATA
- A clean comp grid (Markdown table): comp address | sold/active | price | sq ft |
$/sqft | days on market | one-line key difference
- A 60-word "what didn't work" paragraph using the expired/withdrawn comp(s)
- A 60-word "current competition" paragraph using the active comps

CONSTRAINTS:
- Plain language, no jargon. "Days on market" instead of "DOM" on first use.
- The price range you state must be defensible from the comps shown — don't
inflate it.
- 8th-grade reading level
- Tone: direct, warm, professional. No "I'm confident we can get you top dollar"
or "your home is unique and special". Both of those make sellers trust you less.
- Length: 1.5–2 pages worth, not more.

Works in: ChatGPT · Claude · Gemini · Always verify every comp detail; AI will sometimes summarize numbers wrong

What this delivers vs. the default 22-page CMA

	Default CMA packet	AI-generated narrative
Length	22+ pages	2 pages
Read by seller	Skimmed	Actually read
Price defended	Implied	Explicitly reasoned
Honest about cons	Rarely	Yes, by structure
Recommendations specific to property	Boilerplate	Tailored

You still hand them the full comp data — but you put it in an appendix, not at the front. The narrative is what wins the conversation.

How to use it in the actual appointment

Print 2 copies. One for them, one for you.

Walk through page 1 verbally. Don’t read it — narrate. Pause at the price range and let them respond. Most sellers’ first reaction is the actual number they had in their head; this is where your honesty about cons earns its keep.

Then hand them page 2 and the appendix to read at their own pace.

Mistakes to avoid

Letting AI invent a comp’s details. Verify every comp number AI puts in prose against your MLS pull.
Inflating the range to win the listing. Sellers can sniff this. The agent who quotes the same range as the others, with better reasoning, wins.
Skipping the prep recommendations. This is the section that signals you actually know what their home needs vs. just running comps.

The bigger pattern

This same prompt structure works for every “translate raw data into a narrative for a non-expert” job an agent has — buyer market summaries, neighborhood investment outlooks, even quarterly market updates for past clients. You provide the data and the call; AI provides the structure and the language.

The work that’s still yours: knowing the market, picking the comps, making the honest call. Those don’t outsource. The page-2 prose, though, is what AI was made for.