State, local website improvements crucial as more residents use AI assistants to get important info, report says
As more residents turn to AI assistants for information about public services, state and local governments must ensure their websites provide accurate, accessible information those systems can reliably use, according to a new report.
The report, Making Government a Trusted Source in AI System released Wednesday by Georgetown University’s Beeck Center for Social Impact and Innovation, says that because residents now rely on summaries provided by ChatGPT, Claude or Gemini, governments should make their official information easier for AI crawlers to ingest and relay to to users.
It identities three factors that governments can control to ensure their digital and web assets are compatible with the automated crawlers: access, making sure the AI crawlers can reach public-facing government content; accuracy, ensuring information on government websites is current and correct; and structure, organizing the content so AI systems can reliably extract and relay it.
Deficiencies in these areas, the report claims, can make it less likely that an assistant retrieves, interprets or cites the government’s information correctly, and therefore probable that citizens are not getting the correct or up-to-date information they may need. Governments should address all three topics to make official information more transparent, the report says.
To evaluate access, governments should review their website security settings to ensure legitimate AI crawlers can reach public information without compromising security or overwhelming servers, the report said. Common security and bot-management tools used by governments may unintentionally block AI systems from accessing public information. To mitigate those risks, the report recommends a tiered approach that allows verified crawlers from major AI providers to access, evaluates others individually, and applies graduated restrictions to unverified bots.
The report illustrates how website design can affect the accuracy of AI-generated answers. In one case, the researchers discovered that OpenAI’s ChatGPT provided a figure for Mendocino County, California’s public safety spending that was 53% higher than the county’s own budget tool reported. After evaluation, the report found that the county’s JavaScript-powered interactive dashboard may have made the underlying figures harder for AI systems to retrieve reliably.
By comparison, a Larimer County, Colorado webpage that had straightforward information was accurately extracted and relayed by ChatGPT about special-event permit timelines.
To improve the three factors — access, accuracy and structure — the report recommends that governments assign clear ownership over the improvements, take concrete and corrective actions, and then verify that the change is working. Any corrective steps should be revisited annually to keep site and content access, accuracy and structure up to date.