We have an uncanny ability to model the physical assets of a city — its roads, buildings, utilities, flood plains. But when it comes to understanding the people those assets are built for, we are largely operating blind. This gap between what cities measure and what residents actually experience is the defining challenge of modern urban planning.
This article is based on a presentation delivered at the 61st ISOCARP World Planning Congress in Riyadh in December 2025, where we introduced a methodology for applying AI-driven sentiment analysis to urban decision-making. What follows is the argument for why this matters, how the technology works, and what it looks like in practice.
The Planner's Dilemma: Measured Performance vs Lived Experience
Cities generate unprecedented volumes of data. We can measure housing accessibility, employment density, transport coverage, and urban vitality with increasing precision. But these quantitative metrics leave a fundamental question unanswered: do the people who live in these places actually feel that they work?
A district might score highly on employment density, yet residents report feeling that opportunities are scarce. A neighbourhood could be well-resourced with facilities, yet show unexpected patterns of dissatisfaction. These contradictions reveal something important: conventional analysis routinely overlooks lived reality.
The traditional data sources that planners rely on compound this problem. Census data is collected once and rarely updated. Surveys may pre-date the issue being analysed. Different regions use different methodologies, making comparison difficult. And where data is missing entirely, prediction becomes impossible.
“A city's functionality depends on its infrastructure, but its performance depends on how the people who live there behave.”
This is not an abstract point. Research into the BedZED eco-housing development in London found that among 24 identical homes, the worst-performing unit consumed three times more energy than the best — the only variable was occupant behaviour. Infrastructure provides the potential; human behaviour determines the outcome.
Three AI Technologies That Change Everything
Three distinct AI capabilities, when combined, allow planners to move from modelling physical systems to understanding how those systems are actually experienced. Each answers a different type of planning question.
Natural Language Processing: What are people saying?
NLP acts as a listener at scale — processing thousands of conversations simultaneously across social media, review platforms, forums, and news sources. It extracts emotional tone, identifies key concerns, detects geographic references, and classifies topics. In the GCC context, this requires Arabic-native models that understand Khaleeji, Najdi, and Hejazi dialects, not just Modern Standard Arabic. Generic global AI models miss 30–40% of sentiment nuance in regional Arabic expression.
Computer Vision: What do the spaces actually look like?
Computer vision analyses images at scale — classifying space types, identifying user demographics and activity patterns, assessing design elements and condition, and reading cultural context. Trained on millions of images from GCC urban environments, these models understand local architectural styles, social gathering patterns, and cultural appropriateness in ways that globally-trained models cannot.
Retrieval-Augmented Generation: Does the evidence support the claim?
RAG functions as a cross-checking research assistant. When sentiment data surfaces an issue — say, frustration with traffic in a particular district — RAG can pull verified data sources to confirm or challenge the finding. If traffic data shows 40% higher congestion than average, the sentiment moves from opinion to evidence. Sentiment without verification is opinion. Verified sentiment is evidence for action.
From Sentiment to Satisfaction: Why Binary Analysis Falls Short
Traditional sentiment analysis classifies expression as positive, negative, or neutral. For urban planning, this is inadequate. The statements “the traffic is unbearable” and “the traffic could be better” both register as negative, yet they represent very different levels of urgency and require very different responses.
The methodology we developed adapts the Net Promoter Score framework — validated across 400+ companies over 20+ years — to urban contexts. Instead of a binary classification, each expression is scored on a continuous 0–10 scale and mapped to an 8-tier intervention framework calibrated against the SERVQUAL service quality dimensions.
Scores of 9–10 indicate a Promoter Zone: celebrate and maintain. Scores of 7–8 sit in the Passive Zone: maintain and track. Scores of 5–6 represent the Concern Zone: monitor trends. Scores of 3–4 signal the Risk Zone: investigate root causes. And scores of 0–2 represent Crisis: act immediately.
This approach has been validated against face-to-face intercept surveys across GCC cities, achieving an r=0.84 correlation — meaning the AI-derived scores closely track what researchers find when they speak to people directly.
Making Sentiment Spatial
A satisfaction score is only useful if you know where it applies. Every piece of analysis is geolocated — anchored to districts or hex-grids — so that spatial patterns become visible. A district-level map immediately reveals disparities that vanish in aggregate data.
Green zones (scores 8–10) indicate high satisfaction. Amber zones (4–7) flag emerging issues worth monitoring. Red zones (0–4) demand immediate attention. When these satisfaction maps are overlaid with physical data layers — green space coverage, transit networks, industrial proximity — correlations emerge that point toward causation, not just correlation.
Low mental health satisfaction in areas lacking parks. Mobility dissatisfaction highest where transit coverage is poor. Air quality complaints clustered near industrial zones. When satisfaction drops match physical characteristics, you have evidence for intervention — not assumption.
The Data Revolution: From Years to Hours
Perhaps the most transformative aspect of this approach is cadence. Traditional census data updates every 5–10 years. The non-traditional data sources feeding this methodology update continuously: economic transaction data daily, traffic and transit data in real-time, mobile-derived population data daily, and facility reviews and social content hourly.
In the GCC, where 99.7% of internet users in Bahrain, 95.7% in KSA, and 92.3% in the UAE are active on social media, the volume of available urban intelligence is extraordinary. Every post, review, and interaction is potential urban data waiting to be structured.
From Insights to Action
The end goal is not understanding for its own sake. The methodology is designed to drive three categories of action: infrastructure investment targeted where gaps are highest, evidence-based policy and regulatory adjustments, and hyper-local programmes designed for specific community needs.
The most valuable insights emerge when sentiment diverges from quantitative indicators. An area with high job density but low perceived opportunity suggests a skills mismatch or wage issue. A well-resourced neighbourhood with unexpected dissatisfaction points to quality perception differing from quantity. These contradictions are where the real planning value lies.
What This Means for Practice
For planners, policymakers, and urban consultants looking to incorporate these approaches, we propose a five-step framework: identify knowledge gaps in current datasets, map available non-traditional data sources, define satisfaction indicators aligned to planning objectives, establish baseline measurements, and monitor change over time.
The point is not that every organisation needs to build these capabilities from scratch. The point is that understanding what AI can do equips planners to ask better questions and evaluate tools more effectively.
The companion articles in this series examine two case studies in detail: how this methodology was applied to measure pilgrim satisfaction across 1.67 million Hajj attendees, and how 2.7 million data points shaped new housing guidelines for the city of Al Ain.
Explore the GUS Platform
The methodology described in this article is operationalised through GUS, SilaCities' urban intelligence platform. GUS integrates NLP, computer vision, and RAG across 70+ API endpoints to deliver real-time urban insights.