New AI Case Study and Tribal Broadband Data Resources

Today we released three new resources for those studying broadband infrastructure on Tribal lands: a case study on the use of large language models to generate descriptive content entitled, “Caveat Promptor: A Case Study of Tribal Land Descriptions,” an interactive site with tribal land data at tribalgeo.com, and the “2026 TPI Tribal Lands Broadband Report: Broadband Infrastructure in Tribal Areas (2015-2025).” Together, these resources provide broadband data and demographic context for 706 Tribal lands across 38 states and territories, and examine the quality of AI-generated text describing them.

KEY HIGHLIGHTS

• Case study documenting how structured prompting can redirect large language models away from biased cultural framing toward task-relevant description
• 706 Tribal land profiles covering federal and state reservations, off-reservation trust lands, Oklahoma and state tribal statistical areas, Alaska Native Corporation and village areas, and Hawaiian homelands. 38 state directories with broadband coverage data spanning 2015–2025 and FCC filings current as of June 2025. Integrated data sources combining Census, FCC, speed data, and boundary files in a single report-driven reference. Per-land profilesincluding federal AIANNHCE identifiers, land descriptions, quarterly speed trends, availability, adoption by income, and infrastructure status

TPI Senior Fellow Sarah Oh Lam explains that the project grew out of the opportunity to use AI to improve visibility into Tribal lands and their broadband profiles. “We set out to harness the power of AI to bring greater visibility to Tribal lands and the state of broadband on them,” said Sarah Oh Lam, Senior Fellow at the Technology Policy Institute. “With so much potential in these areas in broadband buildout, spectrum allocation, data center siting, energy infrastructure, and AI deployment, researchers and policymakers benefit from having visibility on Tribal lands, and we hope these resources contribute to their understanding.”

Here are the three resources:

1) “Caveat Promptor: A Case Study of Tribal Land Descriptions”

The case study, “Caveat Promptor: A Case Study of Tribal Land Descriptions,” examines how large language models synthesize knowledge for policy planning purposes and asks what happens when default answers are offensive. The paper presents a case study of Tribal content in which AI models produced text that appears fluent and confident but may be thematically skewed toward biased cultural framing. The authors, Sarah Oh Lam and Annella Charee Tucker, observed this problem while drafting the 2026 TPI Tribal Lands Broadband Report. Prompted without detailed guidance, models produced outputs organized around heritage, spirituality, and tradition rather than the geographic and infrastructure characteristics needed for broadband planning. The paper argues that this output poses a practical risk for any domain in which AI-generated text may be applied without careful review and demonstrates that structured prompting can redirect outputs toward task relevance. The problem, while real, is largely addressable at the prompt level.

The authors conclude that implications are broader than this case study: “These lessons are not limited to Tribal lands or to broadband planning. The same pattern of fluent, confident text organized around the wrong framing, is likely to appear wherever AI is applied to domains where available source material is narrow, or dominated by outsider perspectives, or skewed toward a particular genre. Rural communities, small nations, specialized technical fields, minority language groups, and any subject where the written record is thin relative to the complexity of the subject are all candidates for the same kind of default. As AI-generated text becomes embedded in government reports, grant applications, and planning documents, the risk is not that these outputs will be obviously wrong. The risk is that they will be plausible enough to go unquestioned.” Download the article here.

2) TribalGeo.com

The interactive site, tribalgeo.com, presents Tribal land profiles for search and direct discovery. Visitors can browse the 706 Tribal land pages, navigate 38 state directories, review source documentation, and subscribe to a newsletter for future updates. The site enables researchers and policymakers to locate broadband performance and demographic information for specific Tribal lands without needing to reconstruct the underlying data pipeline themselves. Visit tribalgeo.com to see more.

3) 2026 TPI Tribal Lands Broadband Report

The “2026 TPI Tribal Lands Broadband Report: Broadband Infrastructure in Tribal Areas (2015-2025)” is a databook that compiles broadband data across lands falling within the U.S. Census Bureau’s American Indian, Alaska Native, and Native Hawaiian Area (AIANNH) definition, including federal and state reservations, off-reservation trust lands, Oklahoma Tribal statistical areas, state Tribal statistical areas, Alaska Native Corporation and village statistical areas, and Hawaiian homelands. The 3,590-page report, generated with data compiled by Sarah Oh Lam, Nathaniel Lovin, and Annella Charee Tucker, presents broadband data across 706 Tribal land profiles in 38 states and territories, with a broadband coverage window spanning 2015 through 2025 and the latest FCC filing data current as of June 2025. Download the report here.

For more information about the TPI Tribal Lands project, please contact Sarah Oh Lam at [email protected].