Research post
The Language
of a Listing
What 38,913 MLS descriptions reveal about price, market time, concessions, and seller leverage
Almost everything a listing description tells you about sale performance is a description of the house, not an effect of the writing. Two exceptions survive: how much punctuation the agent used, and how much they shouted.
What the study measures
A language study with a narrow claim.
This is an observational study of public MLS remarks. It asks whether the words and structure in a description are associated with how a closed listing performed, after accounting for the property, geography, timing, agent, and office context visible in the export. It does not ask whether an agent can cause a better sale by inserting a phrase.
The analysis starts with 38,913 closed listings from January 2 through July 31, 2026. It examines nine outcomes, from close price relative to the original list through days in MLS, concessions, and price reductions. The sample is concentrated in the Front Range, but covers 59 counties.
Listing remarks are records of decisions made around a property. As-is can describe condition. Motivated seller can describe pressure that already exists. Price reduced can describe a change that has already happened. The findings are diagnostic before they are persuasive. The clearest discretionary signals concern writing style, especially punctuation and sentence structure.
Close / original list
96.64% mean
The primary gross price outcome.
Close / final list
98.84% mean
Negotiation relative to the last ask.
Net close / original list
95.51% mean
After seller concessions.
Actual days in MLS
23 days median
Market time, with a long right tail.
Any seller concession
61.57%
Share of closed listings.
Any price reduction
39.29%
Share of usable records.
Read every coefficient as a reported association. A reliable pattern can still be a label for the home, a response to market history, or a fragile result in a narrow group.
The raw pattern
What does the language of a listing reveal about performance? This study follows 38,913 closed listings, nine outcomes, and one period from January 2 through July 31, 2026. The sample spans 59 counties, with the largest concentration in Colorado's Front Range. The first pattern is persuasive, and wrong.
A typical remark contains 212 words, 11 sentences, and 19 words per sentence. It contains 1 exclamation mark, no question marks, and no all-caps tokens at the median. The source field caps near 2,000 characters, compressing the upper tail.
Mean close / original list, with median DOM underneath
The raw DOM bars flatten after adjustment. The price association remains at +0.239 points per 75.5 words, while DOM is -0.31%, q = 0.689.
Exclamation groups, raw versus adjusted
0 marks
97.02% price
17 days, 33.44% reduction
1 marks
96.76% price
21 days, 37.62% reduction
2 marks
96.34% price
28 days, 43.18% reduction
3+ marks
95.87% price
37 days, 49.98% reduction
Controls explain 30.8% of close-to-original variation, 31.4% of net variation, and 53.4% of log DOM. That is why the raw view cannot be trusted. The adjustment design is in the appendix.
What survives adjustment
The tone and structure signals are narrower than the raw gradients, but more useful. Exclamation marks are the most consistent adverse signal: -0.181 points on price, -0.200 on net, +4.50% DOM, +0.66 points concession probability, +2.25 points reduction probability, and +0.152 points reduction size per 2.08 additional marks. All caps move in the same direction at roughly half the size, per 9.11 tokens.
Sentence count is separate from length. It is associated with +2.46% DOM, +0.66 points concession probability, and +0.69 points reduction probability, while word count has no adjusted DOM association. Longer remarks carry a 1.03-point higher probability of any concession. Conditional on a concession, they show about 2.47% fewer concession dollars, a 0.035-point lower concession share, and a 0.106-point smaller reduction without a reliable change in reduction probability.
The length and concession trade
Any concession
+1.03 points
longer remarks
Concession dollars
-2.47%
conditional on a concession
Concession share
-0.035 points
conditional on a concession
Reduction size
-0.106 points
no reliable probability change
$-1,357
Arithmetic translation of a reported association at this list price. It is not a forecast. For context, +0.47 points at $750,000 is about $3,526. As-is is about -$18,262 and motivated seller about -$27,969 at the same price.
These are associations, not returns from inserting a phrase. Character count overlaps with word count and is not additive. A complete description can be useful because it records the home accurately, not because the coefficient proves that length creates value.
The words are diagnosing, not persuading
Phrases that look like strong performance drivers are often labels for the property and its history. As-is is associated with -2.435 points on close/original and -9.25% DOM at the same time, plus a 13.62-point lower concession probability. Speed and price move in opposite directions. Investor special is associated with a 32.51-point lower concession probability, fixer with -13.76 points, and sweat equity with -10.32 points. Move-in ready is +0.214 points and +3.65% DOM. Turnkey is +0.317 points with no DOM finding.
motivated seller
motivation, close/original -3.73, DOM +71.67
price reduced
motivation, close/original -3.32, DOM null
TLC
condition and distress, close/original -2.93, DOM null
as-is
condition and distress, close/original -2.44, DOM -9.25
sweat equity
condition and distress, close/original -1.98, DOM null
new HVAC
improvement, close/original +0.60, DOM null
remodeled
improvement, close/original +0.53, DOM null
updated
improvement, close/original +0.47, DOM null
multigenerational
audience, close/original -0.36, DOM +6.15
oasis
lifestyle, close/original +0.34, DOM -5.82
walkable
location, close/original +0.33, DOM -8.56
turnkey
improvement, close/original +0.32, DOM null
beautiful
praise, close/original +0.31, DOM null
move-in ready
improvement, close/original +0.21, DOM +3.65
schedule a showing
call to action, close/original -0.20, DOM null
investor special
condition and distress, close/original null, DOM null
fixer
condition and distress, close/original null, DOM null
new roof
improvement, close/original null, DOM null
minutes from
location, close/original null, DOM +4.57
don't miss
call to action, close/original null, DOM +3.79
see it today
call to action, close/original null, DOM null
won't last
call to action, close/original null, DOM -12.55
renovated
improvement, close/original null, DOM null
spectacular
praise, close/original null, DOM null
stunning
praise, close/original null, DOM null
gorgeous
praise, close/original null, DOM null
seller concession
incentive, close/original null, DOM null
rate buydown
incentive, close/original null, DOM null
closing-cost credit
incentive, close/original null, DOM null
financing incentive
incentive, close/original null, DOM null
priced to sell
motivation, close/original null, DOM null
retreat
lifestyle, close/original null, DOM null
entertaining
lifestyle, close/original null, DOM null
conveniently located
location, close/original null, DOM null
amazing
praise, close/original null, DOM null
close to
location, close/original null, DOM null
dream home
praise, close/original null, DOM null
first time buyer
audience, close/original null, DOM null
hurry
call to action, close/original null, DOM null
investor opportunity
audience, close/original null, DOM null
lock and leave
lifestyle, close/original null, DOM null
must see
call to action, close/original null, DOM null
one of a kind
praise, close/original null, DOM null
rare opportunity
praise, close/original null, DOM null
steps from
location, close/original null, DOM null
work from home
audience, close/original null, DOM null
Price-reduced language is associated with -3.316 points and +86.80% DOM. Motivated seller is -3.729 and +71.67% DOM. These phrases partly restate the outcome because the export captured final remarks rather than launch remarks. Location language also separates: walkable is +0.325 points and -8.56% DOM, oasis is +0.340 points and -5.82% DOM, minutes from is +4.57% DOM, while close to and steps from are null.
Concrete terms show a different pattern from generic filler. The strongest negative automatic terms include opportunity at -0.409 points, attached garage at -0.347, and property at -0.288. Positive terms are specific: beautifully +0.379, modern +0.316, fully +0.299, windows +0.290, covered +0.272, cabinetry +0.252, countertops +0.232, floors +0.240, and views +0.239.
114 terms reached significance in at least one outcome. These are exploratory because discovery and estimation used the same sample.
Only 5 of 300 selected terms reached significance for DOM, compared with 74 for close-to-original-list.
Template contamination is part of this story. 1,370 listings, 3.52% of the sample, share exactly duplicated remark text across 600 groups. Builder and agent templates can make generic terms appear significant without making them persuasive.
What the data does not support
Calls to action, selected outcome associations
Calls to action are mixed. Don't miss is associated with +3.79% DOM and +2.23 points concession probability. Schedule a showing is -0.198 points on close/final. See it today is +9.84 points on reduction probability. A null means nine tests failed to clear the threshold in this sample, not that a phrase never matters.
New roof has no price-ratio association on any of the three price outcomes, but carries a 2.72-point higher probability of a price reduction. New HVAC carries the largest improvement coefficient at +0.600 points, on 1.0% prevalence. Multigenerational is -0.365 points, +6.15% DOM, and +2.72 points reduction probability. Won't last is -12.55% DOM on 408 listings with no price outcome surviving correction. It is a flagged fragile result.
What to do with this
Write with the narrow finding in view. These are the five actions the study supports:
- 1. Write complete descriptions. The price association holds at +0.239 points per 75.5 words, though the speed advantage does not survive adjustment.
- 2. Cut exclamation marks. They are the most consistent adverse discretionary signal in the study.
- 3. Cut all caps. They move in the same direction, with smaller magnitude.
- 4. Write fuller sentences rather than more of them. Sentence count is associated with +2.46% DOM, +0.66 points concession probability, and +0.69 points reduction probability.
- 5. Use condition and feature language accurately, because it describes the home, not because the coefficient is a return on the word.
The next study should use immutable launch-time remark snapshots, include canceled and expired listings, pre-register the phrase set, and validate findings in a later market period. That design would distinguish words chosen before performance from words added because of performance.
Methodology appendix
Question and scope. The study tests observational associations between public remarks and nine outcomes for closed listings.
Construction. Records without at least 20 usable remark characters were excluded. Outcome-specific plausibility filters preserve a listing for other outcomes when one field is invalid.
Cross-fitting. Five folds train adjustment models on four folds and predict the held-out fold. Both outcome and language variables are residualized against property, geography, timing, agent, and office controls.
Inference. Standard errors are clustered by listing agent. Benjamini-Hochberg false-discovery-rate adjustment is applied within each outcome and result family.
Threats to validity. Final remarks may differ from launch remarks. The sample excludes listings that did not close. Property quality, staging, photography, buyer competition, financing conditions, and edit timing are not fully observed. One-at-a-time estimates overlap, templates create dependence, rare phrases can be unstable, and the period is a single market window. Conditional concession outcomes exclude transactions without positive concessions. Flesch reading ease is approximate. No causal identification is claimed.
Adjustment model performance
The performance metrics remove some observable listing heterogeneity. They do not prove that all confounding has been removed.
REcolorado® MLS data; analysis and interpretation by Cooper Thayer. Based on information from REcolorado®, Inc. for the period January 2 through July 31, 2026. Not all properties were listed and/or closed by Company. This representation is based in whole or in part on content supplied by REcolorado®, Inc. REcolorado®, Inc. does not guarantee nor is it in any way responsible for its accuracy. Content maintained by REcolorado®, Inc. may not reflect all real estate activity in the market. For non-commercial use only.