The SNAP program is used by tens of millions of U.S. citizens who need nutrition assistance. In a test by independent AI evaluator, VALS, the most commonly used free AI models from Grok, ChatGPT, Claude, and Gemini returned accurate information on SNAP about 50% of the time. Getting the wrong information means people who need SNAP benefits might not have the right information for proof of work or miss a recertification deadline.
This is the new reality for anyone who needs government information, from Library of Congress research to NASA mission news. Whether federal agencies realize it or not, people increasingly turn to AI for answers.
Most people are using the freely available versions, and the AI responses don’t consistently use official sources and are frequently wrong. In a February 2026 benchmark, the Open Data Institute tested more than 22,000 citizen questions against gov.uk content. Models frequently failed to prioritize or cite official sources, or to admit when they didn’t know the answer.
For federal agencies, accuracy isn’t optional. The 21st Century Integrated Digital Experience Act (IDEA) already requires agency websites to be accurate, searchable, and free of duplicate or outdated pages. OMB memo M-23-22 tells agencies to keep content current and consolidate duplicate pages. That same work is what helps keep AI answers accurate, both on an agency’s own site and everywhere else.
Call it generative engine optimization or answer engine optimization; if you want accurate information to be found by AI tools, your approach to search needs to change. For civil servants, GEO means adapting to a world where AI-generated overviews answer citizens’ questions first.
Learn how to optimize your government website to ensure the correct information is displayed.
Why generative engine optimization means something different for government
For brands, GEO is a visibility contest similar to SEO because showing up in AI answers ahead of competitors translates directly to more leads and sales. But no one is competing with the GSA to publish per diem rates.
The competition for federal agencies is information aggregators, subreddits, outdated pages on their own websites, and outright misinformation. When AI picks one of those sources as authoritative, the public gets the wrong answer, often without any indication that the information is bad. For agencies, GEO is about maintaining public trust and providing equal access to accurate, public information.
The good news is that agencies start from a strong position because most AI models already view government websites as authoritative. “The challenge is making the content understandable, singular, and accessible,” says Jenna Westbrook, Public Sector Website Strategist at WordPress VIP.
Models trust .gov content when they can find it and parse it. Maintaining accuracy becomes difficult when content is scattered across sites, duplicated across pages, or structured in a way that’s hard for AI to parse.
Where AI gets government information wrong
When AI misrepresents government information, the root cause can be traced to a handful of common content issues that have plagued information management since agencies started publishing content online.
- Duplicate or conflicting pages: At GSA’s Federal AI Hackathon, one example illustrated how AI might return the wrong phone number for Social Security, because the agency publishes multiple numbers in different places on its website. When pages have conflicting information, AI picks one answer, which might be the wrong one.
- Information locked in PDFs or scanned documents: When data and guidance are published in PDFs instead of web pages, it’s more difficult for AI tools to parse the information because structured markup like HTML is expected.
- Stale pages: When pages with outdated information compete with current ones while AI is reasoning over a site, there’s a risk that AI will repeat the outdated answer.
- Data tables that lose context: Tables lacking clear headers or captions become a string of meaningless numbers when read as raw HTML.
- Sprawl across bureaus and programs: When an agency has multiple sites all running on separate hosting platforms, there’s an increased chance that information will be inconsistent across the sites.
They’re content governance problems that existed before AI. Now they’re just more visible.
What agencies can do today to improve accuracy in AI search
Improving AI accuracy starts with auditing what the answers say right now. Make each answer singular by creating one source of truth that stays up to date. Make it understandable using plain language and machine-readable structures. And be sure AI can access that source of truth so that the correct answer wins when someone asks a question.
Audit what AI already says about you
Which 20 questions is the public asking your agency most often? Sourcing them from your site analytics and call center data is a good starting point. Run each question through ChatGPT, Gemini, Copilot, Claude, and Perplexity. Log whether the answer was right and which source was cited, then repeat the check quarterly and after any major changes to your content catalog. This audit is a good candidate for automation using an AI agent, but can also be run manually.
Make one page the source of truth
Consolidating duplicate pages, while also redirecting retired pages, reduces the number of places information gets repeated and makes progress toward compliance with 21st Century IDEA.
Write answer-first, plain-language content
Lead with the answer, then the detail. The Plain Writing Act already requires federal agencies to write clearly. That same clarity helps AI extract the right sentence from a block of text. Descriptive headings and short question-and-answer blocks help both AI and human readers find answers quickly.
Add structure AI can read
Clear headings in text and tables paired with visible dates create understandable page structures. Schema markup adds additional structure that gives AI more information.
GovernmentOrganization, FAQPage, Dataset, NewsArticle, and DateModified markup all have their place on agency pages. FAQPage was deprecated as a signal for rich results on Google in May 2026, but still has relevance for AI crawlers and other search engines. Frequently downloaded PDFs should be converted to HTML, which allows AI to parse the information.
WordPress VIP can help you import millions of PDFs automatically.
And any effort you make toward meeting the WCAG 2.0 AA guidelines required for Section 508 compliance has the add-on effect of improving AI-readability.
Decide on AI crawler access deliberately
Blocking AI crawlers doesn’t stop AI from answering questions about your agency, but it will mean the answer comes from somewhere else. For public information, it’s better to make pages easy for reputable AI crawlers to access so people get answers sourced from the authority on the topic. AnswerLine, a tool created by WordPress VIP, helps you see which crawlers have access and what’s visible.
That doesn’t mean you need to open the floodgates to all bot traffic. On WordPress VIP, agencies can manage bot traffic using Edge Workers, which keeps public pages open while protecting things like private user sessions.
Keep the site fast and available
The same effort you put into making your site faster for users also matters for AI. Infrastructure designed to scale during traffic spikes keeps your authoritative answers available during high-demand events like public emergencies, monthly and quarterly data releases, and any major announcements.
When economists are all downloading the latest jobs report from the Bureau of Labor Statistics, AI needs access to that information too, or it will switch to referencing someone else’s summary of the data.
Measure AI visibility and referrals
Track AI referral traffic and crawler activity alongside your quarterly prompt audit. Public sector customers on WordPress VIP are already seeing AI referral traffic. Tools like Parse.ly, with its Search Console integration, separate AI referral from search, which shows the pages getting traffic from AI.
Keep content fresh and trusted
Freshness is hard for agencies that make updates to public information on a fixed schedule. Using machine-readable “last updated” dates with accurate XML sitemap lastmod values helps AI interpret freshness. Updating existing pages instead of publishing new copies at new URLs avoids resetting page authority.
For self-audits, rerun the AI prompt checks after major content changes. Search Console and server logs can help you understand which pages are fetched by AI crawlers and when they fetched them.
Publish where AI looks, but keep the website authoritative
AI tools draw on Reddit and other popular community sites in addition to official sites. Having official agency accounts on those platforms can help, but every copy of information shared directly somewhere else is one more thing to audit when the source of truth changes. A short answer with a link to the canonical page helps prevent more duplicates.
Some large agencies are considering AI-powered search on their own sites. That’s a useful modernization step, but it doesn’t change the reality that most users will ask ChatGPT or Google without ever visiting the agency site.
AI-powered search requires one accurate, well-structured, current page per answer regardless of who hosts the search. Get your content catalog in order so any answer engine, yours or someone else’s, has the right source to draw from.
Protect content and image authenticity
AI tools and website visitors both need to trace content back to its source. Using a consistent .gov domain and clear attribution on data and images makes agency content easier to verify when someone wants to know whether something is authentic or AI-generated. As Content Credentials (C2PA) become more widely adopted, it will be possible to consistently attach verifiable origin data to images and media.
Accuracy is a governance problem, not a plugin
Adding AI search to a government site won’t fix duplicate or stale content. Accuracy still requires the same publishing rigor that takes contributor input and runs it through an approval process that determines what goes live and when.
“All the things a site owner can do to make the site more accessible and usable to humans will help them be visible and legible to AI bots,” says Westbrook. “Like accessibility, there’s always room for improvement.”
This is where your content management system helps most. Structured content models keep facts in the fields they belong to, published on the right pages. Editorial workflows and approvals decide who can publish updates to official data and when those updates are allowed. Revision history creates an audit trail that shows what changed and who changed it. Approaching governance as a multisite effort keeps bureau and program sites consistent and helps reduce duplication of information.
Security still matters. Opening public information to AI crawlers shouldn’t mean private data is exposed too. WordPress VIP is currently the only FedRAMP Moderate authorized WordPress platform, and also holds GovRAMP and TX-RAMP certifications for state agencies.
Organizations already see the GEO impact of using WordPress VIP. Pew Research Center went from almost zero ChatGPT referral traffic to ChatGPT as its second-largest referrer 30 days later. NASA migrated 126,171 pages onto WordPress VIP, the kind of scale where governance must work across the whole site.
Give AI the right answer to repeat
Back to the user searching for SNAP benefit information. This time, they find the exact requirements to prove eligibility in their community. Claude Haiku summarizes the process and refers them to the state website relevant to their benefits.
Most agency sites start in good shape. Taking three specific steps this quarter will make them better:
- Run your agency’s top 20 most-asked questions through the major AI tools and log the outcome so you can fix anything that’s wrong.
- Consolidate duplicate and conflicting pages that the audit uncovers.
- Convert high-traffic PDFs to plain-language HTML with machine-readable update dates.
When an agency doesn’t publish the clearest answer, it doesn’t disappear from AI search. Someone else’s answer takes its place.
See how your agency’s content looks to AI.
Frequently asked questions
What is generative engine optimization for government?
Generative engine optimization (GEO) for government is the practice of making agency content easy for AI tools to find, understand, and repeat accurately. The goal isn’t generating more leads or sales like GEO for corporate entities; it’s making sure the public gets a correct, current answer from the official source.
Why do AI chatbots give wrong answers about government information?
Usually because the official answer is stored in a way that makes it hard to find or is contradicted by another source. Duplicate pages, outdated information, and data that can’t be extracted from PDFs give AI tools a wrong version to repeat.
Should government websites block AI crawlers?
Blocking AI crawlers doesn’t stop AI from answering questions about an agency, so if information is public, it probably shouldn’t be blocked from crawlers. A block on AI means it will look for a different source to generate an answer. That source won’t be the authority. Bot management tools can help protect everything that needs to stay private, without putting a blanket block on the whole site.
Do government websites need an llms.txt file?
No. Google doesn’t require llms.txt for its AI features, and support for llms.txt varies widely across the broader AI tool space. It’s an optional, low-effort way to point AI tools toward an agency’s most important pages, as long as they get updated consistently.
How can an agency check what AI tools say about it?
Identify the 20 questions the public asks most frequently and test them in ChatGPT, Gemini, Copilot, Claude, and Perplexity. Record whether the answer was accurate and what source was cited. Repeat this process every three months and after any major content changes.
How can agencies keep AI answers current when content changes?
Update pages instead of publishing new versions and stamp them with a machine-readable last-updated date. Be sure sitemap lastmod values are accurate too. Redirect any page you deprecate to the authoritative replacement. Test your updates by prompting AI answer engines and verifying that the new information appears in the responses.
How can agencies protect the authenticity of their content and images?
Use Content Credentials (C2PA) for images and media assets, publish on a consistent official .gov domain, and add proper attribution to data and images. All three steps make it easier for AI tools and readers to know content came from the agency and not somewhere else.
