What 27,142 Scanned Homes Reveal About Distressed Property Signals

Most published numbers about distressed housing come from records: tax rolls, foreclosure filings, vacancy surveys. Those tell you about a property’s paperwork, not its condition. We wanted to know what you actually see when you look at every house in a county, one at a time, and score what the picture shows. So we did that for eight counties. This is what the imagery said.

Quick answer: We scored 27,142 homes across eight rural counties in Tennessee and Georgia using street-level imagery and a vision model. About 5 percent showed clear visible distress. More than one in four residential parcels could not be assessed at all, because no street-level imagery of them existed.

Ugly House Finder research, published July 2026. Imagery scored April to May 2026.

How common are visibly distressed homes?

Across all eight counties, 1,364 of 27,142 scored homes (5.0 percent) scored 3.5 or above on a 0 to 5 distress scale. That is the threshold where damage is clear and unambiguous rather than a matter of judgment. A much larger group sat in the middle, showing real but milder neglect.

Clear distress means a distress score of 3.5 or higher. Per-county scored counts reflect each county’s street-level imagery coverage rate.

Which signs of distress show up most often?

The vision model records specific named indicators rather than one opaque score, so we can count how often each one appears. These figures come from Carter County, Tennessee, the county in this set scored under the full indicator schema, across all 17,924 scored homes.

Two things stand out. An overgrown yard is so common that on its own it means almost nothing, which is exactly why single-signal lead lists produce so much noise. The signals that actually separate a distressed house from a merely untidy one are the structural ones, and those are rare: broken windows on 3.3 percent of homes, visible structural damage on 10.3 percent, fire damage on one home in a thousand.

Severity matters as much as presence. Of the 3,420 homes with roof damage, 1,098 were scored at moderate or severe rather than minor. Of the 789 homes that read as abandoned, 656 were at that higher severity, which fits, because a house rarely reads as abandoned in a subtle way.

Does absentee ownership predict distress?

Yes, and the size of the effect is worth being precise about, because it tends to get overstated. In Carter County, homes whose owner does not live at the property scored a mean distress of 2.52 against 2.39 for owner-occupied homes. That is a real gap but a small one in absolute terms.

The difference is clearer at the tier level. Among absentee-owned homes, 56.6 percent scored at 2.5 or above, against 44.9 percent of owner-occupied homes. That is a gap of 11.7 percentage points, measured across 5,088 absentee and 12,835 owner-occupied properties.

The practical reading is that absentee ownership is a useful filter and a poor standalone signal. It shifts the odds by roughly a quarter in relative terms. It does not tell you the house is in bad shape.

Does a home’s age predict distress?

More cleanly than ownership does. Sorting Carter County’s scored homes by decade built produces a straight monotonic decline. The older the housing stock, the higher the share showing visible distress, with no reversal at any step.

A pre-1940 home is roughly 1.6 times as likely to show visible distress as one built since 2000. For anyone deciding where to spend time in an unfamiliar county, the age of the housing stock is a better first cut than the ownership record.

City-level companion scans

The eight counties above are rural. We have since run the same random-sample method against three large cities, where the measured rates run far higher: Baltimore at 44.9 percent, Philadelphia at 43.0 percent, and Detroit at 37.4 percent of homes showing some visible neglect. We have also measured five whole metro counties, which run lower because they include suburbs: Harris County (Houston) at 22.9 percent, Hillsborough County (Tampa) at 20.8 percent, Milwaukee County at 19.5 percent, Cook County (Chicago) at 18.6 percent, and Cuyahoga County (Cleveland) at 9.3 percent.

Where remote assessment fails

This is the finding we did not expect, and it is the one that most affects how anyone should read condition data about rural markets.

Of 37,391 residential parcels across the eight counties, 10,242 had no usable street-level imagery. That is 27.4 percent, and it is not evenly spread. Coverage ranged from 79.4 percent in Calhoun County, Georgia down to 44.8 percent in Quitman County, where more than half the homes have never been photographed from the road.

The uncomfortable part is the direction of the relationship. Across the seven Georgia counties, imagery coverage correlates negatively with measured distress, at r equals negative 0.74. The counties with the least coverage are the counties where the homes we could see were in the worst shape.

With only seven counties this is suggestive rather than conclusive, and we are not going to claim more than that. But the plain implication is that remote condition assessment probably undercounts distress in exactly the places that have the most of it, because the least-photographed roads and the most-neglected housing tend to be the same rural back roads. Any number derived from street-level imagery alone, including ours, should be read as a floor rather than a full count.

How we did this

• Sample. Every residential parcel in eight counties: Carter County, Tennessee, and Baker, Calhoun, Clay, Early, Miller, Quitman, and Randolph Counties, Georgia. This is a full census of those counties, not a sample within them.
• This is not a national sample. All eight are rural counties in the Southeast, and they were selected because they were already being scanned for other reasons, not at random. The 5 percent figure describes these counties. It should not be read as a national distressed-housing rate.
• Method. For each parcel we retrieved street-level imagery aimed at the structure and scored it with a vision-language model against a fixed distress taxonomy derived from our own hand-labeled training set. The model returns named indicators with a severity rating, plus a distress score from 0 to 5.
• Thresholds. A score of 3.5 or above is treated as clear distress. Between 2.5 and 3.5 indicates visible but milder neglect. Below 2.5 is unremarkable.
• Engines. The two county groups were scored by two different vision models, and neither is the model we run in production today. Cross-group comparisons should be treated as approximate for that reason. Within-county findings, which is where all the indicator and correlation analysis sits, are unaffected.
• Dates. Imagery was scored in April and May 2026. Street-level imagery is itself often one to three years old, so these figures describe recent condition, not this morning’s.
• What is not in here. This corpus contains no aerial-imagery scoring and no tax-delinquency data. Those signals exist in our platform but were not enabled for these runs, so nothing here should be read as incorporating them.
• Known error mode. Purely visual scoring over-calls distress in heavily forested and affluent areas at a rate of roughly one in six, because dark cedar and A-frame homes can read as fire-damaged, and neighboring structures or tarps get misattributed. We corroborate visual-only findings against a second signal before treating them as leads.

Common questions

What counts as a distressed property?

Here it means visible physical distress, meaning damage or neglect you could see standing in the street. That is deliberately narrower than the common usage, which often folds in financial distress like tax delinquency or foreclosure. A house can be financially distressed and look immaculate, and it can be falling apart with every bill paid on time.

How accurate is scoring a house from a photograph?

Good enough to rank, not good enough to trust blindly. The model reliably separates a house with a collapsed roof from one with a tidy lawn. It struggles at the margins, and it has a known bias toward over-calling in forested and affluent areas at roughly one in six. That is why the output is a ranked list to work through rather than a verdict on any single address.

Why do so many rural properties have no imagery?

Street-level imagery is captured by vehicles driving public roads. Properties on private drives, unpaved back roads, or long rural setbacks often were never photographed, and some roads have not been revisited in years. In the most rural county we examined, this affected more than half of all homes.

Does this mean 5 percent of American homes are distressed?

No. These are eight rural Southeastern counties chosen for operational reasons, not a random national sample. Urban housing stock, coastal markets, and newer suburban development would all read differently. We would need a broader and properly randomized sample to say anything national, and we have not run one.

Can I get this kind of scan for my own county?

Yes. The same pipeline runs against any county where parcel data is available, which currently covers roughly 55 to 60 million residential parcels across 37 states. You can run a neighborhood scan yourself in the app, or request a full county scan.

Scan your own market

Draw an area on the map and see what the imagery says about it, or request a full county scan. Start scanning, or see plans and pricing.

All figures in this report are computed directly from our scored property records and can be regenerated from source. Counts reflect the corpus as of May 2026.