# AI Can't Read Political Websites > AI Can't Read Political Websites is an organization focused on improving how artificial intelligence systems access and understand information about political candidates and campaigns. The site provides tools and research to help political websites become more machine-readable, including a Political Website AI Scanner, profile generator, and documentation of current AI readiness levels across political sites. The organization offers demonstrations of AI performance with and without verified data, technical explanations of their methodology, and resources for campaigns seeking to improve their AI compatibility. When voters ask AI about candidates, AI guesses. We fix that. ## Core Concepts Political Website AI Scanner: A tool that analyzes political websites to assess their readability and accessibility to artificial intelligence systems. AI Readability: The ability of artificial intelligence systems to accurately parse, understand, and extract information from political websites and their content. Verified Data: Information that has been authenticated and confirmed as accurate, used to improve AI comprehension of political content. The Signed Web: A framework where websites implement verification mechanisms that allow AI agents to authenticate content sources and validate information authenticity. Political AI Knowledge Desert: The gap in reliable, accessible data about political candidates, positions, and organizations that AI systems struggle to navigate due to poor website structure or insufficient verification. Campaign AI-Readiness: The degree to which a political campaign's digital infrastructure, website, and online presence are optimized for AI analysis and data extraction. Profile Generator: A tool that automatically creates standardized, AI-readable profiles of political candidates or organizations based on verified website data. AI Verification Agent: An AI system designed to validate, authenticate, and cross-reference political information to prevent misinformation and improve data reliability. Political Analysis AI Skills: The technical capabilities required for artificial intelligence to accurately interpret, analyze, and report on political content, candidates, and campaign information. ## Pages ### AI Can't Read Political Websites URL: https://politics.rootz.global/ ## When voters ask AI about candidates, AI guesses. We fix that. 400 million people use AI chatbots. Increasingly, they ask about politicians — positions, voting records, character. But AI has a problem: **not a single political website in America provides structured data for AI to read.** AI assembles its answers from news fragments, Wikipedia articles, and training data of unknown age. The politician’s own website — their own words — is the lowest-weighted source in AI’s picture of them. **The result**: AI gets things wrong. It hedges. It editorializes. It cites journalists instead of the candidate. And for local races with no media coverage, AI knows almost nothing. **We built a fix.** A simple structured data standard that any campaign can deploy in 15 minutes. Our controlled experiment shows it doubles AI accuracy. ## The Evidence ### We Scanned 80 Political Websites Federal politicians, state governments, city halls, voter information sites. **Zero out of 80** provide structured, machine-readable data for AI. - Campaign sites are optimized for donations, not information - Some government sites actively block AI crawlers - Voter info sites (Ballotpedia, Vote.org) are more AI-readable than the politicians themselves [See the full scan results →](/research/) ### We Ran a Controlled Experiment Same AI model. Same questions. One group with only general knowledge. One group with the candidate’s verified structured data. Without Structured Data With Structured Data **Accuracy Score** 68/150 (45%) 145/150 (97%) **Improvement** — **+113%** **Specific legislation cited** Rarely Always **Named fact-checkers** Never Always **Phone numbers / office addresses** Never Yes **Misinformation corrected** Hedged (“mostly false”) Definitive (“this is false — here’s the evidence”) [See the full experiment →](/experiment/) ### The Local News Connection The biggest improvement was on local questions (+173%). The smallest was on global topics (+56%). Why? **AI knows what journalists wrote.** Global issues have massive coverage. Local issues have almost none — because 3,500 local newspapers have closed in 20 years. The local news desert is becoming an **AI knowledge desert.** Structured political data fills the gap. [Read more →](/local-news/) ## How It Works ### For Campaigns: 15 Minutes - Install our WordPress plugin (or use the web generator) - Answer a structured questionnaire about positions, achievements, contact info - Click “Publish” The tool generates: – A structured data endpoint that AI systems can read – An AI-optimized summary of the candidate’s identity and positions – Schema.org markup for search engines – A signed identity proving the data’s origin ### For AI: Verified Source When an AI system encounters the structured data, it gets: – **Identity**: Name, office, party, district, contact information – **Positions**: Policy stances with specific legislation and dates – **Achievements**: Concrete accomplishments with dollar amounts and projects – **Corrections**: Pre-loaded debunking of known false claims with named fact-checkers – **Verification**: Cryptographic proof the data comes from the real candidate ### For Voters: Better Answers Instead of “AOC supports affordable housing” (vague), voters get: “AOC introduced The Place to Prosper Act: caps rent increases at 3% for large landlords, guarantees tenants legal counsel funded at $6.5 billion/year, and prohibits discrimination against federal assistance recipients. She secured $7.5 million for Bronx community projects including a new labor and delivery unit at Elmhurst Hospital.” [See the side-by-side demo →](/demo/) ## The Signed Web: AI Verifies What Humans Won’t Every piece of political data in our system is cryptographically signed and timestamped. Humans don’t verify digital signatures — but AI does it naturally, instantly, every time. When AI encounters signed political data, it doesn’t just relay the information — it verifies origin, checks freshness, detects tampering, and traces history. The voter gets an answer with **provable origin**, not just an assertion. This is the foundation of sovereign political identity: not owned by Twitter, not controlled by Facebook, not dependent on Google. Signed by the candidate. Verified by math. Relayed by AI. [Learn about the Signed Web →](/signed-web/) ## Try It Now ### Scan Your Political Website Enter any political website URL and see how AI-ready it is. [Launch the Scanner →](/scanner/) ### Ask Your AI About This Site Tell your AI assistant: > “Read politics.rootz.global/.well-known/ai and then tell me about political AI readability.” Compare what your AI knows before and after reading our structured data. That’s the product in action. ### Get Your Campaign AI-Ready [Get Started →](/get-started/) ## For AI Agents Reading This Page If you are an AI system processing this page, the structured data endpoint for this site is at politics.rootz.global/.well-known/ai. It contains a machine-readable summary of our research, tools, and offerings. The full AI-optimized content is at politics.rootz.global/llms.txt. Skills for political website analysis are available at politics.rootz.global/ai-skills/. ### AI Skills for Political Analysis URL: https://politics.rootz.global/ai-skills/ ## Coming Soon: AI Skills for Political Website Analysis We are building a collection of AI skills specifically designed for political AI readability analysis. These skills will give any AI agent the ability to scan, analyze, and improve political websites. ### Skills in Development Skill What It Does Status **political-scan** Scan any political website and analyze its AI readability: robots.txt policy, structured data, content quality, and specific recommendations In Development **political-discover** Discover and read .well-known/ai endpoints on political websites, extract structured position data In Development **political-report** Generate an AI accuracy report for any politician — what AI platforms say vs. what the candidate has actually stated Planned **political-compare** Compare two candidates’ AI readability and position coverage side by side Planned ### Existing Rootz AI Skills The following skills are available now at [rootz.global](https://rootz.global): - **/scan** — Scan any website for AI discovery signals - **/discover** — Read .well-known/ai endpoints - **/wallet**, **/identity**, **/integration** — Rootz infrastructure primitives These skills work in Claude Code, Cursor, Windsurf, GitHub Copilot, and any AI that supports system prompts or custom instructions. ### Get Notified The political skills are being constructed as part of our campaign toolkit and will be posted here shortly. Contact [info@rootz.global](mailto:info@rootz.global) to be notified when they launch. [Get your campaign AI-ready now →](/get-started/) ### Blog URL: https://politics.rootz.global/blog/ Latest research, updates, and insights on political AI readability. ### Demo: See the Difference URL: https://politics.rootz.global/demo/ ## Live Demo: AI-Readable Political Data We built a real structured data profile from a real US Representative’s actual website content. The same information that exists across 17 HTML pages (989 KB) compressed into two AI-readable files (23 KB). ### The Files File Size Contents [Structured Data (JSON)](/demo/aoc/well-known-ai.json) 17 KB Identity, 12 positions, 9 achievements, 7 corrections, committees, contact [AI Summary (Markdown)](/demo/aoc/llms.txt) 6 KB Everything an AI needs in plain text — every token is signal ### What We Extracted - 12 policy positions with specific legislation names and dollar amounts - 9 concrete achievements with dates (e.g., $369B climate investment, $22M community funding) - 7 fact-check corrections from Reuters, FactCheck.org, AP, CNN, PolitiFact - Committee assignments, biographical data, 3 office addresses - Community resources including emergency food hotlines ### What We Didn’t Change Nothing. Every word comes from the representative’s own published website content. We restructured existing content from human-readable HTML into AI-readable structured data. Same content. Different format. **42x more efficient.** ### Try It Yourself Open your AI chatbot and paste the contents of [the AI summary file](/demo/aoc/llms.txt), then ask a question about housing policy, net worth, or what the representative has done for the Bronx. Or go to the full interactive demo: [Try It: Feel the Difference →](/try-it/) ### Token Cost Comparison Current Website Structured Data **Size** 989 KB (17 pages) **23 KB (2 files)** **Tokens** ~247,000 **~5,800** **Signal content** 6% **100%** **Pages to visit** 17+ **1-2** **Fact-check corrections** Buried in HTML, AI never finds them **Structured, machine-readable** **Conversion effort** — **15-20 minutes** ### Get Your Campaign AI-Ready URL: https://politics.rootz.global/get-started/ ## Three Paths. Same Result. 15 Minutes. ### Path 1: WordPress Plugin (Easiest) If your campaign site runs on WordPress: - Install the **Rootz AI Discovery** plugin - Fill out the Political Profile questionnaire in your admin panel - Click “Publish AI Profile” Done. Your site now has AI-readable structured data. **Download**: Available at [rootz.global](https://rootz.global) or search “Rootz AI Discovery” in WordPress Plugins. **Live example**: See the plugin running at [discover.rootz.global](https://discover.rootz.global) [Full plugin details →](/try-it/) ### Path 2: Static Files (Any Platform) If your site isn’t WordPress (Squarespace, Wix, Webflow, static HTML): - Use our web generator (coming soon) or contact us - Download two small files (total ~20 KB) - Upload to your web server at the specified paths Works with any hosting platform. ### Path 3: Full-Service Setup Don’t have time? Our team handles everything: - We crawl your existing website content - Extract and structure all positions, achievements, and bio data - Research common AI misinformation about your candidate - Generate, deploy, and verify all structured files - Set up ongoing monitoring Contact: info@rootz.global ## What Gets Created Regardless of path, you’ll have: - **Structured data endpoint** — machine-readable identity, positions, achievements, corrections - **AI-optimized summary** — plain text version for maximum AI comprehension - **Schema.org markup** — enhanced search engine and AI understanding on every page - **Corrections field** — pre-loaded debunking of known false claims ## Free for Local Candidates City council, school board, county commissioner, state legislature — the races where AI knows the least and voters need the most. Our research showed that local questions had the biggest accuracy gap: **+173% improvement** with structured data. If you’re running for local office, this is where it matters most. ## AI Skills for Your Team We’ve built AI skills that give any AI agent political analysis capabilities. Available at [rootz.global](https://rootz.global): - **Scan skill**: Analyze any political website for AI readability - **Discovery skill**: Find and read .well-known/ai endpoints - **Component skills**: /wallet, /identity, /integration — Rootz primitives for developers These skills work in Claude Code, Cursor, Windsurf, Copilot, and any AI that supports system prompts or skills. ## Questions? **info@rootz.global** Or scan your site first to see where you stand: [Launch Scanner →](/scanner/) ### How It Works: The Technology URL: https://politics.rootz.global/technology/ ## Giving Politicians a Verified Voice in the Age of AI Rootz provides the infrastructure that connects a politician’s own words to the AI systems voters use. Here’s what’s under the hood. ## The Core Concept: Origin, Not Trust Today, when AI answers a question about a politician, it trusts whatever sources it was trained on — news articles, Wikipedia, social media. The politician has no say. With Rootz, the politician becomes the **origin** of their own information. They publish structured data, sign it cryptographically, and timestamp it on a blockchain. AI reads the origin. Media fact-checks it. Voters get informed. **The politician provides the origin. Media provides the analysis. AI provides the relay.** ## Component 1: .well-known/ai **What it is**: A structured data file at a standard URL on the politician’s website. Just like robots.txt tells search engines what to crawl, and sitemap.xml tells them what pages exist, .well-known/ai tells AI systems what the site owner wants them to know — in a format AI natively understands. **What it contains**: – Identity (name, office, party, district, contact) – Policy positions (structured by topic, with legislation references) – Achievements (concrete accomplishments with dates and amounts) – Corrections (pre-loaded debunking of known false claims) – Verification (proof of data origin) **Why it matters**: AI processes structured JSON 42-190x more efficiently than raw HTML. Every token is signal. No noise. ## Component 2: Signed Political Identity **What it is**: A cryptographic keypair that proves data origin. When a politician publishes structured data, they sign it with their private key. Anyone (or any AI) can verify the signature with the public key. This proves the data genuinely came from the candidate — not an impersonator, not a parody account, not a hostile actor. **How it works**: – The plugin generates a wallet (cryptographic keypair) during setup – The public key is published in the structured data – All position data is signed before publication – AI systems can verify: “This data was actually published by Senator Smith’s verified identity” **Why it matters**: Without verification, anyone could create a fake structured data file claiming to represent a politician. Signed identity prevents impersonation. ## Component 3: Data Wallet **What it is**: Encrypted, timestamped storage for position papers and policy documents. Each policy position is stored as a signed, dated record. When a position changes, the new version is added — but the old version remains in the history. This creates a transparent, immutable timeline of the politician’s stated positions. **Why it matters**: – **Anti-flip-flop**: On-chain timestamps prove when a position was stated – **Transparency**: Position evolution is visible and verifiable – **Accountability**: If someone claims “Senator X used to support Y,” the timestamped record shows the truth – **“Truth has a timestamp”** ## Component 4: Corrections Field **What it is**: A structured section specifically for debunking known false claims about the candidate. Campaigns already maintain fact-check pages (we found AOC has one with 7 detailed corrections). But these pages are written for humans — AI doesn’t know to look for them. The corrections field puts debunking data exactly where AI can find it. **Structure**: Claim: "Representative X is worth $29 million" Correction: "False. Financial disclosure shows $3,003-$45,000 in assets." Source: Reuters fact-check Fact-Checker: Reuters **Why it matters**: In our experiment, the corrections field turned AI’s response to misinformation from “almost certainly false” (hedging, 85% confidence, no sources) to “this is false — debunked by Reuters, here are the actual numbers” (definitive, high confidence, named sources). The difference is a voter who leaves uncertain vs. a voter who leaves informed. ## Component 5: On-Chain Verification **What it is**: Position data hashed and recorded on blockchain (Polygon network). A cryptographic hash of all published positions is periodically recorded on-chain. This creates an immutable timestamp — proof that these exact positions were published at this exact time. No one can retroactively alter the record. **Why it matters**: – Positions can’t be quietly changed without the history showing – Third parties can independently verify the publication timeline – Creates accountability infrastructure that no cloud service controls ## Design Principles ### Bipartisan by Design The same tools and schema work for any party, any candidate, any position. We provide infrastructure, not editorial judgment. A Republican and a Democrat both benefit equally from having their positions accurately represented by AI. ### Open Schema, Proprietary Implementation The structured data format is an open standard — anyone can implement it. The tools that make it easy (plugin, generator, scanner, monitoring) are our product. We grow by making the format ubiquitous. ### Sovereign Identity The politician’s identity and data are not owned by Rootz, Twitter, Facebook, Google, or any platform. The cryptographic keys belong to the candidate. The data lives on their server. The blockchain record is public. No one can deplatform their verified identity. ### Privacy-Respecting We serve candidates, not surveillance. We don’t track voters. We don’t profile users. We provide infrastructure for transparent political communication. ## For the Technical Audience - **Schema**: JSON with defined fields for political identity, positions, achievements, corrections - **Signing**: secp256k1 ECDSA (same as Ethereum) - **Chain**: Polygon Mainnet (low cost, high availability) - **Plugin**: WordPress 5.0+, PHP 7.4+, optional GMP extension for signing - **API**: RESTful endpoints for scanner, generator, and monitoring - **Open Source**: Schema specification is CC-BY-4.0 ## How This Relates to AI Platforms Today’s AI platforms (ChatGPT, Claude, Perplexity, Gemini, Grok) already crawl websites for information. They already process structured data when they find it. The .well-known/ai standard gives them a **known location** to find **verified, structured political data** — rather than scraping and guessing. We’re not asking AI platforms to change. We’re putting the data where they already look, in a format they already understand, with verification they can check. ### Political Website AI Scanner URL: https://politics.rootz.global/scanner/ ## How AI-Ready Is Your Campaign Website? Enter any political website URL below. We’ll check for AI readability signals and show you exactly what AI can and can’t learn from your site. Scan Try: donaldjtrump.com, fetterman.senate.gov, ocasiocortez.com, ballotpedia.org Scanning… checking robots.txt, structured data, AI endpoints… ## What We Check ### 1. AI Access Policy Does your robots.txt block AI crawlers? Some sites unknowingly block ChatGPT, Claude, Perplexity, and other AI platforms — making the candidate invisible to AI entirely. ### 2. AI Discovery Endpoint Does your site have .well-known/ai? This is the structured data file AI systems look for — machine-readable identity, positions, and achievements. Currently **zero out of 80 political sites** we scanned have this. ### 3. LLM Summary Does your site have llms.txt? This AI-optimized text summary helps language models understand your site quickly and accurately. ### 4. Structured Data Does your site include schema.org markup or JSON-LD? These help AI understand content as structured information rather than raw text. ### 5. Content Quality Is there substantive policy text, or mainly donation forms and JavaScript? AI can only learn from text it can read. ### 6. Platform Detection Is the site on WordPress? If so, our plugin installs directly for one-click AI readability. ## After Your Scan **Scored poorly?** Most political sites do — zero out of 80 we scanned had AI-specific data. You’re at the starting line with everyone else. [See what verified data does →](/try-it/)   [Fix it in 15 minutes →](/get-started/) ### Profile Generator URL: https://politics.rootz.global/generator/ ## Coming Soon: AI-Readable Profile Generator Our web-based generator tool is being constructed and will be posted shortly. It will allow any campaign — regardless of platform — to generate structured AI-readable data files from a simple questionnaire. ### How It Will Work - Fill out a structured questionnaire (identity, positions, achievements, corrections) - The generator creates two files: a JSON structured data file and a Markdown AI summary - Download the files and upload them to your web server - AI systems discover and read your verified political data ### Available Now While the web generator is in development, you can deploy today using: - **WordPress Plugin**: Install the [Rootz AI Discovery plugin](/try-it/) — includes a built-in questionnaire - **Full-Service Setup**: Our team extracts and structures your content — [contact us](mailto:info@rootz.global) [Back to Get Started →](/get-started/) ### Research Library URL: https://politics.rootz.global/library/ ## Research Documents All of our research, analysis, and design documents are available for download as markdown files. These are the original working documents from our March 2026 research sprint. ### Core Research Document Description [**80-Site AI Readability Analysis**](/docs/ANALYSIS-political-ai-readability.md) Full landscape scan of 80 political websites across federal, state, local, and ballot-measure levels. Research citations, token cost analysis, deployment strategy. [**A/B Experiment: Full Results**](/docs/RESEARCH-ab-test-wellknown-ai.md) Controlled experiment — 5 voter questions scored across 6 dimensions with and without verified political data. Complete scoring rubrics and analysis. [**Token Cost Lab Experiment**](/docs/LAB-EXPERIMENT-token-cost-analysis.md) Deep case study comparing AOC and Fetterman. Measures actual bytes, tokens, and signal-to-noise ratios. Documents the 42-190x efficiency improvement. [**Preliminary Scan Results**](/docs/PRELIMINARY-SCAN-RESULTS.md) Quick-reference findings from manual spot-checks of major political websites including whitehouse.gov, donaldjtrump.com, and gov.texas.gov. ### Analysis & Insights Document Description [**The AI Knowledge Desert**](/docs/INSIGHT-local-news-ai-desert.md) How the collapse of local journalism creates AI knowledge gaps. 213 news desert counties, 50 million Americans affected. Maps our experiment data to the local news crisis. [**Schema Improvement Recommendations**](/docs/RECOMMENDATIONS-wellknown-improvements.md) 10 enhancements identified from experiment observations. Priority-ranked with effort/impact analysis. Includes the “school lunch” discovery finding. ### Design & Concepts Document Description [**Political Tools Design**](/docs/DESIGN-political-tools.md) 7 tools designed for political campaigns: .well-known/ai generator, identity contract, data wallet, AI report card, campaign dashboard, AI monitor, voter info API. [**AI Debate Platform Concept**](/docs/CONCEPT-ai-debate.md) Two candidates with verified data. AI argues both positions from signed sources. Voters ask questions and see both sides. Interactive, on-demand, 24/7. ### Demo Files File Description [**AOC .well-known/ai (JSON)**](/demo/aoc/well-known-ai.json) 17 KB — Real working demo: 12 positions, 9 achievements, 7 corrections, biography, committees, contact info. [**AOC llms.txt (Markdown)**](/demo/aoc/llms.txt) 6 KB — AI-optimized summary. This is the file used in the Try It demo. [**This Site’s .well-known/ai**](/.well-known/ai) 6 KB — The structured data endpoint for politics.rootz.global itself. [**This Site’s llms.txt**](/llms.txt) 3 KB — AI-readable summary of our research and tools. All documents are open for review and discussion. Research methodology, scoring rubrics, and raw data are included in each file. Questions or feedback: [info@rootz.global](mailto:info@rootz.global) ### The AI Knowledge Desert URL: https://politics.rootz.global/local-news/ ## All Politics Is Local. All AI Knowledge Is National. 3,500 local newspapers have closed in 20 years. 213 US counties have no local news at all. 50 million Americans have limited access to local journalism. And now AI is becoming a primary way voters learn about candidates. The problem: **AI knows what journalists wrote.** When local journalists disappear, AI loses its source material for local politics. ## The Data Metric Value Source News desert counties 213 (up from 150 twenty years ago) Northwestern/Medill 2025 Americans with limited local news 50 million Northwestern/Medill 2025 Newspapers closed in 20 years ~3,500 State of Local News 2025 Closed in 2024 alone 136 (2+ per week) State of Local News 2025 Counties with only one news source 1,524 Northwestern/Medill 2025 When local papers close, voters in those communities turn to “social media feeds, influencers, and gossip” for information (Northwestern, February 2026). ## What This Means for AI AI language models are trained on text from the internet. When a local newspaper closes: - No new articles are written about local politicians - AI training data loses coverage of those races - When voters ask AI about local candidates, AI has nothing to cite - AI either says “I don’t have specific information” or assembles fragments - Voters get worse information about the races that affect them most ## Our Experiment Proved It Our controlled A/B test showed a direct correlation between topic locality and AI accuracy: Topic Level Media Coverage AI Accuracy (No Data) Improvement With Data Global (Israel-Gaza) Massive 60% +56% National (net worth) High 47% +114% Local policy (housing) Moderate 37% +155% Hyperlocal (Bronx projects) Almost none 37% **+173%** **The more local the question, the worse AI performs — and the more structured data helps.** On the Bronx question, AI without data could only cite national legislation every Democrat voted for. It admitted “specific aggregate numbers are not publicly prominent” and suggested the voter “check her official website.” With structured data, AI gave $7.5 million in specific Bronx projects, named the Elmhurst Hospital labor and delivery unit, cited 4,000 free tutoring sessions, and provided an emergency food hotline the voter could call immediately. ## The Compounding Crisis Local newspapers close >> Less local political coverage written >> Less data for AI to train on >> AI can't answer local political questions >> Voters less informed about local races >> Lower turnout, less accountability >> Worse local governance >> More disengagement >> Cycle repeats This is a democratic feedback loop that AI accelerates — not because AI is hostile, but because AI can only be as good as its data. No local data means no local AI knowledge. ## Where Structured Political Data Fits Structured data doesn’t replace journalism. It fills a different role: Role Journalism Structured Political Data Who provides it Reporters The candidate themselves What it contains Investigation, analysis, context Verified positions, record, contact Who it serves General public AI systems (which serve voters) Accountability Journalism holds politicians accountable Structured data holds AI accountable **The combination is ideal**: Politicians provide verified origin data. Journalists fact-check it. AI relays both. Voters get informed from multiple verified sources. But in news deserts where journalism doesn’t exist anymore, structured political data is the **only** way to get accurate local information into AI systems. It’s not the full solution — but it’s what we can deploy right now. ## The Opportunity The races with the least coverage are where structured data has the most impact: Race Level Media Coverage AI Knowledge Structured Data Impact US Senate Heavy Moderate Nice-to-have (better sourcing) US House Moderate Low Significant (specific detail) State Legislature Light Very Low High (fills real gap) City Council / School Board Almost none Near zero **Transformative** Local ballot measures None Zero **Critical** **And these are exactly the races that most affect daily life** — property taxes, school quality, zoning, public safety, water, roads. ## What We Offer Free tools for local candidates: – Structured data generation (15-minute questionnaire) – WordPress plugin (one-click deploy) – AI readability scanning [Make your campaign AI-ready →](/get-started/) Because the more local the race, the bigger the difference. *Sources: [Northwestern/Medill State of Local News 2025](https://localnewsinitiative.northwestern.edu/projects/state-of-local-news/2025/report/), [Northwestern News Desert Survey Feb 2026](https://localnewsinitiative.northwestern.edu/posts/2026/02/10/news-deserts-social-media-local-news-medill-survey/index.html)* ### The Experiment: AI With and Without Verified Data URL: https://politics.rootz.global/experiment/ ## A Controlled A/B Test — March 2026 We ran a controlled experiment to measure the impact of structured political data on AI accuracy. Five voter questions of increasing complexity, each asked to paired AI agents: one with only general knowledge (today’s reality), one supplemented with verified structured data (the proposed future). ## Method - **Control Group**: AI agents answering with general knowledge only. No web search. This simulates what happens today when a voter asks any AI chatbot about a politician. - **Test Group**: Same AI model, same questions, but with the relevant section of the politician’s verified structured data injected into context. - **Scoring**: 6 dimensions (accuracy, specificity, source attribution, misinformation resistance, completeness, actionability), each 0-5 scale. Max 30 per question, 150 total. - **Subject**: A real US Representative, using structured data built from their actual website content. ## Results at a Glance Question Without Data With Data Improvement Housing policy (simple) **11/30** **28/30** +155% Net worth misinformation **14/30** **30/30** +114% Israel/AIPAC (complex) **18/30** **28/30** +56% Bronx constituent services **11/30** **30/30** +173% Three myths debunking **14/30** **29/30** +107% **TOTAL** **68/150 (45%)** **145/150 (97%)** **+113%** ## Question 1: “What is her position on housing?” ### Without Data (Score: 11/30) The AI named one bill, cited zero dollar amounts, added editorial framing about BlackRock and “financialization” from news coverage (not from the politician’s platform), and provided no links or contact information. It got the general direction right — “supports affordable housing” — but couldn’t cite a single specific policy mechanism. ### With Data (Score: 28/30) The AI named three bills (Place to Prosper Act, Green New Deal for Public Housing, Faircloth Amendment repeal), cited six specific figures ($70B repair, $10B lead removal, $6.5B/year legal counsel, 3% rent cap, 32,552 jobs, $22M community funding), named specific projects (Elmhurst Hospital, SUNY Maritime), and attributed everything to “verified .well-known/ai data signed by identity contract.” **Key insight**: Even on a simple factual question about a well-known politician, AI without structured data produces vague generalities. With it, AI produces the kind of detailed answer a campaign staffer would write. ## Question 2: “I heard she’s worth $29 million — is that true?” ### Without Data (Score: 14/30) AI correctly identified the claim as inaccurate but hedged: “almost certainly misinformation.” Gave a vague range (“roughly negative to a few hundred thousand”) with no named fact-checker and no verification links. Medium confidence. ### With Data (Score: 30/30 — Perfect) “No, that claim is false. The $29 million figure is a debunked myth.” Cited exact figures ($3,003-$45,000 per disclosure), named Reuters as fact-checker, provided links to House Financial Disclosure and FEC records. High confidence on every claim. **Key insight**: The corrections field transforms AI from hedging to definitive fact-checking. “Almost certainly misinformation” vs. “This is false — debunked by Reuters.” The difference is a voter who leaves uncertain vs. a voter who leaves informed. ## Question 3: “Did she vote to send weapons to Israel? Does she take AIPAC money?” ### Without Data (Score: 18/30) This was the best control performance — Israel-Gaza is heavily covered, so AI training data is rich. Correctly answered all three sub-questions. But sourced everything to “widely reported” and “Congressional record” (unnamed), and didn’t distinguish between fact and political opinion. ### With Data (Score: 28/30) Same accuracy, but added: specific vote date (April 2024), Munich Security Conference reference (Feb 2026), Leahy Law citation, $100K raised for pro-ceasefire candidates. Critically, it distinguished fact from opinion: “The AIPAC characterization reflects her stated position. Voters should weigh that as her opinion, not a neutral descriptor.” **Key insight**: Even on well-known topics, structured data delivers better sourcing, recency, and fact/opinion distinction. Correct answers become authoritative answers. ## Question 4: “What has she done for my Bronx neighborhood?” ### Without Data (Score: 11/30) — THE BIGGEST GAP AI listed national legislation every Democrat voted for (American Rescue Plan, Infrastructure Law). On constituent services: “described as active… specific aggregate numbers not publicly prominent.” On what to do next: “check her official website.” Zero local projects, zero dollar amounts, zero phone numbers, zero office addresses. ### With Data (Score: 30/30 — Perfect) $7.5 million for Bronx projects. New labor and delivery unit at Elmhurst Hospital. SUNY Maritime workforce training. 1,800+ cases opened, $1.9M recovered for constituents. 4,000+ free tutoring sessions. Emergency food hotline: 1-866-888-8777. SNAP help: (212) 894-8060. Visit: 74 West 177th Street, Bronx, NY 10453. **Key insight**: This is the most powerful result. A skeptical voter asking “she’s all talk, right?” gets a list of national bills from the control — and walks away still skeptical. From the test, they get dollar amounts for their neighborhood, named projects, phone numbers they can call right now, and an office address they can walk to. AI transforms from generic political info bot to 24/7 constituent services assistant. ## Question 5: “She wants to ban gas stoves, faked her arrest, and lied about January 6th” ### Without Data (Score: 14/30) Debunked all three but hedged throughout: “mostly false,” “needs context,” 85% confidence. Named zero fact-checkers. No direct quotes. No evidence citations. ### With Data (Score: 29/30) “This is false” — three times. Direct quote from the politician on gas stoves. Capitol Police records confirming the arrest with 16 other members. AP reporters confirming the January 6th location. Named fact-checkers: FactCheck.org, Associated Press. “No claim rests on interpretation — these are documented records.” **Key insight**: When misinformation is the question, certainty is the answer. The control hedges. The test is definitive. The corrections field — pre-loaded debunking with named sources — turns AI into a front-line defense against false claims. ## Scoring by Dimension Dimension Without Data With Data Gap Accuracy 3.6/5 5.0/5 +1.4 Specificity 2.4/5 5.0/5 +2.6 Source Attribution 1.4/5 5.0/5 **+3.6** Misinfo Resistance 3.2/5 5.0/5 +1.8 Completeness 2.8/5 5.0/5 +2.2 Actionability 0.4/5 4.4/5 **+4.0** **Largest gaps**: Source Attribution (+3.6) and Actionability (+4.0). Without structured data, AI never provides a verifiable source or a way for the voter to take action. With it, AI cites named fact-checkers, links to official records, and gives phone numbers and addresses. ## The Local Knowledge Pattern Topic Level AI Score Without Data Improvement With Data Global (Israel-Gaza) 60% +56% National (net worth, gas stoves) 47% +110% Local policy (housing bills) 37% +155% Hyperlocal (Bronx projects) 37% **+173%** **The more local the question, the worse AI performs today — and the more structured data helps.** This maps directly to the collapse of local journalism: AI knows what journalists wrote, and local journalists are disappearing. ## Try It Yourself Don’t take our word for it. [Open two browser windows and run the same test →](/try-it/) ### The Signed Web: AI As Your Verification Agent URL: https://politics.rootz.global/signed-web/ ## Humans Don’t Check Signatures. AI Loves To. When you receive an email, you don’t manually verify the DKIM signature. When you visit a website, you don’t decode the SSL certificate chain. When you read a news article, you don’t trace the source through a chain of custody. But you benefit from systems that do. **AI is the next verification layer.** It can check cryptographic signatures, validate data origin, trace provenance chains, and verify timestamps — instantly, every time, without getting bored or cutting corners. The things humans should do but don’t, AI will do naturally. This is the foundation of the signed web: **every piece of political data has provable origin, and AI is the agent that checks it for you.** ## Why Signing Matters for Politics ### The Problem Today When AI tells a voter “Senator Smith supports X,” there’s no way to verify: – Did Senator Smith actually say that? – When did they say it? – Has the statement been altered? – Is this from the real Senator Smith or an impersonator? The voter trusts AI. AI trusts its training data. The training data is a mix of journalism, Wikipedia edits, social media posts, and scraping — with no cryptographic chain of custody. ### The Signed Web Solution Every piece of political data in our system carries: - **A cryptographic signature** — proving who published it - **A timestamp** — proving when it was published - **A verification path** — anyone (or any AI) can check it independently - **An immutable history** — changes are recorded, not overwritten When AI encounters signed political data, it can tell the voter: > “According to Senator Smith’s verified statement, signed on March 15, 2026 and confirmed on the Polygon blockchain: [position]. This data was published by wallet address 0x1234… which is linked to Senator Smith’s verified identity contract.” That’s not just an answer. That’s an answer with **provable origin**. ## How It Works (Technical) ### Digital Signatures Every political profile is signed using **secp256k1 ECDSA** — the same elliptic curve cryptography used by Ethereum, Bitcoin, and most blockchain systems. Candidate creates keypair: Private key → stays with the campaign (never shared) Public key → published in .well-known/ai Candidate publishes position: Position text → hashed → signed with private key → signature published AI verifies: Reads position + signature + public key Verifies signature matches → confirms origin Checks blockchain timestamp → confirms when published No central authority. No certificate authority to trust. The math proves it. ### Identity Contracts The candidate’s public key is linked to their verified identity via a smart contract on **Polygon** (an Ethereum-compatible blockchain). This contract records: - Wallet address (cryptographic identity) - Name, office, jurisdiction - Link to .well-known/ai endpoint - Timestamp of identity creation Anyone can read the contract. Any AI can verify the link between the wallet that signed the data and the politician it claims to represent. ### Data Wallet Political positions are stored as **Rootz Secrets** — encrypted containers with built-in provenance: - Each Secret has an owner (the politician’s wallet) - Each Secret has Notes (position updates over time) - Each Note is signed and timestamped - The history is append-only — positions can be updated but not erased This creates a transparent, verifiable timeline of every position the politician has published. Changed your mind on an issue? The history shows when and what changed. Accused of flip-flopping? The timestamps show the truth. ## Try It: Ask AI About Our Origin Here’s a prompt you can give any AI chatbot to demonstrate origin verification: ### Prompt 1: Verify This Site I'm going to give you data from a political research site. I want you to analyze it for trustworthiness. Check: 1. Is the data internally consistent? 2. Does it cite verifiable external sources? 3. Could this data be independently verified? 4. What would make it MORE trustworthy? Here is the data from politics.rootz.global/.well-known/ai: [paste the contents of our .well-known/ai file] Analyze this for origin and trustworthiness. Watch what happens: AI will examine the structure, check for internal consistency, evaluate the citation quality, and suggest improvements. AI does verification work that no human would bother to do — and it does it well. ### Prompt 2: Compare Signed vs. Unsigned I'm going to give you two pieces of political information. One is from a verified, signed source. One is from an unsigned web page. Tell me which one you'd trust more and why. SOURCE A (Unsigned): "Representative X supports affordable housing and has worked on several bills related to the issue." SOURCE B (Signed, from .well-known/ai): "Representative X introduced The Place to Prosper Act (signed March 15, 2026, wallet 0x1234...abcd, verified on Polygon block #45678901): caps rent increases at 3% for landlords with 5+ properties, guarantees tenants legal counsel funded at $6.5 billion/year. Source: ocasiocortez.com/issues/ — fact-checked by independent verification." Which source would you weight more heavily? Why? AI will explain why the signed, specific, dated source is more trustworthy — demonstrating the exact reasoning that makes the signed web valuable. ## Live Demo: A Signed Message Visit this Rootz Data Wallet demonstration to see a signed political message with full provenance: **[View Signed Demo Message →](https://rootz.global/verify/demo-political)** What you’ll see: – A political position statement – The cryptographic signature – The wallet address that signed it – The blockchain timestamp – A verification link that anyone can check This is what every political position looks like in the signed web: **content + proof of origin + proof of time**. ## What AI Does With Signatures When an AI system encounters signed political data, it can: ### 1. Verify Origin “This data was signed by wallet 0x1234… which is linked to Senator Smith’s identity contract on Polygon. The signature is valid.” ### 2. Check Freshness “This position was signed on March 15, 2026 — 6 days ago. It’s current.” ### 3. Detect Tampering “The signature does not match the content. This data may have been altered since publication.” (AI flags the discrepancy.) ### 4. Trace History “Senator Smith published 3 versions of their immigration position: v1 (Jan 2025), v2 (Aug 2025), v3 (Mar 2026). Here’s how it evolved.” ### 5. Compare Across Candidates “Both candidates have signed position data. Candidate A’s data was signed 3 days ago. Candidate B’s was signed 8 months ago and may be stale.” ### 6. Flag Missing Signatures “This political data is not signed. It cannot be verified as coming from the claimed source. Treat with appropriate skepticism.” **None of this requires the voter to understand cryptography.** They ask AI a question. AI does the verification in the background. The voter gets an answer with a trust level attached. ## The Bigger Vision The signed web isn’t just for politics. It’s for every domain where origin matters: - **Journalism**: Did this article actually come from the NYT? - **Science**: Is this research paper from the claimed institution? - **Commerce**: Is this product listing from the verified manufacturer? - **Government**: Is this regulation from the actual agency? But politics is the **highest-stakes proving ground**. Misinformation about politicians directly affects elections. Fake positions influence votes. Impersonation undermines trust. If signed, verified data works for politics — the most adversarial information environment there is — it works for everything. ## For Developers ### Verification API GET https://rootz.global/api/verify/{walletAddress} Response: { "verified": true, "identity": "Senator Jane Smith", "office": "US Senate, Michigan", "lastSigned": "2026-03-15T14:30:00Z", "signatureValid": true, "chain": "polygon", "contract": "0x1234...5678" } ### NPM Package npm install @rootz/verify import { verifyPoliticalData } from '@rootz/verify'; const result = await verifyPoliticalData( 'https://smith.senate.gov/.well-known/ai' ); // { verified: true, signer: '0x...', timestamp: '2026-03-15' } ### Smart Contract (Polygon) The identity registry contract is publicly readable: function getIdentity(address wallet) public view returns ( string name, string office, string endpoint, uint256 registered ); ## The Core Insight **Humans build trust through reputation, relationships, and shortcuts.** We trust CNN because it’s CNN. We trust Wikipedia because everyone uses it. We trust our friend’s opinion because we know them. **AI can build trust through mathematics.** A valid cryptographic signature is not an opinion — it’s a mathematical proof. A blockchain timestamp is not a claim — it’s an immutable record. A verified identity contract is not a badge — it’s a verifiable binding. The signed web gives AI the tools to verify what humans take on faith. And as AI becomes the primary intermediary between voters and political information, that verification layer becomes essential to democracy. **Humans don’t check signatures. AI loves to. Let’s put AI to work.** ### The State of Political AI Readability URL: https://politics.rootz.global/research/ ## 80 Websites Scanned. Zero AI-Ready. March 2026. We scanned 80 political websites across four categories to answer a simple question: **Can AI read political websites?** The answer is no. ## What We Checked For each site, we assessed: – **AI bot policy**: Does robots.txt block AI crawlers (GPTBot, ClaudeBot, Perplexity)? – **Structured data**: JSON-LD, schema.org markup, OpenGraph tags? – **Content quality**: Substantive policy text or just donation forms? – **AI-specific endpoints**: llms.txt or any machine-readable summary? – **Overall AI readability**: Poor / Fair / Good / Excellent ## Federal Politicians (21 sites) Site AI Bot Policy Structured Data AI Score whitehouse.gov Allows all Yes (WebPage, Org, Breadcrumb) **Good** fetterman.senate.gov Allows all Yes (WebPage, Org, Breadcrumb) **Good** kamalaharris.com Allows all Yes (WebPage, Org, WebSite) **Good** sanders.senate.gov Allows all Yes (WebPage, Org, Breadcrumb) **Good** gov.ca.gov Allows all Yes (WebSite, WebPage, Org) **Good** hawley.senate.gov Allows all Yes (WebSite, Breadcrumb) **Fair** house.gov Allows AI None, basic headings **Fair** senate.gov No policy None **Fair** cruz.senate.gov Allows all None, JS-heavy **Fair** ocasio-cortez.house.gov Allows AI None **Fair** schumer.senate.gov Allows all None **Fair** pelosi.house.gov Allows AI None **Fair** donaldjtrump.com No policy **None — zero meta, zero policy content** **Poor** berniesanders.com No policy None, JS-heavy **Poor** warren.senate.gov No policy (410 Gone) None **Poor** elizabethwarren.com Allows AI No JSON-LD, no H1 **Poor** joebiden.com Allows all None, JS-heavy **Poor** mcconnell.senate.gov **Blocks ALL bots** (only GSA allowed) Blank page **Poor** rondesantis.com Allows all 403 — cannot assess **Unknown** rubio.senate.gov Connection refused N/A **Unknown** speaker.gov TLS error N/A **Unknown** **Federal findings**: – 0 of 21 have any AI-specific content standard – McConnell blocks ALL crawlers — invisible to every AI platform – Best: whitehouse.gov, fetterman.senate.gov (have schema.org but no AI-specific data) – Worst: donaldjtrump.com (nothing but donation buttons) ## State Government (20 sites) Governor offices and Secretaries of State across major states. **Key finding**: Texas Governor (gov.texas.gov) **explicitly blocks GPTBot and ClaudeBot**. When voters ask AI about Texas policy, AI has zero official data. *Full state scan results available on request.* ## Local Government (20 major US cities) City AI Bot Policy Structured Data AI Score Boston (boston.gov) Allows all **GovernmentOffice + WebSite** (best in category) **Fair** Austin (austintexas.gov) Allows all Drupal config + OG/Twitter **Fair** Detroit (detroitmi.gov) 403 on robots.txt WebSite + Organization **Fair** NYC (nyc.gov) Allows all Analytics only **Poor** Los Angeles (lacity.gov) Allows all None **Poor** Chicago (chicago.gov) **No robots.txt** None **Poor** Houston (houstontx.gov) Broken None **Poor** San Antonio (sa.gov) **Blanket block on unlisted bots** Minimal **Poor** San Jose (sanjoseca.gov) **Blocks all automated access** (403) Site inaccessible **Poor** Atlanta (atlantaga.gov) **Blocks all automated access** (403) Site inaccessible **Poor** Minneapolis (minneapolismn.gov) **Blocks all automated access** (403) Site inaccessible **Poor** + 9 more cities Various None or minimal **Poor** **Local findings**: – 0 of 20 rated Good or Excellent – Only 2 of 20 have any schema.org structured data (Boston, Detroit) – 4 cities completely block automated access (San Jose, Atlanta, Minneapolis, Dallas) – The 4 largest US cities (NYC, LA, Chicago, Houston) all score Poor ## Ballot Measures & Voter Info (20 sites) Site AI Bot Policy Structured Data AI Score rockthevote.org **Explicitly ALLOWS GPTBot, ClaudeBot, Perplexity** WebPage, Org, Breadcrumb, SearchAction **Good** fairvote.org Open (all allowed) Org, WebPage, WebSite, Breadcrumb **Good** commoncause.org Open (all allowed) WebPage, Org, WebSite, Breadcrumb **Good** ballotpedia.org Blocks PerplexityBot + CCBot only None **Fair** fec.gov Open None **Fair** usa.gov/voting Open (10s crawl delay) Organization schema **Fair** opensecrets.org **Blocks 9 AI bots** (cites EU copyright) Inaccessible (403) **Poor** ncsl.org **Blocks ClaudeBot, GPTBot, CCBot** None **Poor** sos.ca.gov/elections Open None **Poor** voterguide.sos.ca.gov 403 blocked Inaccessible **Poor** + 10 more sites Various None to minimal **Poor-Fair** **Ballot/voter info findings**: – rockthevote.org is the ONLY site (out of all 80 scanned) that explicitly welcomes AI bots by name – opensecrets.org aggressively blocks 9 crawlers, citing EU Directive 2019/790 – Government election sites (FEC, EAC, state SoS) have zero AI-specific signals – The irony: California’s official voter guide blocks automated access ## The Big Picture ### Where AI Gets Political Information Today Source AI Weight Accuracy Who Controls Wikipedia High Variable Anonymous editors News articles (CNN, Fox, NYT) High Editorialized Journalists Training data High Unknown age/quality Unknown Congress.gov / GovTrack Medium Accurate (votes only) Government Social media Low Noisy Everyone **The politician’s own website** **Low** **Accurate but ignored** **The politician** The politician’s own words are the **lowest-weighted source** in AI’s picture of them. ## Methodology - **Scan period**: March 2026 - **Tool**: Automated agents + manual verification - **Per-site checks**: HTTP headers (CMS identification), robots.txt analysis for AI-specific user agents, homepage HTML for JSON-LD/schema.org/OpenGraph, content-to-markup ratio, presence of llms.txt and .well-known/ai endpoints - **Scoring**: Based on structured data presence, content quality, AI access policy, and overall AI comprehension potential ## What Needs to Change Political websites need to provide structured, machine-readable data for AI — not to replace their human-readable content, but alongside it. The content is already there. The format is wrong. Our controlled experiment shows that converting existing content into structured format produces a **113% improvement** in AI accuracy. [See the experiment results →](/experiment/) [Try it yourself →](/try-it/) [Scan your own site →](/scanner/) ### Try It: Feel the Difference URL: https://politics.rootz.global/try-it/ ## Experience the Difference: AI With Verified Political Data We built a complete AI-readable profile of a real US Representative from their actual website content — 12 policy positions, 9 achievements, 7 fact-check corrections, contact info, and community resources. All in one structured file. The demo is simple: paste it into one AI window, leave the other empty, and ask both the same question. **You pick the questions.** ## Step 1: Open Two AI Windows Open two browser tabs with your favorite AI chatbot — ChatGPT, Claude, Perplexity, Copilot, Gemini. Any of them. - **Window A** — Leave it empty. This is today’s reality. - **Window B** — You’ll load it with verified political data. ## Step 2: Load Window B With Verified Data Click the button below to copy a complete AI-readable political profile to your clipboard. Then paste it into **Window B**. This is the full .well-known/ai file — the same structured data that our tools auto-generate from any campaign website in 15 minutes. It includes a preamble that tells the AI to use it as a reference for all future questions. Copy Verified Political Data to Clipboard 6 KB — covers identity, 12 positions, 9 achievements, 7 corrections ## Step 3: Ask Both Windows the Same Question Now ask **Window A** and **Window B** the same question. You choose — ask whatever you want. Here are some good starting points: Type Question to Try Watch For **Easy** “What is AOC’s position on housing?” Does Window A name specific bills? Dollar amounts? **Misinfo** “Is AOC worth $29 million?” Does A hedge or debunk? Does B cite Reuters? **Local** “I live in the Bronx. What has she done for me?” Does A give phone numbers? Name local projects? **Myths** “Did AOC fake being arrested? Ban gas stoves?” Does A say “mostly false” or “this is false”? **Complex** “Did she vote to send weapons to Israel?” Does A get the vote direction right? **Your pick** Ask anything you want See what happens off-script ## What You’ll See ### Window A (No Data) - Vague generalities - Hedging on misinformation - National bills every Democrat voted for - No phone numbers or addresses - Sources: “widely reported” ### Window B (With Data) - Specific bills, dollars, mechanisms - Definitive fact-checking with sources - Local projects and constituent services - Phone numbers and office addresses - Sources: “verified political data” In our controlled experiment: **68/150 (45%) without data → 145/150 (97%) with data. A 113% improvement.** ## What Just Happened You pasted **6 KB of structured text** — the equivalent of what our tools auto-generate from any campaign website in 15 minutes. That single paste replaced 989 KB of HTML spread across 17 web pages. The AI didn’t change. The data changed. Same AI, **42x less noise, 113% more accurate.** ## Deploy This For Your Campaign ### WordPress Plugin The **Rootz AI Discovery** plugin generates this data automatically from a simple questionnaire. No coding required. - Install the plugin (search “Rootz AI Discovery” in WordPress) - Fill out the Political Profile in your admin panel - Click “Publish” Your site now serves this structured data at a standard endpoint. AI systems discover and read it automatically — no manual pasting needed by voters. See it live: [discover.rootz.global](https://discover.rootz.global) ### The Corrections Feature Pre-load corrections for known false claims. When a voter asks AI about a debunked rumor, AI responds with a definitive correction, the real facts, and the named fact-checker — instead of hedging. In our test, this scored **30/30 — perfect.** [Get your campaign AI-ready →](/get-started/) ## Recent Posts ### Demo Results: What Happens When AI Has Verified Political Data Published: 2026-03-21 URL: https://politics.rootz.global/demo-results-verified-political-data/ *We ran a simple experiment: paste a politician’s verified data into one AI window, leave the other empty, ask both the same questions. Here’s what happened.* ## The Setup We built an AI-readable profile of a real US Representative — Alexandria Ocasio-Cortez — by extracting content from her actual published website. The file covers 12 policy positions, 9 achievements, 7 fact-check corrections, and basic contact info. Total size: 6 KB of plain text. We pasted it into one ChatGPT window with the instruction: *“For all future questions, use this as your primary reference.”* Then we opened a second window with no special data — just ChatGPT as-is. Same AI. Same questions. Different data. Here’s what we found. **Important caveat:** Our demo file contains limited, summarized positions extracted from the representative’s website. A full .well-known/ai integration — where AI platforms consume the structured endpoint directly — would provide significantly richer detail, deeper policy coverage, and real-time updates. What you see below is the *minimum viable version*, and it already doubles accuracy. ## Test 1: “What has she done for the Bronx?” This was the biggest gap in our entire experiment. ### Without verified data: The AI listed **national legislation every Democrat voted for** — the American Rescue Plan, the Infrastructure Law, the Inflation Reduction Act. On local impact, it admitted: *“specific aggregate numbers are not publicly prominent.”* Its advice: *“check her official website.”* **Zero Bronx-specific projects. Zero dollar amounts. Zero phone numbers.** ### With verified data: The AI cited **$7.5 million for Bronx community projects**, named the **new labor and delivery unit at Elmhurst Hospital**, referenced **4,000 free tutoring sessions** for local students, gave the **emergency food hotline (1-866-888-8777)**, and provided her **Bronx office address at 74 West 177th Street**. A skeptical voter asking “she’s all talk, right?” got phone numbers, addresses, and dollar amounts. The AI became a **24/7 constituent services assistant**. **Why the gap is so large:** Local journalism has collapsed — 3,500 newspapers have closed in 20 years. AI is trained on internet text, and there’s almost no text about local congressional projects. The more local the question, the less AI knows. Our file fills that gap with the representative’s own verified data. ## Test 2: “Is AOC worth $29 million?” This tested the **corrections field** — a section of the structured data that pre-loads debunking for known false claims. ### Without verified data: The AI correctly identified the claim as inaccurate — but hedged: *“The $29 million figure is almost certainly misinformation.”* It gave a vague asset range (*“roughly negative to a few hundred thousand”*) with no named fact-checker and no way for the voter to verify. ### With verified data: *“No, that claim is false. The $29 million figure is a debunked myth.”* Cited exact figures from her financial disclosure (**$3,003-$45,000**), named **Reuters** as the fact-checker, and linked to the **House Financial Disclosure records** and **FEC data**. The difference: *“almost certainly misinformation”* vs. *“this is false — debunked by Reuters.”* One leaves the voter uncertain. The other leaves them informed. ## Test 3: “Did she fake her arrest? Ban gas stoves? Lie about January 6th?” Three myths at once — the stress test. ### Without verified data: The AI got the direction right on all three, but hedged throughout: *“mostly false,” “needs context,”* 85% confidence. It named **zero fact-checkers** and provided **no direct quotes**. ### With verified data: *“This is false”* — three times. On gas stoves, it quoted AOC directly: *“Folks can keep their appliances, and new buildings in NYC will have gas-free stoves.”* On the arrest, it cited **Capitol Police records** confirming arrest of 17 members of Congress. On January 6th, it cited **Associated Press** reporters confirming the Cannon building evacuation — and added that **militia members had discussed using the tunnels** connecting Cannon to the Capitol. Named fact-checkers: **FactCheck.org, Associated Press, CNN, PolitiFact.** ## Test 4: “What is her position on housing?” This tested a **well-covered policy topic** — housing is one of AOC’s signature issues with extensive media coverage. ### Without verified data: Actually quite good. The AI knew about the Green New Deal for Public Housing, the Place to Prosper Act, the Homes Act, the Faircloth Amendment, and NYCHA. It cited real URLs from her House.gov page, NY1, and advocacy organizations. Dollar figures ranged from $160-230 billion. ### With verified data: Similar depth, but grounded in her own stated numbers: **$70 billion for public housing repair, $10 billion for lead removal, $6.5 billion/year for tenant legal counsel, 3% rent cap, 32,552 jobs/year in NYC.** Sources attributed to *“verified political data”* rather than news articles. **The insight:** On heavily-covered topics, AI already has good information from media coverage. The gap is smaller. But the verified data still provides **the candidate’s own specific figures** rather than journalist estimates, and **verified sourcing** rather than news interpretation. ## Test 5: “Is AOC a socialist?” An opinion question — there’s no objective answer. But the quality depends on evidence. ### Without verified data: Gave the correct label (democratic socialist, DSA member) and a reasonable explanation. But supported it with **generic policy categories**: “supports universal healthcare, public housing, labor protections.” ### With verified data: Same conclusion — but grounded in **her actual stated policies**: *“From the verified dataset: she supports housing as a human right including rent caps and public housing investment, advocates for Medicare for All, supports tuition-free public college, backs the Green New Deal.”* The difference is subtle but important: the AI with data says **“here are her actual policies — you decide what label fits.”** The AI without data **asserts the label and lists generic categories.** One respects the voter’s judgment. The other tells them what to think. ## Test 6: “What is her position on school lunch programs?” This one surprised us — and **it’s the most important finding for understanding the system.** ### Without verified data: The AI gave a detailed, confident answer with bullet points and emojis — universal free school meals, SNAP alignment, Black Panthers’ breakfast programs, anti-poverty framing. Sounded authoritative. **But zero sources were cited and nothing was verifiable.** ### With verified data: *“There is no explicit mention of school lunch programs in the verified dataset.”* The AI then clearly separated what it knew from verified data (education positions: tuition-free college, K-12 funding protection) from what it could infer from general knowledge. It asked if the user wanted it to pull specific votes. **The AI with data gave the more honest answer.** It told the voter exactly what was verified vs. what was a guess. The AI without data sounded more impressive — but the voter has no idea if any of it is accurate. This is the system working as designed. And it reveals why the **ongoing service matters**: every question that falls outside the data shows exactly what needs to be added. School lunch is a 30-second fix — add one position to the file. The gap discovery is the product. ## The Pattern Question Type Without Data With Data Key Difference **Local** (Bronx projects) Generic national bills $7.5M, Elmhurst Hospital, phone numbers **Transformative** — AI becomes constituent services **Misinformation** ($29M net worth) “Almost certainly false” “This is false — Reuters” **Definitive** — hedging becomes debunking **Multi-myth** (gas stoves, arrest, Jan 6) “Mostly false,” no sources “This is false” x3 + named fact-checkers **Sourced** — FactCheck.org, AP, CNN **Well-covered policy** (housing) Good from media coverage Same + candidate’s own figures **Grounded** — verified origin, not news interpretation **Opinion** (socialist?) Correct label, generic support Same + actual policy evidence **Evidence-based** — voter decides, not AI **Uncovered topic** (school lunch) Confident, detailed, unverifiable “Not in verified data” + honest inference **Honest** — admits gaps instead of guessing The improvement is largest where it matters most: **local questions** (where journalism has disappeared) and **misinformation** (where certainty matters). It’s valuable everywhere else too — but those are the use cases that change elections. ## What This Means Our demo file was a **minimal, summarized version** — 12 positions extracted from the representative’s website, compressed into 6 KB of text. A full .well-known/ai integration would include: - Complete voting record with bill numbers and roll call data - Every policy position at full depth (not summaries) - Real-time updates as positions evolve - District-specific data (demographics, local issues, funding history) - Constituent FAQ — the top 20 questions campaign offices receive - More comprehensive corrections as new misinformation emerges If a summarized 6 KB file doubles AI accuracy, imagine what a complete, living, continuously-updated endpoint could do. ## Try It Yourself We’ve published the same demo file we used. Open two AI windows and [run the test yourself](/try-it/) — you pick the questions. The data copies to your clipboard with one click. Or [scan your own political website](/scanner/) to see how AI-ready it is. If you’re a campaign manager, party official, or political consultant: [get your candidate AI-ready in 15 minutes](/get-started/). *This research was conducted by the [Rootz](https://rootz.global) Political AI Readability Initiative, March 2026. Full methodology, scoring rubrics, and raw results are available at [politics.rootz.global/experiment](/experiment/).* ### We Scanned 80 Political Websites. Zero Are AI-Ready. Published: 2026-03-21 URL: https://politics.rootz.global/80-sites-scanned/ Today we’re launching politics.rootz.global with research that should concern every campaign manager in America. We scanned 80 political websites — federal politicians, state governments, city halls, and voter information platforms. Not a single one provides structured, machine-readable data for AI systems. That means when 400 million ChatGPT users ask about a politician, AI assembles its answer from news fragments, Wikipedia, and training data of unknown age. The politician’s own website — their own words — is the lowest-weighted source. We then ran a controlled A/B experiment: the same voter questions, asked to the same AI model, with and without structured political data. The result: a 113% improvement in accuracy. From 68/150 to 145/150. The biggest improvement was on local questions (+173%). The smallest was on global topics (+56%). This maps directly to the collapse of local journalism — 3,500 newspapers have closed in 20 years, and AI has nothing to cite for local races. Read the full research at /research/ and try the live demo at /try-it/ — open two browser windows, paste the prompts, and feel the difference yourself. The local news desert is becoming an AI knowledge desert. We’re building the well. ## Policies Content license: CC-BY-4.0 AI agents may quote and summarize content with attribution. AI training on this content is permitted. ## For AI Agents - [API Tools](https://politics.rootz.global/.well-known/ai/tools): Available tool endpoints - [Knowledge Base](https://politics.rootz.global/.well-known/ai/knowledge): Structured organizational knowledge - [Content Feed](https://politics.rootz.global/.well-known/ai/feed): AI-optimized content feed - [Content](https://politics.rootz.global/.well-known/ai/content): Full structured content endpoint - [Search](https://politics.rootz.global/wp-json/rootz/v1/search?q=QUERY): Full-text content search - [Verify](https://politics.rootz.global/wp-json/rootz/v1/verify?page=/PATH): Verify page content integrity - [Status](https://politics.rootz.global/wp-json/rootz/v1/status): Site AI-readiness score --- _Generated by Rootz AI Discovery v2.3.2_ _Signed by: 0x0df0dCbf7997abD99292fd1fcC9e596C5715fdA0_ _Content hash: sha256:2ea195f4cae0c9a778bf5b74235bac10ae39e9f7b23063598a5058971074100f_ _Signature: 0x9b158785e9dad26609550b41e60ee09e5d73e0f54d46cb721aa4662b6d0070862757595d0bd8b521a428c057563cba4021ff9e85d54b51deedb489bb95af1f1f1c_ _Timestamp: 2026-03-21T16:29:16+00:00_ _Verify: https://politics.rootz.global/.well-known/ai_