When a home buyer raises their hand on a builder's website at 9 p.m., what happens next? We measured 1,058 Sun Belt homebuilder websites from the outside, with no fake leads, and asked three AI assistants which builders they recommend.
Figures refer to regional builders with two or more selling communities unless stated. Brackets are 95% confidence intervals.
Two in three offer a buyer no way to get an answer at 9 p.m.: no chat, no AI assistant, no text line and no online tour booking. Correcting for chat widgets our detector misses, the share is still at least 56%. Among the public nationals it is 42%.
National builders are ahead on instant response (+25 points), ad measurement (+45), lot maps (+27) and published prices (+27). Their homepages score lower on Google's lab speed test (median 37 vs 49), and real-visitor load times are about the same.
About one in six regional builders runs any live chat, mostly general-purpose tools. Builder-specific AI assistants appear on fewer than 3 in 100 sites.
Asked for the best new home builders in 20 Sun Belt metros, ChatGPT, Claude and Gemini named about half of the regional builders at least once; only 28% by all three. Builders with visible ad measurement and lot maps were named about three times as often, after controlling for size and state. This is an association, not proof of cause.
Most builders do publish prices on their community pages, and most builders running Google ads do carry an Ads tag. The gap is not a missing tag. Whether that tag counts the right thing cannot be seen from outside.
Share of regional builders with 2+ communities (n = 293).
Source: Enroke Research crawl of builder homepages and one community page per builder, September 27, 2026. Chat recall on the hold-out validation was 0.60 (precision 1.00), so chat is undercounted.
31% of these builders collect a lead through a form but give the buyer no immediate path to a response. Published research on lead response found the odds of qualifying a lead fell 21-fold when response time went from 5 to 30 minutes (Lead Response Management Study, InsideSales.com and MIT).
Share of builders with each capability. Regional n = 293; nationals n = 19.
| Capability | Regional | Nationals | Gap |
|---|---|---|---|
| Instant response path | 32.8% | 57.9% | +25 |
| Live chat or AI assistant | 17.4% | 47.4% | +30 |
| Google Ads measurement | 44.7% | 89.5% | +45 |
| Price on community page | 57.3% | 84.2% | +27 |
| Lot map on community page | 36.2% | 63.2% | +27 |
| Lead form on community page | 44.1% | 73.7% | +30 |
| Self-guided tour booking | 2.7% | 15.8% | +13 |
| Google Analytics 4 | 79.9% | 84.2% | +4 |
Fisher's exact test, Holm-corrected: instant response +25 points (p = .04); ad measurement +45 points (p < .001). With 19 nationals, their intervals are wide.
ChatGPT, Claude and Gemini, each with live web search, answered three buyer questions for each of 20 Sun Belt metros (180 answers). Regional builders with 2+ communities in those metros: n = 160.
| Assistant | Builders named |
|---|---|
| Any of the three | 52.5% |
| ChatGPT | 46.2% |
| Claude | 36.9% |
| Gemini | 36.9% |
| All three | 27.5% |
| Associated with being named | Odds ratio | 95% interval |
|---|---|---|
| Visible ad measurement | 3.5× | 1.5–8.3 |
| Lot map | 3.3× | 1.4–8.2 |
| More communities (log) | 2.4× | 1.2–4.5 |
| Instant response path | 2.1× | 0.8–5.4 |
| Price published | 1.1× | 0.5–2.8 |
Logistic model with state controls (n = 150). Models: gpt-5.5, claude-opus-5 and gemini-3.8-flash, asked September 27, 2026. Answers vary between runs. Larger builders are both better equipped and better known, so these are associations, not causes.
Five hypotheses and their pass rules were written before any data was collected. Three did not pass as written, and we report them that way.
| Hypothesis | Result | Verdict |
|---|---|---|
| H1 A quarter or more of active Google advertisers lack an Ads tag and call tracking | 15% [6–34], n = 26 | Not supported |
| H2 Most builders have no instant response path | 67% [62–72]; at least 56% corrected | Supported |
| H3 Nationals lead by 20+ points with similar lab speed | +25 and +45 points; nationals 12 points slower | Not as written |
| H4 Fewer than half publish prices and a quarter show lot maps | 57% price; 36% lot map | Not supported |
| H5 A system signal goes with being named by AI | Measurement 3.5×, lot map 3.3× | Supported |
1,260 domains: Enroke's 2026 prospect research, builders found through Google for 57 Sun Belt cities, and 22 public homebuilders. 109 were not builders; 95 could not be measured, including 58 that answered automated visits with a bot challenge, which we did not bypass.
Each site was rendered in a browser; tools were identified by the web addresses they load, never by words on the page. Ad activity comes from Google's Ads Transparency Center (275 domains before Google limited lookups). No leads were ever submitted to any builder.
Independent reviewers labeled screenshots of 150 sites blind to the detectors. On the 45-site hold-out: price 91% agreement, lot map 82%, real-time chat 91% (recall 0.60), homepage form 84%.
We do not estimate any builder's ad spend, we do not name individual builders, and we do not treat associations as causes.
The full report adds what this page leaves out:
We use your email to send you this report and occasional research. No spam, and we never share it.