The Attribute Table Is Where AI Earns Its Lunch

You've spent the morning fighting an attribute table that's half-right. The CAD handoff came in with layer names instead of asset types, and now you're manually reclassifying several hundred features. Meanwhile, the AI conversation in GIS has moved on to imagery analysis— object detection, land cover classification, rooftop assessment. That work is real. But it's not where most GIS hours go.

The AI tools for automating attribute table work already exist inside ArcGIS. Most GIS professionals don't know they're there— because the conversation about AI and GIS has been dominated by imagery, not the daily grind of data maintenance. As Fulcrum noted in their 2025 AEC report, AEC projects generate enormous amounts of geospatial data, and most of it never gets used in time to matter¹. Part of the reason: the work of populating and maintaining attribute data is still largely manual.

This guide maps five categories of AI tools to the five most common attribute table problems in AEC work. Each section covers what the tool does, what problem it solves, and where it still falls short. Start with the tool that matches your current problem.

Why Attribute Tables Are the Productivity Bottleneck

Attribute tables are time-consuming because the data that flows into them is never clean. CAD handoffs arrive in inconsistent formats. Inspection records live in free-text notes. Features are missing values that someone needs to calculate or estimate.

The pain points fall into four categories:

• CAD-to-GIS handoff: Attribute data in CAD appears as a mix of text, layer names, block attributes, leader line data, and sheetset information²— none of it maps cleanly to GIS schema. The handoff currently takes anywhere from a day to a week of manual attribute work per project, depending on project size².
• Missing values: Asset records routinely arrive with null condition ratings, installation dates, and asset classifications. These don't fill themselves.
• Field calculations at scale: Date logic, conditional values, classification expressions— each requires writing Arcade or Python. Doing this across thousands of features by hand doesn't scale.
• Inspection records: Free-text notes that contain structured data (condition codes, dates, materials) but aren't formatted as structured data— invisible to spatial analysis until someone extracts it.

One property assessor working with GeoAI tools put it directly: "I do it all from my desktop, because I don't have enough bodies to go outside."³ That capacity constraint is exactly what ArcGIS's AI tools now start to address— one problem at a time.

Plain-English Field Calculations— The Arcade Assistant

The Arcade assistant lets you type what you want a field calculation to do in plain English and it writes the Arcade expression for you. You review it, deploy it, and move on.

According to Esri⁴, the Arcade assistant (in beta as of mid-2026) uses AI to generate ArcGIS Arcade expressions from instructions typed in natural language. The supported expression types include field calculations, pop-ups, label expressions, renderers, clustering expressions, attribute expressions, and error-fixing in existing Arcade code. Platform availability is broad— Map Viewer, Field Maps Designer, ArcGIS Experience Builder, Scene Viewer, ArcGIS Data Pipelines, and the Maps SDK for JavaScript⁵.

The inspection scheduling example Esri documents⁴ is a useful benchmark for what this tool can handle: "When a mobile worker successfully completes a high pressure inspection, it is required to set the due date for the next inspection to 3 months in the future. The due date needs to be scheduled for a business day and not on a weekend." That's a multi-condition expression. You describe it in plain language; the assistant writes it.

"The best code is no code"— and this tool gets closer to that than anything Esri has shipped before. But here's the honest caveat: AI-generated Arcade expressions need review before deployment. The tool reduces time, not accountability. You still need enough GIS knowledge to recognize a logic error in the output.

Filling the Gaps— Predict Missing Values Using AI Model

Predict Missing Values Using AI Model trains a machine learning model on your existing non-null rows and applies it to estimate what the null values should be. It requires no code and no data science background.

Esri's documentation⁶ is precise: the tool "replaces missing values (nulls) in a feature class or table by automatically training a machine learning or deep learning model on available data and applying it to estimate missing numerical values." For AEC infrastructure work, this fits anywhere you have historical records that can inform null estimates— asset condition ratings, elevation values, count fields. If 80% of your asset inventory has condition ratings, the tool trains on that data and estimates the remaining 20%.

Two limitations to name directly. First, this tool is designed for numerical fields— text and categorical fields are handled by the NLP tools in the next section⁶. Second, it requires sufficient non-null data to train a reliable model. Small or imbalanced datasets produce unreliable results. Esri doesn't specify a minimum in public documentation— treat it as a qualitative constraint: the more complete your existing data, the more reliable the predictions.

A practical starting point: if more than half your feature records already have values for the target field, you have enough to train. Run the tool on a test set first and validate results against known values before applying to the full dataset.

The tool is part of the GeoAI Feature and Tabular Analysis toolset in ArcGIS Pro, which also includes Train Using AutoML and Predict Using AutoML⁷.

Feature Classification at Scale— ArcGIS AutoML

ArcGIS AutoML lets you train a machine learning model on labeled feature data and use it to classify other features— without writing code or knowing which algorithm to use.

The Feature and Tabular Analysis toolset⁷ includes two tools: Train Using AutoML (builds the model) and Predict Using AutoML (applies it). It automates the hardest parts. You keep the judgment.

What AutoML automates: data exploration, feature selection, and hyperparameter optimization (the process of tuning the model's internal settings for accuracy)— the hardest parts of traditional machine learning.

The AEC use case is partial asset inventories. If 70% of your infrastructure assets are classified by type, condition, or ownership, AutoML can classify the remaining 30% using the pattern in your existing labeled data. The output is a trained model package (.dlpk) that's reusable on future datasets— meaning work done once can apply to the next project.

AutoML earns its keep when your labeled data is the foundation— and the model becomes the multiplier across every future project.

Mining Free-Text Fields— NLP Entity Extraction

The Text Analysis toolset uses NLP (natural language processing) to pull structured data out of free-text attribute fields. Inspection notes that say "corroded flange, replaced August 2024" can be processed to extract condition, component, and date as separate structured values.

Esri's documentation⁸ describes it clearly: "The Text Analysis toolset contains tools that perform natural language processing on text, where text can be classified or transformed, and entities such as addresses can be extracted."

Entity types the tool can extract from text fields include:

• Addresses (with automatic geocoding to spatial output)
• Dates
• Place names
• Person names
• Monetary values

And for infrastructure asset management, that means something concrete. Maintenance records and work orders that contain location data can become spatial features automatically⁸.

The models behind the toolset are production-grade NLP architectures— the same class used in enterprise search and document processing at scale, including BERT and Mistral⁹. Through the Esri and Microsoft Azure OpenAI collaboration (July 2025)¹⁰, ArcGIS users also gain access to 90+ pretrained deep learning models via the GeoAI toolbox¹¹, and the platform supports custom NLP workflows with open-source and commercial LLMs¹².

In practical terms: you're not building on experimental tools. Inspection records that currently exist as unstructured text become queryable spatial data— the field note saying "corroded flange, replaced August 2024" becomes a condition attribute, a date field, and a work order, all without manual reclassification.

One setup note: the Text Analysis toolset isn't zero-configuration. It requires model selection or, for specialized domains, model training. If your firm works in regulated infrastructure with strict data standards, building an AI governance strategy before deploying NLP at scale is worth the time.

The External AI Path— ChatGPT, Claude, and ArcPy

ChatGPT (OpenAI) and Claude (Anthropic) can write ArcPy scripts for GIS attribute automation, and research shows they do it reliably. A 2024 ACM peer-reviewed study found ChatGPT-4 achieved an 80.5% task success rate generating ArcPy functions across GIS workflows¹³.

The approach is straightforward: describe what you want in plain language, including field names and what the script should do. The AI generates ArcPy. You test it, deploy it. According to Geospatial Training Services¹⁴, AI can generate scripts for:

• Field schema inspection (listing all field names and data types for a feature class)
• Ratio calculations between attribute values
• Multi-step workflows— add fields, buffer, clip, compute acreage
• SQL query generation via the ArcGIS Pro assistant's Query Layer mode, which "turns a plain-language description into a SQL query, explains how it was built, and adds the result to a map"¹⁵

But that 80.5% task success rate means roughly 1 in 5 AI-generated scripts will need correction before they work in production. All AI-generated GIS code should be tested in a nonproduction environment first. This path requires enough Python literacy to recognize a broken script— it's the "bring your own AI" approach, more flexible than Esri's built-in tools and with a higher skill floor.

For context on where this fits in a broader automation approach, the AI automation guide covers how to sequence these external tools alongside platform-native options.

Where AI Still Falls Short

AI tools for attribute tables improve on manual methods, but they don't fix the underlying problem: garbage-in, garbage-out still applies. If source data is missing, inconsistent, or wrong, AI will systematically produce confident-sounding errors.

The specific limitations:

• Training data floor: AutoML and Predict Missing Values both require sufficient non-null training data. Small or imbalanced datasets produce unreliable models. This constraint is underdocumented in Esri's public resources.
• Validation requirement: AI-generated Arcade expressions and ArcPy scripts need review by someone who can recognize a logic error. The ACM study's 80.5% success rate¹³ implies roughly 1 in 5 scripts fail before testing— the tool reduces time, not accountability.
• CAD connector alternatives: Many AEC firms have already invested in GeoBIM or direct CAD-GIS linking tools that reduce the handoff problem. AI still helps with standardization and classification post-import, but it isn't the only answer.
• Government data standards: Jurisdictions with strict attribute schemas may not accept AI-estimated values without manual review and sign-off. Know your compliance requirements before deploying AI estimation at scale.

The frame that holds: AI handles the tedious execution; the GIS professional provides the domain judgment about what the data should mean. That division of labor is what makes this work. The validation step isn't optional.

Where to Start— A Skill-Level Path

Start with the Arcade assistant. It works in your existing ArcGIS Online or Pro environment, requires no model training or Python, and produces immediate results on field calculation problems you're already dealing with.

The right starting point depends on the problem you're solving, not on how much AI experience you have. Don't start with the most powerful tool— start with the one that solves what you're already blocked on. Most GIS professionals can begin using AI for attribute work today, with tools they already have access to.

If mapping AI tools to your firm's specific GIS workflows feels like its own project, that's a concrete problem with a concrete answer. Dan Cumberland Labs helps AEC firms build AI implementation roadmaps that prioritize tools by where the manual hours are actually going. If you're figuring out where external help fits into that picture, what an AI consultant can actually do is worth a read before you decide.

And if you're thinking about this more strategically— not just which tools to try but how AI implementation fits your firm's broader roadmap— AI strategy for your firm is where that conversation starts.

FAQ

Can AI fill in missing values in ArcGIS attribute tables?

Yes— the Predict Missing Values Using AI Model tool trains a machine learning model on existing non-null data to estimate and fill null values automatically⁶. It works primarily on numerical fields; categorical and text fields are handled by the separate Text Analysis toolset. The tool requires sufficient non-null training data— small or imbalanced datasets produce unreliable results.

How does the ArcGIS Arcade assistant work?

Type what you want the field calculation to do in plain English— for example, "set the next inspection date to 3 months from today, on a business day." The assistant writes the Arcade expression⁴; you review and deploy it. It's available in Map Viewer, Field Maps Designer, Scene Viewer, ArcGIS Experience Builder, and more⁵. The tool is in beta as of mid-2026.

Can ChatGPT or Claude write ArcPy scripts for attribute table automation?

Yes. A 2024 ACM peer-reviewed study found ChatGPT-4 achieved an 80.5% task success rate generating ArcPy functions across GIS workflows¹³. Claude (Anthropic) is also reported effective for step-by-step scripting¹⁴. Test AI-generated code in a nonproduction environment before deployment— roughly 1 in 5 tasks will require iteration.

What AI tools does ArcGIS have for classifying feature attributes?

The GeoAI Feature and Tabular Analysis toolset includes Train Using AutoML and Predict Using AutoML⁷— tools that build classification or regression models on labeled attribute data without requiring code or ML expertise. The output is a trained model package that can be reused on future datasets.

Can ArcGIS extract structured data from inspection notes?

Yes— the Text Analysis toolset runs NLP on text attribute fields to extract named entities including dates, addresses, condition descriptions, and asset types⁸. The models are built on production-grade NLP architectures including BERT and Mistral⁹. Note that this toolset requires model selection or training— it's not zero-configuration out of the box.

Closing

The attribute table you started with this morning— 400 features needing reclassification after a CAD handoff— is a half-day job that AI tools now handle differently. Not magically, and not automatically. But the Arcade assistant writes the expression you would have had to write by hand. AutoML extends the classifications you've already done. The Text Analysis toolset pulls structure out of free-text that was previously invisible to analysis.

AI earns its keep in GIS precisely where most hours go: the attribute table. Field calculations, null value estimates, free-text extraction— these are concrete problems with concrete tools. Pick the one that matches the problem you're blocked on right now.

References

1. Fulcrum, "How AI and GIS are shaping the future of AEC" (2025)— https://www.fulcrumapp.com/blog/how-ai-and-gis-are-shaping-the-future-of-aec/
2. Geo Week News, "Fixing the CAD to GIS Handoff" (2024)— https://www.geoweeknews.com/articles/fixing-the-cad-to-gis-handoff/
3. Nearmap, "Modernize property assessment with GeoAI" (2024)— https://www.nearmap.com/blog/modern-property-assessment-geoai
4. Esri, "Mastering field calculations with the Arcade assistant" (2025)— https://www.esri.com/arcgis-blog/products/arcgis-online/geoai/mastering-field-calculations-with-the-ai-arcade-assistant
5. Esri, "What's new in AI assistants (June 2026)" (2026)— https://www.esri.com/arcgis-blog/products/arcgis-online/geoai/whats-new-in-ai-assistants-june-2026
6. Esri, "Predict Missing Values Using AI Model (GeoAI Tools)" (2025)— https://doc.esri.com/en/arcgis-pro/latest/tool-reference/geoai/predict-missing-values-using-ai-model.html
7. Esri, "An overview of the Feature and Tabular Analysis toolset" (2025)— https://doc.esri.com/en/arcgis-pro/latest/tool-reference/geoai/an-overview-of-the-feature-and-tabular-analysis-toolset.html
8. Esri, "An overview of the Text Analysis toolset" (2025)— https://pro.arcgis.com/en/pro-app/latest/tool-reference/geoai/an-overview-of-the-text-analysis-toolset.htm
9. Esri, "An overview of the Text Analysis toolset— NLP models" (2025)— https://pro.arcgis.com/en/pro-app/latest/tool-reference/geoai/an-overview-of-the-text-analysis-toolset.htm
10. Esri, "Esri Collaborates with Microsoft to Bring ArcGIS Users New AI Enhancements" (2025-07-14)— https://www.esri.com/about/newsroom/announcements/esri-collaborates-with-microsoft-to-bring-arcgis-users-new-ai-enhancements
11. Esri, "Esri Collaborates with Microsoft to Bring ArcGIS Users New AI Enhancements— GeoAI pretrained models" (2025-07-14)— https://www.esri.com/about/newsroom/announcements/esri-collaborates-with-microsoft-to-bring-arcgis-users-new-ai-enhancements
12. Esri, "Use third party language models with ArcGIS" (2025)— https://doc.esri.com/en/arcgis-pro/latest/tool-reference/geoai/using-third-party-language-models-with-arcgis.html
13. Association for Computing Machinery, "Automating Geospatial Analysis Workflows Using ChatGPT-4" (2024)— https://dl.acm.org/doi/10.1145/3678717.3695760
14. Geospatial Training Services, "Using AI Assistants to Generate ArcGIS Pro Python Scripts for Automation" (2024)— https://geospatialtraining.com/using-ai-assistants-to-generate-arcgis-pro-python-scripts-for-automation/
15. Esri, "Try the ArcGIS Pro assistant 3.7 beta and get more done in ArcGIS Pro" (2026)— https://www.esri.com/arcgis-blog/products/arcgis-pro/announcements/try-the-arcgis-pro-assistant-3-7-beta-and-get-more-done-in-arcgis-pro