Research in the age of AI

Has AI changed our view on
real estate markets, theories
and research approaches?

Thies Lindenthal · University of Cambridge · 2026

Moving up the value chain...
Am I still useful in the research process? Does a human competitive edge remain?

Comparison of Core Research Competencies (my self-perception)

## Can AI generate good research ideas?<span class="subtitle"><br/>Eight ways to prompt the LLM: offering context, methods, and ideas from other fields</span>

* **A — Naïve:** no context, LLM draws on training only
						* **B — Full REE corpus:** all 1,676 abstracts provided as context
						* **C — Cluster-seeded:** one idea per research cluster; 51 ideas
						* **D-G — Method/paradigm transfer:** Ideas and methods from Econ/Finance or Psychology
						* **H — Citation-guided:** LLM steered toward high-impact territory

<div style="margin-top:0.6em; padding:0.5em 0.8em; background:#f4f4f4; border-left:3px solid #3B9AB2; font-size:0.6em; line-height:1.5; font-family:monospace;">
						<strong>Prompt (Condition A):</strong> "Generate 10 original real estate economics and finance research ideas. Each idea should be genuinely novel — not an incremental extension of standard hedonic or mortgage models, but a contribution that could open a new research programme. Think across the full breadth of the field: housing markets, commercial real estate, urban economics, real estate finance, mortgage markets, land use, REITs, climate risk, discrimination, neighbourhood effects, and so on."
						</div>

50 years of real estate research
1,676 REE articles + top-journal subsets embedded and clustered into a research space

Where do AI-generated ideas land?
Each dot is a generated idea projected into the research space

A: Naïve generation, based on LLM training only

B: Full REE corpus provided as context

C: Transfer of research methods and ideas from Econ/Finance

D: Transfer of research methods and ideas from Psychology

Universe, all journals

REE

Generated ideas

Comparing generated research ideas
Unconstrained variant (a); 100 ideas per condition except C (51, one per cluster)

Condition	Clusters	HHI	Atypicality	Periph./REE	NN dist./REE	Pred. cit.
A — Naïve	21 / 51	0.070	0.726 ± 0.180	3.16	2.57	11.9
B — Full REE corpus	23 / 51	0.074	0.707 ± 0.248	2.75	2.48	11.9
C — Cluster-seeded	31 / 51	0.043	0.720 ± 0.206	2.89	2.48	11.5
D — Method transfer (econ/fin)	25 / 51	0.089	0.735 ± 0.147	3.70	2.19	13.4
E — Method transfer (psych)	21 / 51	0.112	0.693 ± 0.210	1.85	2.70	11.2
F — Paradigm transfer (econ/fin)	23 / 51	0.080	0.735 ± 0.158	4.36	2.21	13.1
G — Paradigm transfer (psych)	15 / 51	0.161	0.632 ± 0.251	1.74	2.67	11.2
H — Citation-guided	27 / 51	0.066	0.729 ± 0.179	2.80	2.36	12.5

Predicting impact of ideas
Predicted 5-year citations by generation condition — constrained variant

\( log(1 + Citations_i) = \alpha + \beta_1\log(Age_i) + \beta_2 Atypicality_i + \beta_3 NN.dist_i + \beta_4 log(frontier.dist_i) \) \( + \beta_5 cluster.mean.cit_i + Journal.FE_i +\epsilon_i \)
\(R^2 = 0.182\); \(n = 1,074\); evaluated at REE baseline, age = 5

Has AI changed our view on real estate markets, theories and research approaches?

Moving up the value chain... Am I still useful in the research process? Does a human competitive edge remain?

50 years of real estate research1,676 REE articles + top-journal subsets embedded and clustered into a research space

Where do AI-generated ideas land? Each dot is a generated idea projected into the research space