The federal government has a chatbot now, and it runs on two AI models built by companies with very different relationships to the White House. America.gov, unveiled September 29, 2026, pairs Google’s Gemini with xAI’s Grok to answer questions drawn from roughly 29,000 government websites, according to comments U.S. Chief Design Officer Joe Gebbia gave to CNBC. The pitch is simple: one chat box instead of thousands of disconnected .gov pages. The execution, reporters are already finding, is messier than the pitch.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
America.gov Launches as a Single Front Door to 29,000 Sites
America.gov went live publicly on September 29, 2026, positioned by the administration as a consolidated entry point for federal information and, eventually, federal services, according to FedScoop. The site’s chat interface is meant to replace the experience of searching across thousands of individual agency web pages with a single natural-language query box, and GovExec reported in September 2026 that the service requires zero accounts or sign-ins, letting anyone ask a question without creating a profile first. FedScoop described the ambition behind the project as giving Americans “one place for all of their needs from the government.”
President Trump unveiled the site, and Gebbia discussed the technical details in a CNBC interview the same day. The scope is unusually broad for a federal digital project: rather than serving one agency’s internal workforce, America.gov sits in front of information spanning departments as varied as the Social Security Administration, the State Department, and the Department of Health and Human Services, per The Register. Politico reported in September 2026 that the underlying executive order sets a 90-day clock for agencies to integrate with America.gov, giving departments a hard deadline rather than an open-ended rollout window.
How the Chatbot Works, According to Gebbia
Gebbia, the Airbnb co-founder now serving as the administration’s Chief Design Officer, walked CNBC through the mechanics. “You come to America.gov and you ask for what you need, and in a chat interface, we go off on the back end and scan across all those tens of thousands of websites to pull in any information that is specific to whatever your request is, and give you a personalized response that’s only from official sources,” Gebbia told CNBC.
At the launch event, Gebbia framed the retrieval design as the project’s core safeguard against AI models making things up. “Under the hood, America.gov is an extraordinary use of AI designed to improve your life. It’s trained specifically on one official source of truth: The United States Government,” he said, according to FedScoop. That grounding claim is the crux of the administration’s pitch: the models aren’t answering from general training data, they’re retrieving and summarizing content pulled live from .gov sources. PBS’s own September 2026 coverage described the same sourcing model, reporting that every response is drawn exclusively from official federal sources rather than open web content the underlying models might otherwise pull from.
The examples officials cited span mundane but high-traffic tasks: reserving a national park campsite, replacing a Social Security card, checking Medicare eligibility, applying for a child’s passport, or searching federal job listings, per FedScoop and Nextgov’s coverage of the launch.
By the Numbers: 29,000 Sites, 40 Million Daily Visits
The scale claims are the most concrete figures to come out of the launch. America.gov is described as drawing on roughly 29,000 government websites, a number repeated in both CNBC’s and FedScoop’s coverage. Separately, Gebbia said government websites collectively receive about 40 million visits a day, a figure reported by Nextgov that underscores why the administration is betting on a single AI front end rather than a sitewide redesign of every individual agency property.
What’s missing from the public record so far is just as notable. No outlet covering the launch has published a development cost, an annual operating budget, or contract values tied to Google or xAI’s involvement. There’s no confirmed user count from launch day and no adoption target, though GovExec reported in September 2026 that a phase-two rollout is now targeted for early 2027, with benefit transactions, not passport renewals, named as the next capability expected to ship, a timeline that also lines up with the 2027 window cited separately by the Washington Sun.
America.gov Launch Facts at a Glance
| Detail | What’s reported | Source |
|---|---|---|
| Launch date | September 29, 2026 | FedScoop, CNBC |
| AI models named | Google Gemini and xAI Grok | CNBC (Gebbia) |
| Sites indexed | ~29,000 government websites | CNBC, FedScoop |
| Daily traffic baseline | ~40 million visits/day across .gov sites | Nextgov |
| Lead official | Joe Gebbia, U.S. Chief Design Officer | CNBC, FedScoop |
| Agencies involved | National Design Studio (design/build), GSA (operations/support) | FedScoop, Nextgov |
| Transactional features | Not yet live; reported target is 2027 | Washington Sun |
| Disclosed cost or contract value | Not disclosed in any launch-day report | CNBC, FedScoop, Nextgov |
Who Built It: National Design Studio and GSA
Gebbia leads the administration’s National Design Studio, the unit credited with designing and engineering America.gov, while the General Services Administration is reported to run or support the site’s operations, according to FedScoop and Nextgov. That split, a White House design unit driving the product and GSA providing the infrastructure, mirrors how past federal tech modernization pushes have typically been staffed, though the cross-agency scope here is unusually wide.
Google’s Formal Role as Technology Partner
Google’s involvement goes beyond supplying a model. A Google spokesperson, Melanie Gonzalez, framed the company’s role as a formal partnership: “Google is proud to be named a technology partner in this vital initiative, leveraging Gemini to help more than 100 million people access critical public resources with greater speed and ease,” Gonzalez said, according to FedScoop. That 100-million figure is the first concrete adoption target tied to the project, though it reads as an aspiration rather than a measured result this early.
The Model Roster Discrepancy: Grok, Gemini, and Maybe OpenAI
Two Accounts That Don’t Match
Here’s where the reporting gets messy. Gebbia’s account to CNBC named exactly two models: Gemini and Grok. But a separate account of the launch event, carried by the Washington Sun, quoted administration official Michael Oz describing the system as running on “a couple” of AI systems and specifically citing xAI’s Grok alongside OpenAI, not Gemini. Al Jazeera’s coverage lines up with Gebbia’s version, reporting plainly that “the site is powered by Google’s Gemini AI and xAI’s Grok.”
No outlet has reconciled the two accounts. It’s possible OpenAI is involved in a pilot capacity, a broader technical role, or was simply misstated in one report. It’s also possible Oz was speaking about a different, earlier phase of development before Gemini and Grok were finalized as the production models. Anthropic and Microsoft, both major federal AI and cloud vendors elsewhere in government, have no confirmed role in America.gov in any of the coverage reviewed here.
The pairing of Grok and Gemini is politically notable regardless of whether OpenAI turns out to have a role. Grok’s developer, xAI, was founded by Elon Musk, who previously led the Department of Government Efficiency inside the same administration now running America.gov. Readers following the broader pattern of AI agents interacting with federal systems this year may recall OpenAI’s own agents reaching Census Bureau data using leaked credentials, a separate incident that has kept scrutiny of AI-and-government pairings high all year.
Historical Context: From GSA USAi to a Public-Facing Portal
America.gov is not the government’s first attempt at putting generative AI in front of either workers or the public. The GSA has previously promoted a governmentwide AI capability known as USAi, intended to give federal employees access to approved generative-AI tools on shared infrastructure. Individual agencies have also run narrower pilots: document analysis, internal customer service, code assistance, fraud detection, and records processing have all seen agency-specific generative AI trials over the past two years.
What sets America.gov apart is scope. Past efforts were largely internal-facing, aimed at federal employees rather than citizens, or scoped to a single department. America.gov is public-facing by design and spans the entire federal web presence rather than one agency’s slice of it, according to FedScoop and Nextgov’s framing of the launch. That makes it the most ambitious public rollout of consumer-facing generative AI the U.S. government has attempted.
It also arrives in a year when Washington has spent as much energy scrutinizing AI as deploying it. Rep. Maxine Waters pushed for a federal probe and a temporary AI moratorium just days before this launch, and Australia separately summoned OpenAI and Anthropic executives over AI safety concerns. Launching a government-run, public-facing chatbot into that climate is itself a statement of confidence, or at minimum a gamble.
Federal AI Tools Compared
| Initiative | Audience | Scope | Models involved |
|---|---|---|---|
| America.gov | General public | ~29,000 sites, cross-government | Gemini, Grok (per Gebbia) |
| GSA USAi | Federal employees | Governmentwide internal tool access | Not specified in America.gov coverage |
| Agency-specific pilots | Individual agency staff or users | Single department or function | Varies by agency |
| Legacy .gov search/chat tools | General public, single agency | One department’s site only | Pre-generative-AI search systems |
Market Impact: What This Means for Google, xAI, and Rivals
For Google, landing an on-the-record “technology partner” credit on a flagship federal project is a meaningful reference customer, regardless of contract size. Government work carries a credibility signal that spreads to other public-sector and enterprise sales conversations, and Gonzalez’s statement was explicit about reach: over 100 million people are the target audience Google is citing. Readers tracking Google’s broader AI roadmap should note this lands just as Gemini 4 nears its own launch, giving the company a live government deployment to point to right as it pitches a next-generation model.
For xAI, the win is arguably bigger relative to the company’s size and market position. Grok has trailed rivals on independent benchmarks, and xAI’s Grok 4.7 release still lags Claude on several benchmark comparisons, according to prior reporting. A federal government deployment alongside Gemini gives Grok a marquee use case that has nothing to do with chatbot leaderboard rankings and everything to do with political proximity, given Musk’s prior DOGE role in this same administration.
OpenAI and Anthropic, meanwhile, are notably absent from the confirmed model roster, at least according to Gebbia’s own account. That’s a contrast worth sitting with: OpenAI has been positioning itself aggressively around government and enterprise AI work all year, while Anthropic has leaned into safety-focused federal partnerships. Neither shows up here. Whether that’s a deliberate choice, a procurement outcome, or simply unreported is one of the bigger open questions the launch coverage leaves unanswered.
The Transparency Gap Reporters Flagged
Not every outlet covering the launch came away convinced the rollout matched the administration’s framing. Axios reported that America.gov does not prominently identify which AI models power it on the site itself, even though Gebbia separately named Gemini and Grok directly to CNBC. That’s a meaningful gap for a public tool handling questions about passports, benefits, and federal employment: users interacting with the chatbot have no clear, on-site way to know which company’s model is generating their answer.
The Register was more pointed in its assessment, writing that “what you have now on America.gov, then, is just a federal government chatbot powered by Gemini and Grok that refers users to other websites, helpful, but not exactly the central portal Trump promised in his EO.” The gap between the executive order’s ambition, a unified government services portal, and the launch-day reality, an information retrieval chatbot, is the throughline in nearly every piece of skeptical coverage published so far.
Security, Privacy, and Accuracy Risks Nobody Has Answered
What Retrieval-Based Design Does and Doesn’t Fix
A chatbot fielding questions about Social Security numbers, Medicare eligibility, and passport applications is, by definition, handling sensitive personal context even before a user submits a form. Engadget reported in September 2026 that America.gov’s privacy policy commits to retaining zero search histories or messages, though the launch-day coverage reviewed here still doesn’t spell out whether prompts get used to improve or train the underlying commercial models, or what logging practices apply before that no-retention policy kicks in. That gap matters more here than it would for a typical consumer chatbot, because the user base is, in practice, the entire American public interacting with federal services.
Accuracy is the second open risk. Even with retrieval grounded in official sources, as Gebbia described it, federal rules frequently conflict across agencies or change faster than any indexing pipeline can track. None of the coverage explains how America.gov resolves contradictory guidance between agencies, or what recourse a user has if the chatbot gives a wrong answer about, say, benefit eligibility. This year has already produced a steady drumbeat of AI-agent-related federal security incidents, including the congressional push for an OpenAI probe and AI moratorium and separate reporting on a SharePoint vulnerability under an active CISA patching deadline, both reminders that federal IT security scrutiny is already elevated heading into this launch.
Security researchers have also spent much of 2026 documenting how commercial AI models can be manipulated through prompt injection, an attack class where malicious instructions hidden in retrieved content hijack a model’s output. A system designed to scan tens of thousands of external web pages and synthesize their content in real time is, structurally, exposed to exactly that category of risk, though no incident specific to America.gov has been reported as of this writing.
The DOGE Workforce Backdrop
America.gov did not launch in a vacuum. Nextgov and the Washington Sun both tie the project to assistance from former Department of Government Efficiency personnel, situating the chatbot within the administration’s broader push to consolidate federal operations and, in some cases, shrink the federal workforce. Neither outlet reports a specific job-loss figure tied to America.gov itself, and no federal union has issued a public statement about the project in the coverage reviewed here.
Still, the optics are hard to separate from the technology. A single AI chatbot capable of answering questions that previously required navigating individual agency websites, or calling individual agency help desks, arrives at the same moment DOGE-linked consolidation efforts have already reshaped parts of the federal workforce. Whether America.gov reduces demand for human-staffed government customer service, or simply supplements it, is not something any of the launch coverage addresses directly.
What’s Next: Transactions, 2027, and Expansion Plans
The version of America.gov that launched September 29 is, by its own designers’ description, an information layer rather than a transaction layer. Users can ask how to replace a Social Security card or reserve a campsite, but they still complete those actions on the original agency website America.gov points them toward. Broader functionality, actually completing a passport renewal or a Medicare enrollment inside the chatbot itself, is reportedly not expected until 2027, per the Washington Sun’s coverage of the launch remarks.
That phased approach is a reasonable engineering choice for a project this large: retrieval and summarization carry far lower stakes than letting an AI system initiate or complete a binding government transaction on a user’s behalf. But it also means the “front door” framing the administration has used since the executive order is, for now, a front door that still redirects most visitors to the same 29,000 side doors America.gov was meant to replace.
Predictions: Where America.gov Goes From Here
- Model disclosure improves under pressure. Given Axios’s reporting on the missing on-site model attribution, expect the administration to add clearer model disclosure to the site within weeks rather than months, simply to blunt the “hidden AI” criticism.
- The OpenAI discrepancy gets resolved or drops. Either a follow-up report confirms OpenAI’s role in some capacity, or the Michael Oz comment quietly becomes a footnote nobody revisits. A months-long unresolved contradiction about which models power a flagship government tool is unlikely to stand.
- Congressional scrutiny follows. Given existing momentum toward AI oversight hearings this year, expect at least one congressional inquiry or oversight letter specifically about America.gov’s data handling within the next two quarters.
- 2027 transactional rollout slips. Government technology timelines for transactional features, especially ones touching identity verification like passports and Social Security, have a well-documented history of slipping. A 2027 target announced at launch is more likely to become late 2027 or 2028 in practice.
- Other agencies route around it. Individual agencies with strong existing digital service teams may continue building or maintaining their own AI tools in parallel rather than fully deferring to America.gov, mirroring how GSA USAi coexists with agency-specific pilots today.
Frequently Asked Questions
What is America.gov?
America.gov is a new federal government website and AI chatbot, launched September 29, 2026, designed to help users find information across roughly 29,000 government websites through a single chat interface, according to CNBC and FedScoop.
Which AI models power America.gov?
U.S. Chief Design Officer Joe Gebbia told CNBC the site uses Google’s Gemini and xAI’s Grok. A separate account of launch remarks referenced OpenAI instead of Gemini, a discrepancy that has not been publicly resolved.
Who built America.gov?
The site was designed and engineered by the administration’s National Design Studio, led by Joe Gebbia, with operational support reported from the General Services Administration (GSA).
Can I complete transactions like passport renewals on America.gov?
Not yet. The current version focuses on information retrieval and points users to the relevant agency site to complete actions. Transactional features are reportedly targeted for 2027.
Is America.gov free to use?
Yes. As a public government website, America.gov carries no user cost, and GovExec reported in September 2026 that the service runs ad-free as well. Development and operating costs have not been publicly disclosed by any agency or outlet covering the launch.
Does America.gov show which AI model answered my question?
According to Axios’s reporting, the site does not prominently disclose which model is generating a given response, despite officials naming Gemini and Grok in separate interviews.
Is my data safe when I use America.gov?
Engadget reported in September 2026 that America.gov’s privacy policy retains zero search histories or messages, though GSA and the National Design Studio haven’t publicly detailed a model-training policy or query-logging practices beyond that as of this launch.
How is America.gov different from GSA’s USAi tool?
USAi is aimed at giving federal employees access to approved AI tools internally. America.gov is public-facing, built for citizens searching government information, and spans the entire federal web presence rather than internal agency workflows.
Related Coverage
- ChatGPT vs Claude vs Gemini vs Grok: $292 Plan Gap [2026]
- OpenAI Dots Debuts at $500/Month, Undercuts Muse Free Tier [2026]
- Sonnet 5.5 vs GPT-6 Luna vs Gemini 3.8: 20x Gap [2026]
- Googlebook Cheat Sheet: Price, Models, Gemini Features [2026]
- Grok 4.7 API Setup: 12 Steps, $2/$6 per 1M Tokens [2026]
