Talk to the Map: Is Natural-Language GIS the Next Public Geoportal Interface?

For decades, public geoportals have stood as the locked vaults of municipal intelligence. They house a wealth of information—from zoning boundaries and property tax records to real-time air quality metrics and flood hazard maps—intended to drive civic transparency and data-informed policy. Yet, extracting any meaningful insight from these databases has traditionally required a steep technical toll. To answer a simple question about their own neighborhood, a citizen has had to master complex database schemas, write structured query language (SQL) scripts, or navigate the dense, menu-heavy environments of desktop GIS software like ArcGIS or QGIS. Consequently, public spatial data has remained a playground for specialized professionals, leaving community advocates, investigative journalists, and local planners on the outside looking in.

Now, a profound structural shift is underway. By fusing large language models (LLMs) with geographic information science (GIScience), a new paradigm of Natural-Language GIS (NL-GIS) is emerging. This technology completely reimagines how the public interacts with spatial information. Instead of navigating rigid dropdowns or writing code, users can simply "talk to the map," using conversational language to retrieve data, run spatial overlays, and render custom cartographic layouts on the fly. In a comprehensive spatial analytics assessment surveying more than 250 geospatial professionals, conversational interfaces were highlighted as the single most critical technological leap forward in the field, signaling a future where advanced geography is democratized for everyone.

The Continuum of Empowerment: From Google Earth to Conversational Autonomy

To understand the magnitude of this shift, we must look back to the mid-2000s, when mapping technology experienced its first major wave of democratization. The launch of Google Earth in 2005 realized the vision of a "Digital Earth"—an intuitive, three-dimensional virtual globe that turned our planet into an organizing metaphor for global information.

This milestone birthed the era of "neogeography," turning ordinary citizens into amateur map-makers. For the first time, anyone with an internet connection could view high-resolution satellite imagery, overlay GPS coordinates, and crowdsource community-driven maps through platforms like OpenStreetMap. Prominent geographers declared that just as the personal computer had democratized computing, Google Earth would democratize GIS.

Yet, the virtual globe revolution had a glaring limitation: it democratized spatial visualization, but not spatial analysis. While a citizen could pan across satellite images or toggle pre-processed layers, they could not easily perform complex computations. The core engines of spatial logic—calculating pedestrian service buffers, intersecting socioeconomic census tables with environmental hazard zones, or running multi-criteria suitability analyses—remained locked behind expensive software licenses, rigid academic training, and administrative bottlenecks.

Conversational GIS represents the second, and far more radical, wave of spatial empowerment. It shifts the public from passive consumers of pre-made maps to active creators of spatial analysis. By translating natural language directly into automated data pipelines, custom cartography, and statistical charts, NL-GIS puts the computational agency of a professional GIS analyst into the hands of anyone who can write a text message or speak a sentence.

Empowering the Civic Sphere: Grassroots Action, Newsrooms, and Crisis Operations

By removing technical friction, conversational interfaces are turning static municipal datasets into active drivers of localized decision-making. This empowerment is not abstract; it is manifesting as a practical tool for grassroots neighborhood groups, budget-constrained newsrooms, and emergency personnel operating in high-stress environments.

Grassroots Civic Advocacy

In traditional open-data portals, public information is often fragmented across separate, poorly indexed tables. When a community group seeks to evaluate park equity or environmental justice, they are forced to manually download, clean, and join vector shapefiles using desktop software. Conversational GIS collapses this multi-step process into a single, intuitive prompt:

"Show me a map of all residential areas in my ward that are more than a ten-minute walk from a public park, and highlight those that overlap with historically underfunded neighborhoods."

Behind the scenes, the conversational agent parses the request, pulls the park boundaries, runs a pedestrian network analysis using real sidewalk data, intersects the results with census block groups, and displays a styled vector map paired with a narrative explanation.

During local planning meetings, this allows residents to engage in real-time scenario modeling. Rather than waiting weeks for a consultant to run a study, planning boards and neighborhood residents can ask:

"What if we move the proposed transit line two blocks north? How does that change pedestrian access for high-density housing?"

The agent can instantly compute the routing updates, visually adjusting the spatial buffers on a digital twin of the city to show the immediate socioeconomic impact.

Supercharging Local Journalism

Local newsrooms, crippled by shrinking budgets and a lack of dedicated data specialists, frequently struggle to report on systemic urban trends. Conversational GIS tools integrated with open-source Model Context Protocols (MCP) allow journalists to conduct rapid, cross-database investigations.

Using tools like Civic AI Tools—an open-source project that connects generative AI assistants to Socrata-powered open data portals—reporters can perform complex queries across hundreds of municipal databases simultaneously. A reporter can ask:

"Identify the correlation between landlord code violations and older apartment buildings in our borough, and tell me the top three property owners responsible."

The assistant instantly queries the city's housing inspection data, matches it with parcel and corporate registration records, identifies the statistically relevant clusters, and outputs a publication-ready choropleth map and structured chart.

The benchmark for these kinds of capabilities is mapped by datasets like TACO, which tests text-to-SQL performance against thousands of real-world smart city and open data schemas from the U.S. and Beijing, ensuring models can handle vague, multi-database queries.

Rapid Emergency Response

During flash floods, wildfires, or structural emergencies, static maps are too slow to guide first responders. Conversational GIS allows incident commanders to manage active situations using spoken commands and live data feeds.

During a major storm event, an emergency coordinator can verbally query the system:

"Find all elderly care facilities within the current flood zone, calculate their fastest evacuation routes, and map which roads are blocked by rising waters."

By linking conversational agents with real-time hazard mapping platforms, the system pulls live satellite footprints, overlays municipal facility registries, calculates optimal detours around reported closures, and instantly broadcasts dynamic navigation maps to field personnel.

User Role	Traditional GIS Obstacles	Conversational GIS Solutions	Core Civic Impact
Civic Advocate	Opaque data schemas26 Desktop software costs4 Manual geometry joins20	Natural-language prompts20 Automated walkability analysis27 Narrative summaries28	Grassroots equity audits27 Democratic community representation8
Local Journalist	Fragmented city APIs5 Lack of programming skills2	Cross-portal queries32 Automated spatial SQL34	Rapid investigative mapping19 Empirical cross-city reporting13
Emergency Responder	Heavy data-loading latency11 High-stress UI environments24	Real-time speech commands11 Live stream integrations35 Dynamic spatial routing27	Drastically reduced response times27 Dynamic situational awareness11
Urban Planner	High consulting fees20 Protracted modeling cycles27	Conversational scenario testing27 Digital twin interactions13	Inclusive public planning11 Low-cost feasibility studies27

Technical Paradigms: Enterprise Clouds and Open-Source Sandboxes

While these grassroots scenarios paint a revolutionary picture of civic empowerment, they do not materialize out of thin air. Behind the intuitive "ask and receive" interface lies a complex, bifurcated engine room of modern software architecture designed to translate human intent into programmatic geographic logic. The technology driving this shift is currently split between two competing paradigms: massive, cloud-integrated enterprise ecosystems and agile, self-contained open-source sandboxes. Both strive to convert unstructured human speech into precise machine execution, but they do so through vastly different infrastructural pathways.

Enterprise Cloud Frameworks

Geospatial giants are embedding conversational assistants directly into enterprise ecosystems, allowing natural language queries to run natively on cloud data warehouses.

• Carto AI Agents: Carto has transitioned to an "Agentic GIS" model, using models like Claude 3.5 Sonnet and Gemini 1.5 Pro within cloud platforms like Google BigQuery and Snowflake. Operating through Carto’s MCP Server, these agents translate natural-language commands into highly optimized spatial SQL queries. They can dynamically style vector layers, generate statistical histograms directly inside a chat window, and run complex workflows over billions of data points without leaving the secure data warehouse.

• Esri ArcGIS Assistants: Esri has integrated conversational helpers across its suite. This includes the ArcGIS Hub Assistant, which lets public site visitors explore open data using simple prompts, and the ArcGIS Pro Assistant, which writes ArcPy code and triggers routine mapping actions from plain English commands. In Web APIs, developers can embed components like arcgis-assistant directly into client apps, giving users out-of-the-box navigation, data filtering, and help agents.

• Mapbox Location AI: Focused on integrating spatial awareness into external AI systems, Mapbox provides an MCP Server that allows any LLM to perform geocoding, proximity searches, and route calculations. This powers MapGPT, an automotive co-pilot that lets drivers converse about local landmarks, charge levels, and route updates in real time.

Open-Source Architectures

Open-source projects prioritize code transparency and platform independence, allowing developers to create highly customizable and reproducible workflows.

• GISclaw: This agent connects an LLM core with a secure, sandboxed Jupyter-like Python environment pre-loaded with open-source GIS libraries like GeoPandas, rasterio, and libpysal. Using a specialized self-correction loop, GISclaw parses data schemas and automatically corrects its own geoprocessing code based on execution errors.

• AtQuery: A lightweight, offline-capable QGIS plugin that uses local models (via Ollama) to translate conversational prompts into workflow automation, keeping user data entirely on local devices.

Technical Hurdles: Hallucinations and the "Black Box" Map

Yet, whether deployed via multi-billion-dollar enterprise clouds or localized open-source scripts, this conversational pipeline must eventually confront a hard, scientific reality: geography is an unforgivingly rigid discipline. Language models are fundamentally built to predict the next logically plausible word, but the physical world does not operate on linguistic probabilities. When a system designed for semantic fluency is tasked with mathematical and geometric precision, the frictionless user experience can quickly mask devastating analytical errors. This is the realm of spatial hallucinations and the "black box" map—where visually pristine cartography can silently hide structural falsehoods.

Common spatial hallucinations include:

• Projection Failures: A model might attempt to create a $1000$-meter buffer around a transit line but generate code using geographic coordinates (degrees) rather than projected coordinates (meters). This error silently creates an invalid buffer spanning 1,000 degrees, wrapping around the globe.

• Inverted Topological Predicates: Models often struggle with spatial orientation. In a point-in-polygon overlay, the model may invert the target and join parameters, resulting in a completely empty data set or corrupted tables.

• Heuristic Failures: In network routing, models operating without strict physical boundary inputs can suggest impossible paths that cross topological barriers or cut directly through solid structures.

These errors feed into the "black box" problem of AI cartography. In traditional GIS, an analyst must carefully document, audit, and clean every spatial transaction. In a conversational interface, this work is entirely hidden.

A conversational agent will construct and render a beautiful, visually convincing map even if the underlying spatial logic was flawed or data was dropped. This creates a false sense of spatial trust, where community members or municipal leaders may make zoning, development, or safety decisions based on highly misleading geographic outputs.

Moving Toward the 3-to-5-Year Horizon: Google Earth AI and the Spatial Web

Navigating these deep cognitive deficits is the primary challenge defining the current research frontier, yet the blistering pace of innovation suggests that a vastly more capable era of spatial thinking is just over the horizon. Over the next three to five years, the paradigm of "talking to the map" will expand from basic, flat chat queries into dynamic, multidimensional, and highly predictive intelligence engines. Leading this charge is a new generation of planetary-scale models designed to fundamentally bridge the gap between human language and physical Earth observation.

The Launch of Google Earth AI

Google’s introduction of Google Earth AI represents the most significant leap forward in public spatial thinking since the original platform's launch in 2005. Built on Gemini's advanced multimodal reasoning and planetary-scale modeling, Google Earth AI is designed to make complex Earth observation and satellite analysis accessible to non-technical users.

Instead of writing complex code in Google Earth Engine, a citizen or researcher can use zero-shot, natural-language prompts to interrogate the globe. A user can zoom into a neighborhood and ask:

"Where are all the storm drains within 50 meters of public schools in Red Hook, Brooklyn?"

Or, looking at environmental crises, they can ask the model to analyze satellite pixels directly:

"Which global mangrove forests are most similar to mangroves in the Mekong delta in Vietnam?"

These advanced models enable open-vocabulary retrieval, allowing users to locate objects on the Earth's surface (e.g., "find all wind turbines") or spot temporal environmental changes (e.g., distinguishing "pre-construction" from "active construction") simply by describing them.

Additionally, Google Earth AI introduces Population Dynamics Insights (PDI), a first-of-its-kind geospatial embeddings dataset. PDI compresses billions of aggregated, anonymized trends from Google Search, Google Maps popular times, POIs, and local environmental conditions (like weather and air quality) into rich, $330$-dimensional location vectors. This turns static, decade-old census snapshots into real-time, monthly updated behavioral models, allowing public health officials and urban planners to project socioeconomic needs, anticipate disease outbreaks, and model changing neighborhoods through plain-language interfaces.

Agentic Digital Twins & Co-Scientists

Beyond conversational search, the horizon holds fully autonomous GIS co-pilots. Platforms like the GIS Co-Scientist framework distribute scientific tasks across six specialized sub-agents (understanding, data retrieval, planning, spatial modeling, result analysis, and reporting), while maintaining critical human-in-the-loop checkpoints.

In this near-future city planning landscape, citizens and planning boards will interact with complex urban simulations via voice. They will verbally stress-test neighborhoods, modeling real-time urban shifts like pedestrian-only plazas or the localized impacts of a 500-year flood event on neighborhood infrastructure.

Establishing Civic Trust: Model Context Protocol as a Governance Layer

Unlocking this highly anticipated future, however, requires more than just raw computational power or sophisticated planetary embeddings; it demands a robust method for establishing civic trust. For municipal governments tasked with serving the public, the primary barrier to adopting these conversational interfaces is not a lack of vision, but a lack of control. To prevent automated mapping tools from descending into a wild west of hallucinated statistics and security vulnerabilities, pioneering cities are deploying a new, standardized architectural buffer known as the Model Context Protocol (MCP).

Pioneered by Santiago Garces, the Chief Information Officer of Boston, the city’s Open Context project represents a major milestone in public data governance. When a resident queries Boston's open data portal through an AI assistant, the model is not given direct access to municipal databases or allowed to generate arbitrary code. Instead, the MCP server translates the resident's natural language request into a highly structured tool call. The middleware executes the query natively within Boston’s secure open data repository, generates the SQL against the live database, and returns only the verified, grounded datasets or maps back to the client.

Citizen Query

"How many active code violations are on my street?"

AI Assistant (LLM)

Translates query to
standard tool call

Boston Open Context (MCP)

(Agentic Middleware)

Constructs deterministic
grounded SQL

Secure City Data Portal

(e.g., Socrata, Esri, Opendatasoft)

Executes query and returns
verified tabular/spatial data

AI Assistant (LLM)

Renders map and summary
with source attribution

✓ Validated Map delivered to Citizen

This "Agentic Middleware" approach solves several critical public policy hurdles:

1. Dramatically Lowering Hallucinations: By binding the AI to validated, structural database schemas, it prevents the model from fabricating non-existent data, numbers, or records.

2. Computational Efficiency: Running queries natively on the open data portal's own database engine avoids overloading the AI’s limited context window, delivering faster, lower-cost, and more accurate answers.

3. Robust Security & Privacy: The MCP layer prevents arbitrary code execution, shielding municipal IT databases from cyber-attacks. It also aligns with strict AI safety policies; in Boston, AI systems are banned from processing data that could impact a resident's civil liberties, property rights, or public safety. The open data portal functions as a safe, transparent, and low-risk testing ground to refine these conversational interfaces before they are scaled across other areas of government.

Through open-source initiatives, Boston is working to share its MCP server configuration with other municipalities worldwide, paving the way for a standardized, secure conversational GIS layer across global cities.

A New Era of Collective Spatial Reasoning

The evolution of public mapping over the last two decades reveals a clear trajectory toward total democratization. In 2005, the virtual globe revolution represented by Google Earth gave humanity a collective viewport—allowing the public to witness the physical reality of the planet and browse pre-made layers of geographic data. Yet, the power of spatial computation and deep geographic inquiry remained locked behind academic, financial, and programmatic gatekeepers.

Natural-Language GIS is breaking those gates down. By transforming the interface from technical code to natural, human conversation, we are entering the second wave of public spatial empowerment. In this new landscape, a neighborhood activist, a local journalist, or a crisis coordinator can construct complex geoprocessing workflows, run statistical overlays, and generate insightful maps simply by describing their intent.

As we look toward the 3-to-5-year horizon, frameworks like Google Earth AI and the secure, standardized guardrails of the Model Context Protocol will define this transition. The maps of tomorrow will no longer be silent, static products designed by select experts for passive consumption. Instead, they will operate as active, conversational partners—grounded in real-world facts, secure by design, and ready to help anyone, anywhere, interrogate the geography of their physical world.

• Sources

  1. From Data to Knowledge: A Modular Framework for AI- Enhanced Enrichment of Open Data Portals - CEUR-WS.org, https://ceur-ws.org/Vol-4127/paper31.pdf

  2. Designing a Large Language Model Based Open Data Assistant for Effective Use - KIT, https://publikationen.bibliothek.kit.edu/1000171237/154024533

  3. FY 2022 Adopted IT Plan - Fairfax County, https://www.fairfaxcounty.gov/informationtechnology/sites/informationtechnology/files/assets/itplan/2023-adopted/fy2023itplan.pdf

  4. On the Use of LLMs for GIS-Based Spatial Analysis - MDPI, https://www.mdpi.com/2220-9964/14/10/401

  5. arXiv:2007.10503v1 [cs.SE] 20 Jul 2020, https://arxiv.org/pdf/2007.10503

  6. GISclaw: An Open-Source LLM-Powered Agent System for Full-Stack Geospatial Analysis - arXiv, https://arxiv.org/html/2603.26845v1

  7. GeoAnalystBench: A GeoAI benchmark for assessing large language models for spatial analysis workflow and code generation - arXiv, https://arxiv.org/html/2509.05881v1

  8. View of Many, many maps: Empowerment and online participatory mapping | First Monday, https://firstmonday.org/ojs/index.php/fm/article/view/1620/1535

  9. What is CARTO?, https://docs.carto.com/getting-started/carto-in-a-nutshell

  10. What's new in CARTO - Q1 2026, https://carto.com/blog/whats-new-in-carto-q1-2026/

  11. How AI Could Potentially Transform the Future of GIS in Local Government, https://geospatialtraining.com/how-ai-could-potentially-transform-the-future-of-gis-in-local-government/

  12. Top AI Tools for Cartographic Design in 2026 - Dinopix, https://dinopix.ai/resources/ai-design-tools/ai-tools-for-cartographic-design-complete-2025-review-and-comparison

  13. Full article: Tool, toolmaker, and scientist: case study experiences using GIS in interdisciplinary research - Taylor & Francis, https://www.tandfonline.com/doi/full/10.1080/15230406.2020.1748113

  14. Harnessing the power of volunteers, the internet and Google Earth to collect and validate global spatial information using Geo-Wiki - IIASA PURE, https://pure.iiasa.ac.at/id/eprint/11389/1/%E2%80%99Harnessing%20the%20Power%20of%20Volunteers%2C%20the%20Internet%20and%20Google%20Earth%20to%20Collect%20and%20Validate%20Global%20Spatial%20Information%20using%20Geo-Wiki.pdf

  15. Students' Perception of the Pedagogical Approach to Geography Teaching and Learning Through Google Earth Pro - MDPI, https://www.mdpi.com/2227-7102/16/2/268

  16. Neogeography and the delusion of democratisation - Muki Haklay's personal blog, https://povesham.wordpress.com/2012/06/22/neogeography-and-the-delusion-of-democratisation/

  17. The Practice of Neogeography in Community-Based Organizations - SURFACE at Syracuse University, https://surface.syr.edu/cgi/viewcontent.cgi?article=1077&context=thesis

  18. Hacker Cartography: Crowdsourced Geography, OpenStreetMap, and the Hacker Political Imaginary, https://acme-journal.org/index.php/acme/article/download/1237/1032/3647

  19. Participatory GIS - Progress in Human Geography - SciSpace, https://scispace.com/pdf/participatory-gis-a-people-s-gis-1mf9po59kk.pdf

  20. Full article: GeoJSON agents: a multi-agent LLM architecture for geospatial analysis—function calling vs. code generation - Taylor & Francis, https://www.tandfonline.com/doi/full/10.1080/20964471.2026.2615511

  21. An Autonomous GIS Agent Framework for Geospatial Data Retrieval - arXiv, https://arxiv.org/pdf/2407.21024

  22. Generative AI Role in Shaping the Future of Open Data Ecosystems: Synergies amidst Paradoxes | by Nikiforova Anastasija | Medium, https://medium.com/@nikiforova.anastasija/generative-ai-role-in-shaping-the-future-of-open-data-ecosystems-synergies-amidst-paradoxes-00fcfb2518fc

  23. Natural Conversational Interfaces to Geospatial Databases | Request PDF - ResearchGate, https://www.researchgate.net/publication/220606108_Natural_Conversational_Interfaces_to_Geospatial_Databases

  24. Full article: GIScience in the era of Artificial Intelligence: a research agenda towards Autonomous GIS - Taylor & Francis, https://www.tandfonline.com/doi/full/10.1080/19475683.2025.2552161

  25. AtQuery — QGIS Python Plugins Repository, https://plugins.qgis.org/plugins/AtQuery/

  26. GeoContra: From Fluent GIS Code to Verifiable Spatial Analysis with Geography-Grounded Repair - arXiv, https://arxiv.org/html/2605.00782v1

  27. Principles, Applications & Careers - GIS in Environmental Science, https://www.environmentalscience.org/principles-applications-gis

  28. AI Assistants in ArcGIS Provide Practical Help and Give Useful Results | Winter 2026 - Esri, https://www.esri.com/about/newsroom/arcnews/ai-assistants-in-arcgis-provide-practical-help-and-give-useful-results

  29. [2307.07930] GeoGPT: Understanding and Processing Geospatial Tasks through An Autonomous GPT - arXiv, https://arxiv.org/abs/2307.07930

  30. A Geodatabase Design for the Development of a Digital Twin for Urban Environments: A Case Study from Turin, Italy - ResearchGate, https://www.researchgate.net/publication/385116221_A_Geodatabase_Design_for_the_Development_of_a_Digital_Twin_for_Urban_Environments_A_Case_Study_from_Turin_Italy

  31. A growing observatory of examples of how open data from official sources and generative artificial intelligence (AI) are intersecting across domains and geographies. - Policy Repository, https://repository.opendatapolicylab.org/genai/

  32. GitHub - npstorey/civic-ai-tools: Open-source platform connecting AI assistants to government open data — MCP server, curated civic MCP directory, and anti-hallucination framework for all 559 Socrata portals, https://github.com/npstorey/civic-ai-tools

  33. npstorey/socrata-mcp-server - GitHub, https://github.com/npstorey/socrata-mcp-server

  34. TACO: A Benchmark for Open-Domain Text-to-SQL with Ambiguous and Cross-Database Queries - VLDB Endowment, https://www.vldb.org/pvldb/vol19/p1386-fan.pdf

  35. Leveraging GIS for Faster and Smarter Natural Disaster Response - Quarticle, https://quarticle.ch/blog/leveraging-gis-for-faster-and-smarter-natural-disaster-response

  36. Building an “Agentic Middleware” for the City Government: Boston's Experiment with Model Context Protocol - Reboot Democracy, https://rebootdemocracy.ai/blog/boston-agentic-middleware

  37. GIS DAY 2025, https://webapps.transportation.wv.gov/TWS/IT/GIS/GISDAY2025/CopyofGISDAY2025WhatsNewinArcGISShared.pdf

  38. MMP-A*: Multimodal Perception Enhanced Incremental Heuristic Search on Path Planning, https://arxiv.org/html/2601.01910v2

  39. Archived GIS Day 2023 | Geographic Information Services | County of Santa Clara, https://gis.santaclaracounty.gov/gis-days/archived-gis-day-2023

  40. GISclaw: A Comprehensive Open-Source LLM Agent System for Realistic Multi-Step Geospatial Analysis - arXiv, https://arxiv.org/html/2603.26845v2

  41. Geospatial Trends in 2025: The Latest Industry Evolutions - Korem, https://www.korem.com/geospatial-trends-in-2025-the-latest-industry-evolutions/

  42. CARTO AI, https://docs.carto.com/carto-user-manual/settings/carto-ai

  43. AI Agents - CARTO Academy, https://academy.carto.com/agentic-gis/ai-agents

  44. What's new in ArcGIS Hub (February 2026) - Esri, https://www.esri.com/arcgis-blog/products/arcgis-hub/announcements/whats-new-in-arcgis-hub-february-2026

  45. Introducing AI components (beta) in the ArcGIS Maps SDK for JavaScript - Esri, https://www.esri.com/arcgis-blog/products/js-api-arcgis/developers/introducing-ai-components-beta-in-the-js-maps-sdk

  46. AI Assistant component (beta) | Sample | ArcGIS Maps SDK for JavaScript - Esri Developer, https://developers.arcgis.com/javascript/latest/sample-code/ai-assistant/

  47. Geospatial AI & Location AI for AI Agents - Mapbox, https://www.mapbox.com/location-ai/build

  48. Location AI Overview | Help | Mapbox, https://docs.mapbox.com/help/getting-started/location-ai/

  49. Mapbox Location AI, https://www.mapbox.com/location-ai

  50. Mapbox MapGPT | AI assistant with location intelligence, https://www.mapbox.com/mapgpt

  51. Mapbox MapGPT: Deliver natural conversations with a location-intelligent AI assistant, https://www.youtube.com/watch?v=LGbuBSCEOsc

  52. Esri Tutorials Spotlight: AI Assistants in ArcGIS, https://community.esri.com/t5/education-blog/esri-tutorials-spotlight-ai-assistants-in-arcgis/ba-p/1690825

  53. GeoContra: From Fluent GIS Code to Verifiable Spatial Analysis with Geography-Grounded Repair - arXiv, https://arxiv.org/pdf/2605.00782

  54. Revisiting General Map Search via Generative Point-of-Interest Retrieval - arXiv, https://arxiv.org/html/2605.03397v1

  55. WebChain: A Large-Scale Human-Annotated Dataset of Real-World Web Interaction Traces, https://openaccess.thecvf.com/content/CVPR2026/papers/Fan_WebChain_A_Large-Scale_Human-Annotated_Dataset_of_Real-World_Web_Interaction_Traces_CVPR_2026_paper.pdf

  56. (PDF) GeoContra: From Fluent GIS Code to Verifiable Spatial Analysis with Geography-Grounded Repair - ResearchGate, https://www.researchgate.net/publication/404396800_GeoContra_From_Fluent_GIS_Code_to_Verifiable_Spatial_Analysis_with_Geography-Grounded_Repair

  57. Benchmarking LLM Agents for Environmental Geospatial Analysis Across Frontier and Open-Weight Foundation Models - arXiv, https://arxiv.org/html/2606.12821

  58. The Future of Mapping Is AI Native: Inside Mapbox's Vision for Agents, Real-Time Location Intelligence, and Next-Gen Geospatial Apps - Matt Forrest, https://forrest.nyc/ai-native-mapping-mapbox-kieran-mccann-future-of-location-intelligence/

  59. Google Earth AI, https://research.google/research-areas/google-earth-ai/

  60. Google Earth AI, https://ai.google/earth-ai/

  61. Google Earth AI: Our state-of-the-art geospatial AI models, https://blog.google/innovation-and-ai/products/google-earth-ai/

  62. Geospatial solutions and analytics - Google Maps Platform, https://mapsplatform.google.com/maps-products/geospatial-analytics/

  63. Google Maps Platform adds Earth AI-powered aerial and satellite insights | GIM International, https://www.gim-international.com/content/news/google-maps-platform-adds-earth-ai-powered-aerial-and-satellite-insights

  64. Google Earth AI models and datasets in BigQuery | Google Cloud Blog, https://cloud.google.com/blog/products/data-analytics/google-earth-ai-models-and-datasets-in-bigquery

  65. From Static Maps to Geospatial AI: Announcing Population Dynamics Insights, https://mapsplatform.google.com/resources/blog/from-static-maps-to-geospatial-ai-announcing-population-dynamics-insights/

  66. 5 geospatial AI breakthroughs from Google Next '26 - CARTO, https://carto.com/blog/google-cloud-next-26-geospatial-ai-takeaways/

  67. The use cases of digital earth | Request PDF - ResearchGate, https://www.researchgate.net/publication/228639793_The_use_cases_of_digital_earth

  68. GIS Copilot: towards an autonomous GIS agent for spatial analysis - Semantic Scholar, https://www.semanticscholar.org/paper/GIS-Copilot%3A-towards-an-autonomous-GIS-agent-for-Akinboyewa-Li/a1cea74d534f08e7fd543c008c192560af939eb3

  69. GIS Co-Scientist: A Human-AI Collaborative Framework for Autonomous Geospatial Scientific Inquiry - ResearchGate, https://www.researchgate.net/publication/404620114_GIS_Co-Scientist_A_Human-AI_Collaborative_Framework_for_Autonomous_Geospatial_Scientific_Inquiry

  70. What Is Model Context Protocol (MCP) for AI Security? - Rapid7, https://www.rapid7.com/fundamentals/model-context-protocol-mcp/

  71. Boston's CIO wants the public — and other city governments — to use his open-source agentic AI tools | StateScoop, https://statescoop.com/bostons-cio-open-source-agentic-ai-tools/

  72. AI Agents - CARTO Documentation, https://docs.carto.com/carto-user-manual/ai-agents

  73. AI agents are coming for government. How one big city is letting them in - Fast Company, https://www.fastcompany.com/91504876/boston-cio-santi-garces-on-ai-agents-mcp-open-data