Does local wealth change who AI recommends? Income, home prices and AI answers in 50 markets

Recasi team · Oct 12, 2026 · 8 min read

We compared income, home prices, and other Census measures with 1,000 AI answers across 50 markets. In the highest- and lowest-income thirds, answers named an agent or team at nearly the same rate. Home prices produced the same broad result: we found no clear link. This article describes the comparisons, their limits, and the market-size pattern that did hold up.

The results do not show that local wealth has no effect on every answer. They show no clear relationship in this particular set of markets and questions. Market size was different. Larger cities produced less agreement between the four tools and more citations from smaller agent-directory and rating sites.

How we ran the comparison

We reused the 1,000 answers from our 50-market AI real estate agent study. The answers came from ChatGPT, Gemini, Perplexity, and Google's AI Mode on October 11 and October 12, 2026. We did not ask new AI questions for this follow-up. This analysis therefore examines the same answer set from a different angle.

For each city, we used public Census figures from the American Community Survey 5-year estimates for 2020–2024. We matched all 50 markets to a Census city, called a “place.” None required a county fallback. We examined per capita income, median household income, median home value, population, owner-occupied share, and home broadband share.

We ranked the 50 markets on each measure and checked whether an AI-answer measure rose or fell with it. This is a rank correlation from -1 to 1. A result near zero indicates no steady rise or fall. We also estimated a 95% range by resampling the markets 2,000 times.

Finally, we compared the top third of markets, 17 places, with the bottom third, also 17 places. We ran 78 comparisons in total. Some results can appear important by chance when that many checks are made. We therefore corrected for chance hits and call a pattern “clear” only when it survived that correction.

Income showed no clear link

Per capita income did not change how often answers named an agent or team. In the top-income third, 87.9% of answers named someone. In the bottom-income third, 89.1% did. The correlation was -0.03, with a 95% range from -0.32 to 0.26, which is close to no pattern.

Median household income produced a similar result. The top-income third had 88.2% of answers naming an agent or team, compared with 90.0% in the bottom third. The correlation was 0.00. Higher household income did not make a market more likely to receive named people in these answers.

The agent or team-site measure showed a small difference. Those sites appeared in 72.6% of answers in the top-income third and 69.1% in the bottom third. The correlation was 0.14, with a range from -0.14 to 0.40. That difference could readily be chance, so it is not a clear result.

The number of names per answer also did not support an income pattern. The top-income third averaged 4.4 names, and the bottom third averaged 4.7. An agent in a higher-income market should not infer from this study that income alone changes whether a tool names them. Nor does it show that income changes how many names a response provides.

Home prices showed no clear link

Median home value produced the same broad result. In the higher-home-value third, 90.0% of answers named an agent or team. In the lower-home-value third, 89.7% did. The correlation was 0.03, which is very close to no relationship.

Agent or team sites appeared in 72.6% of answers in the higher-home-value third and 68.2% in the lower third. The correlation was 0.22, with a range from -0.06 to 0.46. It leans upward, but it was not clear after the chance correction. This is a difference worth treating cautiously rather than as a market rule.

Zillow appeared in 57.9% of answers in both thirds. That exact match illustrates the result plainly. A higher-priced market was not automatically a different kind of AI search market in this data. Local prices may affect a business in other ways, but they did not explain the answer patterns we measured.

Market size was the clear pattern

Population produced a different result. In the largest third of markets, the average overlap between two tools' named lists was 5.4%. In the smallest third, it was 9.4%. The correlation was -0.43, with a range from -0.66 to -0.14, and it survived the correction for chance hits.

The tools therefore agreed less in larger markets. That does not mean every answer in a large city will disagree. It also does not mean smaller markets have one fixed list of agents. It means the overall pattern moved in that direction across the markets studied.

Smaller agent-directory and rating sites, beyond Zillow, Realtor.com, Homes.com, and the two biggest agent-matching sites, appeared in 39.4% of answers in the largest third. They appeared in 17.6% of answers in the smallest third. The correlation was 0.47, with a range from 0.21 to 0.68. This pattern also survived the correction.

Other source patterns were less certain. Realtor.com appeared in 14.1% of large-market answers and 24.4% of small-market answers, with a correlation of -0.41. That pattern was likely, but not certain, after the correction. Zillow appeared in 50.9% of large-market answers and 62.4% of small-market answers, with a correlation of -0.32; it did not survive the correction.

The share of answers naming someone did not change with size. It was 89.1% in both the largest and smallest thirds. Size changed agreement and some sources, not whether an answer named an agent or team at all. That distinction keeps the finding within what the data supports.

Other factors did not explain the answers

We also checked owner-occupied share and home broadband share. Neither showed a clear link with any AI-answer measure after the chance correction. Both may affect a local market in other ways. They did not explain the patterns in this set of answers.

What we think the results mean

This interpretation is ours; the study does not prove it. AI answers draw on what is published about agents, including websites, profiles, reviews, and local coverage. Every market in our sample had enough public material for the tools to name people. That may help explain why income and home values did not separate the answers.

Market size may change the number of competing agents and available sources. That interpretation fits the lower agreement and broader mix of smaller directory sites in larger cities. The study cannot show why a tool selected a particular name. It also cannot establish that market size caused the observed difference.

We would not use income or home value as a shortcut for judging AI visibility. A less wealthy market can have a clear public record, while a wealthier market can have a thin or confusing one. The practical work remains the same: make your own information specific, accurate, and easy to check. That is more actionable than a broad assumption about the market.

What this means for agents

In our data, market income and home prices did not decide whether AI named an agent. Published information about you is the part you can improve. In a larger market, tools agreed less, so we recommend checking more than one tool and more than one relevant question. That gives a stronger view than a single result.

Recasi Local is AI search marketing for real estate agents and teams. We make your name, areas, and record say the same thing everywhere online. We also write pages that answer local buyer and seller questions and earn mentions on other websites. We work with one agent or team in each agreed market.

Learn more about AI SEO for real estate agents or review pricing. The governing aim is a public record that helps people understand your work, wherever their research begins. It is not a promise that a particular tool will name you. The evidence should guide the recommendation.

Published information about you is the part you can improve.

A careful conclusion

Local wealth was not a useful shortcut in this 50-market analysis. Market size better explained disagreement between tools and the mix of smaller directory sources. Treat that as a pattern to investigate, not a rule about any single city.

Frequently asked questions

Sources

• U.S. Census Bureau, American Community Survey 5-year estimates (2020–2024)
• Table B19301: Per capita income
• Table B19013: Median household income
• Table B25077: Median home value
• Table B01003: Total population
• Table B25003: Tenure (owner-occupied homes)
• Table B28002: Internet subscriptions in household
• Recasi 50-market AI real estate agent study