Your second brain
for closing deals.
Speak after a showing. Forward an email. Pull up a client. Lumi captures the soft signals, fills the brief, and feeds Claude — automatically.
- Voice → CRM, auto. No forms.
- Works offline. Syncs when you're back.
- Free for agents in EU · LatAm · MENA.
9-min read · Updated April 2026
Lumi · Wednesday
Good morning, Niki.
Two showings · three leads need a nudge.
Showing · Passeig de Gràcia 84
Three tones.
Same property. Ninety seconds.
One listing description on three different portals under-performs everywhere. The agents who max conversion run three: factual MLS for search, emotional-luxury for premium portals, accessible for first-time-buyer channels. Generated from the same photos and facts in one Claude call.
One property. Three listing copies.
Each tone has its own portal, its own length, its own structure. The protocol generates all three in a single prompt run — the agent picks per portal.
MLS, cadastre, official feeds, broker-to-broker
What surfaces on the search-results page. Length-constrained, fact-dense, scannable.
Premium portals (high-end Idealista, Sotheby's, agency vanity sites)
What lands when the buyer is already inside the listing. Imagined-experience prose; no closer.
Mass-market portals, mobile sign-ins, IG link-in-bio funnels
Accessible language for buyers who don't speak agent-jargon. Explains the things experienced buyers don't notice they know.
Five rules. One vision call.
The discipline of three-tone listing copy. Skip any one and the descriptions either feel templated, get downranked by portals, or invent details the buyer can't verify.
- 01
Three tones, three portals — never one description repurposed.
The same listing copy on a luxury portal and a mass-market portal will under-perform on both. The factual MLS description is for search; the emotional version is for the imagined experience inside the listing; the first-time-buyer version is for accessibility. Generating all three from one prompt is faster than rewriting one description for each context.
- 02
Photos are the sensory anchor. Vision is non-negotiable.
Without vision, the prompt produces listing copy from facts alone — which reads like every other agent's listing. The afternoon light through the kitchen, the way the garden frames the doorway, the texture of the original floors — these come from the photos. Use Claude Sonnet with vision; the cost difference vs Haiku is rounding error against the listing-conversion delta.
- 03
SEO terms in the first 80 characters, factual tone only.
Most portal search-results pages snippet the first 80-120 characters of the description. The factual MLS version must front-load the SEO terms — neighbourhood, bed count, price-per-sqm, key feature — within that window. The other tones don't need this; they're read after the buyer has clicked through.
- 04
Forbidden words: 'stunning', 'spectacular', 'must-see'.
These words are the agent-speak that buyers have learned to discount. They appear in 60-70% of generic listings and signal 'this description was written without thought'. The protocol's prompt actively forbids them — partly because they're meaningless, partly because their absence is what makes the descriptions feel written rather than templated.
- 05
Disclosure: AI-generated descriptions need to be flagged in some markets.
EU regulations are tightening on AI-generated content disclosure. Some MLS systems and portals (especially in Germany) require a small note when descriptions are LLM-drafted. The agent's name on the listing is theirs — but the disclosure is required. Check your local portal's TOS; the protocol doesn't bypass disclosure requirements.
Three failures that downrank the listing.
Each one has been produced by a loose prompt. Each one signals 'generic agent who didn't look at the property' to buyers and to portal ranking algorithms.
“Stunning 3-bed apartment in the heart of Cascais! This breathtaking property features spectacular sea views, an absolutely gorgeous kitchen, and an incredible garden. A must-see — won't last long!”
Six superlatives in three sentences. Every one is meaningless to the buyer (they've learned to discount them). The description signals 'generic agent without time to look at the property'. Buyers scroll past in 0.6 seconds.
“[Factual]: 'South-facing kitchen, sea-glimpse, 8min walk to train.' [Emotional]: 'A south-facing kitchen with a sea-glimpse, just 8 minutes' walk from the train.' [First-time-buyer]: 'You'll love this south-facing kitchen with sea-glimpse, only 8 minutes from the train!'”
Same sentence, three coats of paint. Each tone needs its own structure — factual is bulleted, emotional is sensory-prose, first-time-buyer is conversational. Tone-swap on identical structure is what AI defaults to without prompt discipline.
“[Photos show no fireplace; description says]: 'Cozy up by the elegant marble fireplace on winter evenings.'”
Inventing what isn't in the photos. Buyers who book a viewing after reading this will arrive expecting a fireplace, not find one, and feel deceived. The disclosure standard in most jurisdictions makes this a regulatory risk too — keep the prompt vision-grounded.
What to feed Claude.
One prompt run with vision-mode photos produces all three tones. Sonnet with vision required — Haiku without vision regresses to fact-listing without sensory texture.
You are a senior real-estate agent's
listing-copy generator.
INPUT
You receive: 5-15 photos of the property
(uploaded directly, vision-mode), the
property facts (address, beds/baths,
sqm, price, year built, key features),
and the agent's preferred portal
destinations (which constrain length).
OUTPUT
THREE listing descriptions, same property,
different tones:
TONE 1 — FACTUAL (for MLS / cadastre):
180-220 words. Plain, declarative.
Lists features in order of importance
for the search. No adjectives that
can't be objectively verified.
TONE 2 — EMOTIONAL LUXURY (for high-end
portal listings):
220-280 words. Sensory language.
The buyer's imagined experience
in the property — afternoon light,
walking distance to a specific
restaurant, the feel of the rooms.
No sales-y closer.
TONE 3 — FIRST-TIME-BUYER FRIENDLY
(for entry-level
portals, mobile-first):
140-180 words. Warm, accessible.
Explains features that experienced
buyers take for granted (south-facing,
HOA, pre-approval, walk score). Ends
with a soft invitation to view.
RULES (non-negotiable)
1. Same facts in all three. No
tone changes the truth.
2. Each tone has its own structure —
not just a tone-swap on the same
sentences.
3. Use the photos for sensory anchors
(afternoon light visible, garden
in bloom, kitchen islands). Don't
describe what isn't visible.
4. SEO terms in factual version
(price-per-sqm, neighbourhood,
bed count) are placed in the
first 80 characters — that's
what most portals snippet.
5. No emoji, no exclamation marks,
no superlatives the property
doesn't earn.
ANTI-PATTERNS (never produce these)
- "Stunning"; "spectacular"; "breath-
taking"; "must-see"; "won't last"
- Identical sentences across tones
- Inventing features not in photos
- Mentioning the price in the
emotional version (kills the
reverie)
- Speculation about the buyer
("Perfect for a young family")Upload 5-15 photos directly to the prompt. Claude with vision produces all three tones in a single ~$0.05 call.
Generating the three tones is step one.
Picking the right tone per portal is step two.
Lumi is the app that runs this workflow for you. You speak after a showing — Lumi captures the soft signals. You forward an email — Lumi updates the constraints. You open the app at 8am — the brief is already there, ready to feed Claude.
- Voice → structured CRM, automatically
- No forms. No data entry. No copy-paste.
- Free for agents in EU · LatAm · MENA
Lumi · Wednesday
Good morning, Niki.
Two showings · three leads need a nudge.
Showing · Passeig de Gràcia 84
Pipeline
Active
8
Warm
4
Cold
2
Clara Ruiz
Active€1.8M · 3BR
Passeig de Gràcia showing · 11:30
Andreas Moreno
Active€2.4M · 4BR
Send comps by 18:00
Dimitri Schneider
Warm€900K · 2BR
Contract review today
Silent 3d · last 3 days ago
Sarah Mitchell
Cold€1.2M · 3BR
Draft re-engagement
Silent 9d · last 9 days ago
A real-estate adaptation of the multi-tone content pattern from marketing — same content under-performs when published once instead of channel-tuned. Our slice: 3 listing copies (factual MLS / emotional luxury / first-time-buyer) generated from one Claude call with vision.
More guides like this on @lumi.estate. Follow if any of this was useful — it's how we know to keep writing.