Notes from Việt Nam on law, technology, languages, history, culture and travel.

Written for Chatbots Too: Việt Nam Tells Every Government Website to Speak JSON-LD

Most of Việt Nam’s government websites look much the same: a crest and a long agency name at the top, a carousel of news, a footer full of phone numbers. That look comes from regulation. Since April 2024, Thông tư 22/2023/TT-BTTTT has set the layout, functions and performance targets for every state agency portal.

On 15 October 2026 the Ministry of Science and Technology replaced it. Thông tư 61/2026/TT-BKHCN keeps most of the old checklist but adds a new reader. Government web pages must now be built for search engines and for artificial intelligence systems, using the same structured-data techniques that commercial publishers use to show up in Google results and chatbot answers.

The document

  • Number: Thông tư 61/2026/TT-BKHCN, “Quy định cấu trúc, bố cục, yêu cầu kỹ thuật cho cổng thông tin điện tử và trang thông tin điện tử”
  • Issued by: Minister of Science and Technology Vũ Hải Quân, on the proposal of the Director of the National Digital Transformation Agency (Cục Chuyển đổi số quốc gia)
  • Signed: 15 October 2026. It was published on the ministry’s legal-documents page on 16 October.
  • Effective: 1 December 2026
  • Size: 4 chapters, 17 articles and 6 appendices (39 pages)
  • Legal basis: Luật Chuyển đổi số 148/2025/QH15 and Nghị định 224/2026/NĐ-CP of 24 June 2026, which implements that law. The circular gives effect to Article 11(6) of the decree.
  • Replaces: Thông tư 22/2023/TT-BTTTT of 31 December 2023, which expires on 1 December 2026, subject to the transition rules described below.

It applies to the websites of ministries, provincial and commune-level People’s Committees and their departments. It also applies to every company that builds, upgrades, hosts or connects to those sites.

What is new

1. “Readiness for search engines and AI” (Article 8). This is the headline change. Pages must use a clear heading hierarchy, title and description tags, semantic HTML, keyword-bearing URLs and an XML sitemap. Important content must carry structured data so that “search engines and artificial intelligence systems” can interpret it accurately. Appendix III makes this concrete. The format is JSON-LD, the vocabulary is Schema.org, and dates follow ISO 8601. Two types are mandatory: GovernmentOrganization for the agency (name, URL, description, address, contact point) and NewsArticle/Article for news (headline, datePublished, dateModified, author, publisher, URL). Recommended types include GovernmentBuilding for one-stop service counters with opening hours, Legislation for legal documents (identifier, issue date, date of applicability, legal force, issuing body), DigitalDocument for administrative correspondence and Event. The appendix says the purpose is to help “generative AI” systems answer users’ questions correctly. Agencies are also encouraged to use AI on their own sites, for example for summaries, text-to-speech or personalised content, provided this complies with AI law.

The old circular did have structured content. It required Dublin Core meta tags, and Appendix II of the new circular keeps them, now with explicit references to ISO 15836-1:2017 and ISO 15836-2:2019. But TT 22/2023 contained no JSON-LD, no Schema.org and no mention of AI.

2. Tighter network security language (Article 12). HTTPS with at least TLS 1.2 stays. Under TT 22/2023 a site only had to be “ready to support” IPv6 and DNSSEC. Now it must use IPv6 and apply DNSSEC. New duties cover routine malware scans, system logs kept for verification, backup and restore plans, and a ban on posting broken or infected files.

3. New service-level targets (Appendix IV). The page-load thresholds carry over unchanged: First Contentful Paint ≤ 3 s, Largest Contentful Paint ≤ 4 s, Speed Index ≤ 5.8 s and Time to First Byte ≤ 0.6 s. So do the system targets: average response ≤ 2.5 s, slowest response ≤ 30 s, and active users ≥ 1/6 of concurrent visitors. Two targets are added: uptime of at least 99.5% and a technical error rate of no more than 1% of requests. Concurrency is now split by tier: ≥ 500 concurrent users for a portal and ≥ 200 for a component website.

4. Accessibility: a higher version, but a softer obligation (Article 11). TT 22/2023 required WCAG 2.0 as a minimum. The new text encourages WCAG 2.1 level AA or higher, applied “flexibly” to the features a site actually has, and suggests free open-source tools for self-testing.

5. A new monitoring platform with a longer data list (Article 13, Appendix VI). Sites used to embed the tracking code of the EMC system. Now they must connect to the ministry’s Nền tảng thống kê, đo lường, giám sát, đánh giá triển khai chuyển đổi số, the national platform for measuring digital transformation. It collects visits, page views, unique visitors, bounce rate and session length, as before. The list now explicitly adds traffic source, device details (device type, operating system, browser, screen resolution), visitors’ IP addresses and ISPs, uptime and outage times, and an open-ended category of “other data”.

6. Annual pass/fail ratings (Article 15). The Ministry of Science and Technology will assess state websites every year against the circular and publish the results. Each criterion is rated Đạt (pass) or Không đạt (fail). Agencies can respond to the National Digital Transformation Agency.

7. Flatter site hierarchies (Article 4). Websites of subordinate units, provincial departments and commune People’s Committees are to sit inside the parent ministry’s or province’s portal structure. A component site may not have its own sub-sites.

Who is affected and what to do

  • State agencies have until 1 December 2027 at the latest. Article 16 lets sites already running stay on the TT 22/2023 specifications while they upgrade, but for no more than one year after the effective date. Articles published before 1 December 2026 can keep their old metadata.
  • Vendors of government CMS platforms will do most of the work. The minimum function list in Appendix I includes role-based access, single sign-on tied to the agency’s identity system, modular plug-ins, event logs, versioned backups, Vietnamese search that ignores case and tone marks under TCVN 6909:2001, and subscriptions by email or messaging apps such as Zalo and Viber. JSON-LD templates for the six Schema.org types will now be a standard tender item.
  • Citizens and businesses should, in principle, get more accurate answers when they ask a search engine or chatbot about an office’s opening hours or whether a decree is still in force. Machine-readable legislationLegalForce and legislationDateOfApplicability fields are exactly what such tools need.

Open questions

  • Privacy. Article 14 tells site owners to protect users’ personal data, and Article 15 limits collection to what is “necessary”. At the same time, the central platform now explicitly collects IP addresses and device fingerprints. The circular does not say how long this data is kept or whether visitors are told.
  • Enforcement. Being rated “Không đạt” in a public ranking is the main sanction the circular provides. The circular itself does not penalise a site that misses the 99.5% uptime target.
  • Coverage. Only state bodies are bound. Public service units, state-owned enterprises and the news outlets of party and mass organisations are not named in Article 2.

Sources

This post is general information, not legal advice.