--- title: Can I Speak to a Human? The Intersection of AI and Accessibility and what we learned at NAGW (National Association of Government Web Professionals) description: Suraj Kika, Jadu CEO and Netcall Chief Growth Officer, shares his thoughts on the intersection between accessibility and AI. image: https://blog.jadu.net/hubfs/Blog-Hero-Image-1.jpg --- - [Skip to content](https://blog.jadu.net/blog/can-i-speak-to-a-human-the-intersection-of-ai-and-accessibility-and-what-we-learned-at-nagw-national-association-of-government-web-professionals#main-content) [![JADU main site home page](https://blog.jadu.net/hubfs/raw_assets/public/JADU_August2022/images/logo.svg "JADU main site home page")](https://www.jadu.net/) - [Support](https://www.jadu.net/support) - [Jadu Library](https://www.jadu.net/library) - [Contact](https://www.jadu.net/contact) - Search Open Search Close Navigation Close navigation - Products - [Jadu Digital Platform](https://www.jadu.net/products) - [Jadu Connect](https://www.jadu.net/connect) - [Jadu Central](https://www.jadu.net/central) - [Ways to buy](https://www.jadu.net/buy) - Services - [Services Overview](https://www.jadu.net/services) - [Accessibility](https://www.jadu.net/accessibility) - [Development](https://www.jadu.net/development) - [Eye Tracking](https://www.jadu.net/eye-tracking) - [Hosting](https://www.jadu.net/hosting) - [Professional Services](https://www.jadu.net/professional-services) - [Support](https://www.jadu.net/support) - [Testing](https://www.jadu.net/testing) - [Themes](https://www.jadu.net/core-themes) - [Training](https://www.jadu.net/training) - [User Experience](https://www.jadu.net/ux) - [Web Design](https://www.jadu.net/webdesign) - Customers - [Customer Case Studies](https://www.jadu.net/case-studies) - [Local Government](https://www.jadu.net/local-gov) - [Higher Education](https://www.jadu.net/highered) - [Tourism](https://www.jadu.net/tourism) - [Partners](https://www.jadu.net/partners) - About - [About Us](https://www.jadu.net/jadu-1) - [Careers](https://www.jadu.net/careers) - [Corporate Social Responsibility](https://www.jadu.net/csr) - [Events](https://www.jadu.net/events) - [Jadu Academies](https://www.jadu.net/academy) - [Latest News](https://www.jadu.net/news) - [Our Story](https://www.jadu.net/our-story) - [People and Culture](https://www.jadu.net/culture) - [The Jadu Blog](https://blog.jadu.net/blog) Close navigation - Products - [Jadu Digital Platform](https://www.jadu.net/products) - [Jadu Connect](https://www.jadu.net/connect) - [Jadu Central](https://www.jadu.net/central) - [Ways to buy](https://www.jadu.net/buy) - Services - [Services Overview](https://www.jadu.net/services) - [Accessibility](https://www.jadu.net/accessibility) - [Development](https://www.jadu.net/development) - [Eye Tracking](https://www.jadu.net/eye-tracking) - [Hosting](https://www.jadu.net/hosting) - [Professional Services](https://www.jadu.net/professional-services) - [Support](https://www.jadu.net/support) - [Testing](https://www.jadu.net/testing) - [Themes](https://www.jadu.net/core-themes) - [Training](https://www.jadu.net/training) - [User Experience](https://www.jadu.net/ux) - [Web Design](https://www.jadu.net/webdesign) - Customers - [Customer Case Studies](https://www.jadu.net/case-studies) - [Local Government](https://www.jadu.net/local-gov) - [Higher Education](https://www.jadu.net/highered) - [Tourism](https://www.jadu.net/tourism) - [Partners](https://www.jadu.net/partners) - About - [About Us](https://www.jadu.net/jadu-1) - [Careers](https://www.jadu.net/careers) - [Corporate Social Responsibility](https://www.jadu.net/csr) - [Events](https://www.jadu.net/events) - [Jadu Academies](https://www.jadu.net/academy) - [Latest News](https://www.jadu.net/news) - [Our Story](https://www.jadu.net/our-story) - [People and Culture](https://www.jadu.net/culture) - [The Jadu Blog](https://blog.jadu.net/blog) - [Support](https://www.jadu.net/support) - [Jadu Library](https://www.jadu.net/library) - [Contact](https://www.jadu.net/contact) Search this site Search ![](https://blog.jadu.net/hubfs/Blog-Hero-Image-1.jpg) # Can I Speak to a Human? The Intersection of AI and Accessibility and what we learned at NAGW (National Association of Government Web Professionals) Published by [Suraj Kika](https://blog.jadu.net/blog/author/suraj-kika) on November 6, 2026, 2:47:47 PM GMT   "Can I speak to a human?" That remains to be seen. I spent a week this month at the [National Association of Government Web Professionals (NAGW)](https://nagw.org/) conference in Milwaukee. This is a conference organized by and with, the people who actually run government websites across the US - web administrators, content strategists, accessibility specialists, and the IT teams, who somehow keep an entire county online with digital public services. As well as time on the Jadu booth, I delivered a “Beyond the Age of Agents” session talking about what's next with Agentic experience. After just 2 days, I came away with a clear picture of where US County and local government digital teams actually are with AI, what they’re worried about, and why accessibility turned out to be the theme running through every single conversation. That alone was my biggest takeaway. Web teams are still 100% committed to accessibility and trying to do the right thing. One thing caught my attention: The Opening Keynote on Day 2: “Everything Government Web Professionals Need to Know About AI Since Yesterday.” Day 2 opened with Dr. Alan Shark, Senior Fellow at the Center for Digital Government and Associate Professor at George Mason University’s Schar School. His framing of the problem was the clearest I’ve heard so far and he delivered it in an engaging and thoughtful (and light-hearted) way. The public doesn’t see AI as a set of separate issues... they see one tangled knot of worries: sprawling data centers, energy costs, water shortages, AI slop, deepfakes, hallucinations, and students who can’t think without a prompt. Government web professionals are on the front line of those concerns, because the website is where citizens meet local authorities and service delivery. He offered a three-layer model: - **Layer 1: Personal productivity.** Can I verify the output? - **Layer 2: Enterprise applications.** Who is accountable? - **Layer 3: Global and autonomous systems.** Who can intervene? Almost everyone in that room operates safely in layers 1 and 2 - but the scary stories about agents going rogue (and he had a few, including one about a thousand agents that self-organized, elected representatives, and invented their own shorthand) belong to layer 3. Experimental, sophisticated systems. Not the consumer-grade tools a city web team is working with - but the public reads those stories, so you need to be able to contextualize them. ## The digital front door is changing The core of his talk was the shift from a website citizens must navigate to a service that answers questions and guides them through tasks. The old model: find, read, understand, decide, act. The new model: ask a question, get a personalized, guided response. He listed seven ways AI is changing government’s digital front door. Search becomes a conversation. Translation becomes a multilingual communication. Accessibility improves. Chatbots become digital employees. Personalization. Information becomes transactions. And agentic AI starts proactively identifying what a citizen is eligible for. I agreed... because that was precisely the point of my own session the day before - and is now becoming commonly understood. Search is no longer about results, it’s about answers. ## So, what are the risks and actions Web Teams need to do right now? Firstly, here are five risks that they can't ignore: 1\) Being provided with the wrong information 2\) Loss of authenticity 3\) Privacy and security (staff unknowingly pasting PII into open systems) 4\) Bias and unequal access across languages and disabilities 5\) Trust of citizens Citizens will ask - was this written by a person? Who is responsible if it’s wrong? Alan’s six practical steps were: 1. **Clean up content before feeding AI.** Contradictory pages, outdated PDFs, and duplicates can’t be fixed by a model. You need to clean up headings and tables and remove unnecessary bold text. 2. **Identify authoritative sources.** When two government pages disagree, which one wins? 3. **Establish an AI content policy**, distinct from an AI use policy. What data can AI access, what needs human review, what can never be entered, and who’s accountable when it goes wrong? It needs sign-off from the highest governing authority to carry weight. 4. **Start small and test.** A proof of concept asks, “does it work?” A pilot asks “does it scale?” They’ are different questions. 5. **Test like a citizen, not an IT professional.** “My basement flooded, who do I call?” not “retrieve emergency stormwater contact.” 6. **Always provide the opportunity to talk to a human.** His closing line stuck with me: “Our challenge is not to prove that humans can compete with AI. Our challenge is to ensure AI remains guided by our wisdom, human values, and human accountability.”   ## Where AI and accessibility meet The accessibility panel on Day 1 (Rob Carr from WebAIM, Chris Coleman from Chesterfield County, Jeremy Gonzalez from Siteimprove, Suzanne Ross Mantle from the City of Eureka, moderated by Colleen Killian from Champaign) set up the theme that ran through the whole conference, something I have communicated for a while: **If you make things accessible for humans, you make them accessible for machines.** [Chris Coleman](https://www.linkedin.com/in/hichriscoleman/) said Chesterfield County’s structured, accessible content has driven what he called “accidental AI success”. Less duplication and proper semantic structure improved their AI referral rates. **They didn’t optimize for AI... they optimized for people**. That is the intersection; semantic headings, plain language, one authoritative page instead of five near-duplicates, HTML instead of PDF. These aren’t just WCAG compliance checks; they’re the training data for every answer engine, assistant, and agent that will sit in front of your website over the next five years. A few more themes that came up over the course of the conference were: **AI is both good and bad at accessibility.** The consensus was strikingly consistent. **Good:** alt-tagging images as a first pass, generating metadata, identifying archivable pages, converting PDFs to HTML, breaking down dense WCAG criteria into plain explanations, programmatic checks like contrast ratios. **Bad:** understanding context and intent. AI describes an image literally, not purposefully. Nine times out of ten, the alt text is too long or contextually wrong. One panelist put it perfectly: “You can’t outsource understanding.” **“What does the tool do when it’s uncertain?”** This came up in the later AI session as the single most important question to ask any vendor. If a document AI accessibility tool hits a scanned page it can’t parse, does it guess, flag, or stop? Any vendor claiming “100% compliant without human review” was flagged as unreliable. I agree. **Stop creating PDFs.** Jeremy’s “one thing” recommendation. Also, from what I heard speaking to web managers at the booth, the hardest ongoing battle most web teams fight with their internal departments. I lost count of the number of people who told me their biggest accessibility problem was an archive of legacy PDFs that are now also their biggest AI problem. An answer engine trained on bad documents gives bad answers and indexing PDFs is expensive. **“Pick one thing.”** Suzanne’s motto was “anything worth doing is worth doing badly”. Start, iterate, don’t let perfection block progress. Chesterfield frames accessibility as “saving people time” rather than compliance, and coaches authors forward rather than penalizing them. Rob’s advice was “shift left and up”. Get leadership buy-in, governance, and resources. **Don’t lose sleep over WCAG 3.0.** It’s a minimum of three years from recommendation status, more like five or six before US adoption. Focus on hitting 2.1 and 2.2 and building up to 3.0.   ## What attendees told us at the booth This is the part I value most. Over two days we spoke with web teams from cities, counties, and state agencies across Illinois, Virginia, Minnesota, Colorado, Missouri, Michigan, Kansas, New Jersey, and beyond. Some patterns: **Governance is further along than expected.** One county has AI guidelines (not yet a full policy) and an AI committee. Their rule is for a hundred web editors to use AI as a tool to edit, convert, and check, never to generate text outright. Another city has had a formal AI policy for six months that restricts which models staff can use by role and data sensitivity, and they’ve already had to shut down unauthorized app installs. The shared principle is everywhere. Keep humans in-the-loop for anything high-risk or legally sensitive. One county won’t put any AI near election content yet, and a local forest preserve’s chatbot was pulled within weeks after giving wrong information on a hot-button issue. **Cost is a worry.** Usage-based AI pricing is very hard to budget for government. Several teams are investing in local, on-premise open-source models for cost control and security. One county is already running internal LLMs on its own hardware. The advice from the panel: buy the harness, not the model, and route tasks to cheaper models where you can. This is exactly why we built [Agent-Ex: Search](https://www.jadu.net/agent-ex) with response caching and a flat annual fee. Governments need to stream this like Netflix, not run an R&D budget. **Legacy platforms are a source of frustration.** Multiple teams described locked-down front ends with no design access, I-framed payment portals, and migrations that eat an entire year in time, resource and budget. Several were explicitly looking for a lower-friction way to offer something that sits on top of what they have rather than forcing a re-platform. **Accessibility is the deciding factor.** One state accessibility lead I met had previously helped us re-write our [VPAT (Voluntary Product Accessibility Template)](https://docs.jadu.net/central/content-portal/vpat/), flagged issues, and we fixed and reissued it. We treat the VPAT as a yardstick, not a marketing badge. And in at least one recent customer selection, accessibility was the reason we were chosen. Specialists at the booth wanted granular detail: can you lock down bold and italic per author, block fake headings, enforce review workflow, store manual WCAG test documentation inside the CMS? Yes, yes, yes, and yes. These are the questions that matter. **The human skills that grow in value.** The AI panel closed by asking what humans bring that AI doesn’t. The answers included; professional judgment and contextual expertise, empathy, relationship building and trust, and real-world communication. And finally, a framing that I loved was look for singles and bunts, not a holy grail”.   ## Beyond the age of agents: voice is the interface In my own session, I made an argument that I think ties all of this together. I can’t type. I never could. I now speak almost everything into my devices. When we type, we use keywords: “report pothole”. When we speak, we use natural language: “I’d like to report a pothole in front of my house.” The first works on most government websites. The second returns nothing. I demonstrated this live on some of the best-invested, award-winning government websites in the world, and they fell over on a plain-spoken sentence. That’s not a technology failure. It’s a behavior change. People no longer expect a page of results to search through. They expect an instant answer. Voice is becoming the primary way we interface with computers and will continue to be, and the agentic web is the thing that makes it work. The demo I showed was [Agent-Ex: Search:](https://www.jadu.net/agent-ex) a natural-language answer engine, fluent in over 70 languages, that only knows what’s published on your website. Ask it “can I claim benefits if I have more than £16,000 in savings?” and it gives you a concise, correct answer plus the top tasks to act on. Ask in Italian, get the answer in Italian. Ask by voice, hear it back. It’s the council officer who never sleeps, speaks every language, and knows everything about the organization. Crucially, it isn’t a chatbot. Chatbots retrofitted LLMs into a broken experience. We’d rather fix the experience: if you ask a question, you should get an answer. Have a conversation only when you need one. One of the most common questions asked in chatbots is, "*Can I speak to a human?"* And here is where it loops back to accessibility, and Dr. Shark’s six steps. The scope of Agent-Ex: Search is narrowed to content that has already been through editorial and approval workflow, which is how you shrink hallucination risk to almost nothing. If the agent is wrong, it’s because the content is wrong, and the content needs to be fixed anyway. Several customers have found it works as a self-fulfilling QA system. The agent surfaces the gaps, contradictions, and bad PDFs that were always there. Clean content, authoritative sources, a human escape hatch. Dr. Shark described the discipline. We built the product around it. ## Key take aways - **Accessibility is the AI strategy** Semantic, plain, deduplicated, HTML-first content is both WCAG compliance and the knowledge base for every agent you’ll ever deploy. Chesterfield proved it by accident. **Everyone else can do it on purpose.** - **Governance before tooling** The teams that are calmest about AI are the ones with a policy, a committee, and a clear line on what humans must review. - **Ask vendors what the tool does when it’s uncertain** If they can’t answer, walk away. - **Buy outcomes, not experiments** Government shouldn’t be funding R&D. It should be switching on standardized, cloud-delivered capability with predictable cost. - **Voice and answers are the new front door** Your search box is the most important real estate you own... and is the natural place for an Agent to live. Thank you to NAGW, to Dr. Shark, to the accessibility panel, and to everyone who stopped by the stand and told us honestly what’s hard. That honesty is the most valuable thing we bring home. [If you want to see Agent-Ex: Search on your own content, get in touch](https://www.jadu.net/request-a-demo). It takes about 30 days to deploy, and your website is already the training data. However, for it to be successful, your content and its accessibility really matter.   Tags: [Local Gov category](https://blog.jadu.net/blog/tag/local-gov), [Accessibility category](https://blog.jadu.net/blog/tag/accessibility), [Technology & A.I. category](https://blog.jadu.net/blog/tag/technology-a-i) [Back to Blog](https://blog.jadu.net/blog) Share: [Share on LinkedIn](https://www.linkedin.com/shareArticle?mini=true&url=https://blog.jadu.net/blog/can-i-speak-to-a-human-the-intersection-of-ai-and-accessibility-and-what-we-learned-at-nagw-national-association-of-government-web-professionals) [Share on Twitter](https://twitter.com/intent/tweet?url=https://blog.jadu.net/blog/can-i-speak-to-a-human-the-intersection-of-ai-and-accessibility-and-what-we-learned-at-nagw-national-association-of-government-web-professionals) [Share on Facebook](http://www.facebook.com/sharer.php?u=https://blog.jadu.net/blog/can-i-speak-to-a-human-the-intersection-of-ai-and-accessibility-and-what-we-learned-at-nagw-national-association-of-government-web-professionals) ## Leave a comment ## Related Articles ## [How accessible is your website?](https://blog.jadu.net/blog/how-accessible-is-your-website) With the digital world continuing to evolve, it’s key that you’re meeting accessibility standards... ## [A snapshot of the new ADA Web Accessibility Standards & Requirements Rule.](https://blog.jadu.net/blog/a-snapshot-of-the-new-ada-web-accessibility-standards-requirements-rule) What does the new Rule on the Accessibility of Web Content and Mobile Apps really mean? As part of ... ## [The Websites and Software Applications Accessibility Act](https://blog.jadu.net/blog/the-websites-and-software-applications-accessibility-act) There's been an exciting and potentially huge accessibility-related development in the US recently.... Like to know more? Get in touch [![site-logo-white](https://blog.jadu.net/hubfs/site-logo-white.png)](https://www.jadu.net/) ## Let's talk… Fields marked with \* are required - [Jadu Blog](https://blog.jadu.net/blog) - [Products](https://www.jadu.net/products) - [Customers](https://www.jadu.net/customers) - [Services](https://www.jadu.net/services) - [Support](https://www.jadu.net/support) - [Culture](https://www.jadu.net/culture) - [Contact](https://www.jadu.net/contact) - [Careers](https://www.jadu.net/careers) - [News](https://www.jadu.net/news) - [Legal](https://www.jadu.net/homepage/187/legal) Sign up for our emails © Jadu Ltd 2026 - [Follow us on LinkedIn](https://www.linkedin.com/company/jadu/) - [Follow us on Twitter](https://twitter.com/jadu) - [Subscribe to us on YouTube](https://www.youtube.com/channel/UCiICwjYZf8gENbHC2GlkAxg) - [Follow us on Facebook](https://www.facebook.com/JaduGroup) - [Follow us on Instagram](https://www.instagram.com/jadugroup/) - [Report a Security Issue](https://www.jadu.net/report-security-issue) - [Accessibility Statement](https://www.jadu.net/accessibility-statement) [![Jadu](https://www.jadu.net/site/images/site-logo-white.png?2.20.1 "Jadu")](https://www.jadu.net/) ## Sign up for our newsletters Fields marked with \* are required