Grounded AI for Home Search: What 48,315 Arizona Homes Show About Local Logic Data

Max Leblond

17 Nov 2026

Grounded AI for Home Search: what 48,315 Arizona homes show about Local Logic data as an AI input, a Local Logic and HouseQuest customer story

A Local Logic customer story with HouseQuest. November 2026.

Listings describe the house, not the street. Most AI tools fill that gap with plausible but unsourced language, so the location half of a home match is guessed. HouseQuest’s HiQ grounds it in Local Logic neighborhood data. This paper reports what that looks like across 48,315 Arizona homes.

The study at a glance: 48,315 Arizona MLS homes scored with Local Logic data, 15 of 16 Local Logic scores linked to a HiQ trait, and a highest lift in model confidence of 11.7%

At a glance

  • The challenge. Listings describe the house, not the street. Most AI tools fill that gap with plausible but unsourced language, so the location half of a match is guessed.
  • The solution. HiQ’s matching intelligence engine scores every home on 14 traits from the MLS listing plus 16 Local Logic neighborhood scores and their plain-language explanations, requested through Local Logic’s APIs.
  • The results. 15 of 16 Local Logic scores linked to a HiQ trait. Model confidence rose 10.0% to 11.7% on four location-sensitive traits, and trait scores moved both up and down with the neighborhood.
Detail
Customer HouseQuest
Product HiQ, matching intelligence engine
Market tested Arizona Regional MLS
Local Logic data 16 location scores and neighborhood profiles, via API
Scoring model OpenAI o3

“Where you live has a profound impact on your life, and that includes both your home and the neighborhood around it. That’s why we partnered with Local Logic to help people find homes and neighborhoods where they can truly thrive. By combining psychometric science, Local Logic’s neighborhood data and advanced location intelligence, we’ve created a first-of-its-kind home-matching technology.”
Randy Marsden, CEO and Founder, HouseQuest

Summary

Any engine that matches people to homes has to reason about the part of a home a listing can’t describe: the street, the noise, the walk to coffee, the trail at the end of the road. Most AI tools fill that gap with plausible but unsourced language. HouseQuest’s HiQ fills it with Local Logic neighborhood scores.

This paper reports two analyses. The first examines how 16 Local Logic scores relate to HiQ’s 14 home traits across 48,315 Arizona listings. The second rescored 500 current listings three times to separate the effect of Local Logic data from the model’s normal run-to-run variation.

  • 48,315 Arizona MLS homes scored using Local Logic neighborhood data
  • 15 of 16 Local Logic scores had an absolute correlation of at least 0.20 with at least one HiQ trait
  • +10.0% to 11.7% higher model confidence with Local Logic data for four location-sensitive traits, across 500 current Arizona homes

Three traits rose. Community fell.

Local Logic data also changed the scores themselves. Mean trait scores for Retreat, Nature, Community and Safety moved by 8.4% to 17.6%: Retreat, Nature and Safety rose, while Community fell. The data did not simply push every score upward. It raised some traits and lowered others, depending on what each neighborhood is like.

HiQ trait Mean trait score shift with Local Logic data
Retreat +17.6%
Nature +8.6%
Community −16.8%
Safety +8.4%

500 current Arizona listings, scored with and without Local Logic data.

The test does not establish that the resulting scores are more accurate or that buyers make better decisions. That comparison comes next.

1. The challenge: AI can’t see the street

A listing describes a house: bedrooms, square footage, a pool, a renovated kitchen. It says almost nothing about the life around it. Yet that is often what decides whether a home fits. A buyer who wants quiet mornings and trails and a buyer who wants to walk to dinner can read the same listing and need opposite answers.

Language models are good at producing confident descriptions of places. A model can infer neighborhood character from an address, school names and general patterns, but those inferences aren’t facts about the specific location. Structured neighborhood data gives the model address-level facts and explanations it can use.

For a matching engine, the gap is larger. Matching depends on comparing homes across many dimensions. If the location half of that comparison is guessed, the match is only half grounded.

Inside the four walls Outside the four walls
What the listing describes What decides whether it fits
Beds, baths, square footage, pool, casita, single story Street noise, walk to coffee, trails and parks, nightlife, nearby schools
Structured, sourced, already in the MLS record Guessed by most AI tools. Grounded by Local Logic.

Key takeaway: if the location half of a comparison is guessed, the match is only half grounded.

2. The data: Local Logic scores as an AI input

Local Logic quantifies what happens outside the four walls of a home. Four properties make the data usable by an AI model:

  • One scale everywhere. Every score uses the same scale, wherever the home is. A model can compare a bluff in northern Arizona with a block in downtown Phoenix without translating between sources or units. This paper shows all scores from 0 to 10.
  • A reason with every number. Each score comes with a plain-language explanation, such as “virtually no sources of noise nearby.” A model can pass that explanation to the user, so its reasoning cites the data rather than paraphrasing it.
  • Consistent coverage. 18 location scores across more than 250 million addresses in the US and Canada, built from over 100 billion data points across transportation, amenities, education, character, nature and wellness.
  • Built to plug in. Product teams access the data through Local Logic’s APIs, and AI agents can query it through the Local Logic MCP server. HouseQuest requests scores and neighborhood profiles through the APIs each time it scores a listing.

What the data describes, and what it doesn’t

Local Logic data describes a place: how quiet it is, how walkable it is, and what is nearby. It does not judge whether a neighborhood is good or bad, and it does not suggest who should live there. Local Logic does not provide crime data or demographic data in its scores.

HiQ’s 14 traits, including Safety and Kids, are HouseQuest’s own. When Local Logic data informs them, it does so through those same descriptions, such as street noise or how close the nearest schools are. Local Logic’s school scores describe proximity only. They don’t rate schools.

The 16 Local Logic scores in HouseQuest’s scorer

Category Scores
Transportation Pedestrian friendly, cycling friendly, transit friendly, car friendly
Amenities Groceries, shopping, cafés, restaurants, nightlife
Education High schools, primary schools, daycares
Character Quiet, vibrant
Nature and wellness Parks, wellness

3. The application: how HiQ uses the data

HiQ is HouseQuest’s home-matching product. A buyer’s questionnaire produces a 14-trait preference profile. HiQ compares that full profile with each home’s 14 trait scores, then applies the buyer’s stated property and location requirements. Persona names summarize a profile for display; they do not calculate the match score.

Four steps: Local Logic neighborhood input, MLS listing input, HouseQuest trait scoring and HiQ buyer match

Inside the scorer: one request in, three answers per trait out

The current scorer uses OpenAI’s o3 model. Each request contains the scoring instructions, the MLS listing and the Local Logic neighborhood data. For each trait, the model returns a score, a reason and a confidence rating from 0 to 100. The confidence rating is the model’s own estimate of how well the supplied information supports its score. HouseQuest records that rating without calibration.

Diagram: scoring instructions, the MLS listing and Local Logic neighborhood data go into the o3 model, which returns a score, a reason and a confidence rating for each of 14 traits

The Local Logic input, as a HiQ buyer sees it (Desert View, Phoenix). The neighborhood panel shows Local Logic’s 0 to 5 values with their explanations and attribution. This paper doubles them to a 0 to 10 scale.

HiQ neighborhood panel for Desert View, Phoenix, showing Local Logic scores out of 5 with plain-language explanations and Local Logic attribution

From listing to match: one Cave Creek home in HiQ

A buyer opens a listing, sees the neighborhood through Local Logic data, then sees how the home’s 14 traits line up with their own profile, with a written reason behind each score.

A. The listing. MLS facts and an overall match score for this buyer.

HiQ listing view of a Cave Creek, Arizona home showing an 80% match, MLS facts and HouseQuest's reasons the home fits the buyer

B. The match. The buyer’s profile (blue) against the home (green) across all 14 traits.

HiQ match graph comparing the buyer's profile with the home across 14 traits, with the reason behind the home's Retreat score of 8/10

The reason behind Retreat 8/10: “The walled front courtyard, single-story layout, and backyard that opens directly to a 20-acre desert preserve in a neighborhood rated 5/5 for quiet make the home feel secluded and restorative.” The reason cites the Local Logic quiet score (5/5 on the API scale, 10/10 on this paper’s scale).

4. Method

HouseQuest’s matching intelligence engine does not use MLS data to train AI models; it scores each home on 14 traits and correlates those scores with each buyer’s personality profile.

Analysis Listings
Arizona MLS listings scored with Local Logic data (July 14 to November 6, 2026) 48,315
With all 16 scores: correlation analysis 25,191
With a price and a quiet score: quiet and price analysis 48,313

Large-sample dataset

The analysis includes 48,315 Arizona Regional MLS listings scored with the 14-trait version of HiQ between July 14 and November 6, 2026. Listings without stored Local Logic data were excluded. Some included listings have since gone under contract or sold.

Correlation analysis

HouseQuest calculated the correlation between each Local Logic score and each HiQ trait, using the 25,191 listings with a complete set of the 16 scores. The results table reports the strongest absolute correlation for each score. An absolute correlation of 0.20 or more is described as a link. This is an association, not proof that one score caused the other.

Quiet and price analysis

HouseQuest divided listings into four price groups and compared average Retreat scores at the lowest Local Logic quiet value, 0, with scores at the highest values, 9 or 10. The low-quiet groups contain 119 to 176 listings per price group. The high-quiet groups contain 8,727 to 9,550, so the comparison is descriptive rather than a balanced experiment.

Current comparison

HouseQuest drew 500 homes at random from 26,371 active Arizona listings, spanning 77 cities and 11 counties. The production scorer evaluated each home without Local Logic data and twice with it. Every call used the same o3 model, 14 traits, 1 to 10 scale and MLS record; only the Local Logic neighborhood section changed.

Run Setup
Run 1: Baseline o3 model, 14 traits on a 1 to 10 scale, same MLS record, no Local Logic section
Run 2: With Local Logic o3 model, 14 traits on a 1 to 10 scale, same MLS record, Local Logic neighborhood section
Run 3: With Local Logic, repeated o3 model, 14 traits on a 1 to 10 scale, same MLS record, Local Logic neighborhood section

An earlier test pointed the same way

An earlier test, in March 2026, scored 486 Arizona homes once without Local Logic data and once with it. The model’s average confidence across all its traits rose from 55 to 63 out of 100, about 14% higher. The November comparison is the one reported in detail in this paper.

Scale

Local Logic’s API returns scores from 0 to 5. This paper doubles them for display on a 0 to 10 scale. The conversion changes only how the number is shown.

5. Results

Local Logic scores align most strongly with related HiQ traits

Among the 25,191 complete records, 15 of the 16 Local Logic scores had an absolute correlation of at least 0.20 with one or more HiQ traits. Quiet had the strongest relationship, with Retreat at 0.60. Vibrant, nightlife, pedestrian access and cafés had their strongest relationships with Community. Car friendly was below the threshold at 0.17. These are correlations, not estimates of how much Local Logic caused a trait to change.

Bar chart of the strongest correlation between each Local Logic score and a HiQ trait, led by quiet and Retreat at 0.60; 15 of 16 scores pass the 0.20 link threshold

What it means: quiet goes with Retreat, walkable amenities with Community, parks with Nature and nearby schools with Kids.

Quiet streets score about 50% higher for Retreat, at every price

In each price group, listings with a Local Logic quiet score of 9 or 10 had a higher average Retreat score than listings with a quiet score of 0. Quiet is one of the inputs used to score Retreat, so this pattern is expected. The analysis shows that the relationship appears in every price group; it does not prove that quiet alone caused the difference.

Dot chart: in every price group, homes on the quietest streets score about 50% higher for Retreat than homes on the loudest streets

What it means: Retreat follows the street, not just the price tag. The gap between the loudest and quietest streets holds at about 50% from the lowest price group to the highest.

With Local Logic, the engine grew more confident, and moved scores both ways

The largest confidence shifts were Retreat at 11.0%, Nature at 11.7%, Community at 10.3% and Safety at 10.0%. Their corresponding mean trait-score shifts were Retreat +17.6%, Nature +8.6%, Community −16.8% and Safety +8.4%. These percentages divide each mean shift by that trait’s mean without Local Logic, so traits with different starting values can be compared.

Bar chart: with Local Logic data, model confidence rose 11.0% for Retreat, 11.7% for Nature, 10.3% for Community and 10.0% for Safety; mean trait scores changed +17.6%, +8.6%, -16.8% and +8.4%

What it means: the direction matters. Local Logic did not simply raise every score; it made the model more certain while moving each trait toward what the neighborhood data describes.

Safety is a HiQ trait, not a Local Logic score. Local Logic doesn’t provide crime data. In this test, the model had more descriptive information about the neighborhood, such as noise levels and what is nearby, and it reported more confidence in its Safety score as a result.

Kids shows why score and confidence differ. Overall Kids confidence barely changed, but a high primary school score, meaning schools close by, raised the Kids trait score, while a low one lowered it.

6. Two homes, side by side: the engine’s reasons point back to the street

Both examples come from the 48,315-listing analysis. The excerpts are taken from the stored engine reasons. An ellipsis marks omitted words. Numbers in square brackets convert Local Logic’s API values to the 0 to 10 scale used in this paper.

Two Arizona homes compared: a quiet bluff home scoring Retreat 9 with a Local Logic quiet score of 10/10, and a walkable central Phoenix home scoring Community 7 with nightlife, restaurants and groceries at 9/10

The engine’s reason for Retreat (Home 1): “Two private acres on a bluff … with natural buffers, a [10/10] neighborhood quiet score, and a wrap-around deck all signal strong seclusion and sanctuary qualities.”

The engine’s reason for Community (Home 2): “Walkability scores of [8/10 to 9/10] for groceries, cafés, restaurants and nightlife mean social venues cluster within a short stroll …”

In both examples, the engine’s explanation names the Local Logic inputs that support its neighborhood description. These examples show traceability. They do not test whether the description is accurate across all homes.

7. Limitations

What this test shows What it does not yet show
Local Logic scores move with related HiQ traits That the trait scores are more accurate
The model reports more confidence with Local Logic data That buyers make better decisions
Trait scores move in both directions, not only up That one score caused another
The engine’s reasons cite the Local Logic inputs they used That results hold beyond Arizona and one model
  1. Self-reported confidence is not accuracy. The current test shows that o3 reports greater certainty when it receives Local Logic data. It does not show that the trait scores are closer to the truth or more useful to buyers. Adding information can make a model more confident without making its answer more correct, so accuracy requires an independent reference or buyer outcomes.
  2. No outcome validation yet. Results have not been checked against which homes buyers chose, toured or bought.
  3. Correlation is not causation. The correlation analysis shows which Local Logic scores move with which traits. Some links are built in, since Local Logic scores are direct inputs to the traits.
  4. Model variation. Repeated runs showed normal model variation, so HouseQuest used two runs with Local Logic and reports the stable trait-level patterns rather than individual home-level changes.
  5. One market, one model. The current comparison covers 500 randomly selected Arizona homes, one product and one model. The sample spans 77 cities and 11 counties but was not balanced across every Local Logic score, price band or property type, so these percentages are an initial estimate.

8. What this means for builders

  1. Give the model normalized scores, not raw counts. A score on a fixed scale is something a model can compare across neighborhoods. A count of cafés within 500 meters is not.
  2. Pass the explanation with the number. When the model receives “virtually no sources of noise nearby” alongside the score, it can cite it rather than invent its own description.
  3. Make every reason traceable. Each claim an AI makes about a place should point back to an input a user can check.
  4. Validate against outcomes, not just confidence. Confidence is a useful signal during development. Buyer behavior is the real test.

9. Next steps

HouseQuest plans to compare each buyer’s match scores with the homes that buyer saves and asks an agent to show. Tour data will depend on agent reports, and purchase data will come from closed sales in the MLS.

Four buyer signals HouseQuest will compare with match scores: saves and showing requests from HiQ, tours from agent reports, and purchases from MLS closed sales

Frequently asked questions

What is HouseQuest’s HiQ?

HiQ is HouseQuest’s home-matching product. A buyer’s questionnaire produces a 14-trait preference profile, and HiQ compares it with each home’s 14 trait scores, built from the MLS listing and Local Logic neighborhood data.

How does HiQ use Local Logic data?

HouseQuest requests 16 Local Logic location scores and neighborhood profiles through Local Logic’s APIs each time it scores a listing. Its engine, built on OpenAI’s o3 model, reads those scores and their plain-language explanations alongside the MLS listing to score each home on 14 traits.

What changed when the engine had Local Logic data?

Across 500 current Arizona homes, model confidence rose 10.0% to 11.7% on four location-sensitive traits, and trait scores moved in both directions: Retreat rose 17.6% while Community fell 16.8%.

Does higher confidence mean the scores are more accurate?

No. Confidence is the model’s own rating of how well its inputs support a score. HouseQuest plans to validate matches against buyer behavior, such as saves, showing requests, tours and purchases.

Does Local Logic provide crime or demographic data?

No. Local Logic data describes a place, such as how quiet it is, how walkable it is and what is nearby. It does not include crime or demographic data in its scores, and its school scores describe proximity only.

How was the analysis conducted?

HouseQuest analyzed 48,315 Arizona MLS listings scored between July 14 and November 6, 2026, and rescored 500 randomly selected active listings once without Local Logic data and twice with it, changing only the Local Logic section.

See it in action: neighborhood data your AI can cite

Product teams access Local Logic’s location scores through its APIs, and AI agents can query them through the Local Logic MCP server.

Ground your AI in Local Logic data | Try HiQ

About Local Logic

Local Logic is a location intelligence company that quantifies everything outside the four walls of a home. Its data runs on more than 13,000 real estate websites, reaches 50 million users a month, and covers 18 location scores across more than 250 million addresses in the US and Canada.

About HouseQuest

Where you live shapes your happiness. It matters. HouseQuest is reimagining home discovery by combining AI and neuroscience to match buyers with homes that resonate with their personalities and lifestyles. HouseQuest also provides a spatial computing app that turns a look around the neighborhood into an interactive discovery experience, revealing property insights through the phone’s camera. Together, these innovative tools make finding a home personal, engaging, and fun. HouseQuest was founded by serial entrepreneur and Swype co-founder Randy Marsden. Learn more at housequest.com.

Download Case Study