Building a Europe AI Can See: Belgium’s National AI Profile – One Federal Site Proves the Fix

Five federal homepages, one success, and four sites that look the same to a machine.

• National average: 49.8, Grade D. The five sites span 42 points (35 to 77), against 11 for Austria. One site reaches Grade B, none reaches C, four sit in Grade D. Without the B, the average falls to 43.0.
• Four of five have zero JSON-LD. Only health.belgium.be carries structured data, one valid Article block, and it scores 77. No site has Organization or Person schema, and no site links to Wikidata or Wikipedia (knowledge-graph anchoring is 0 on all five).
• Identical scores on four sites. Entity clarity is 40 and schema and metadata is 10 on each of the four D-grade sites. Ties that exact point to a shared template rather than four editorial decisions.
• The lowest score is social security, at 35. It has two competing H1 tags, 272 words, and the lowest score on every one of the five engines.

The profile in one view

Scores are the Inspector’s dimension scores (0 to 100) for each homepage. Read across a row to see a national pattern; read down a column to see an institution’s character.

Shading: 75 and above strong, 50 to 74 moderate, 20 to 49 weak, below 20 absent. Trust and authorship is the Inspector’s E-E-A-T density score.

The structure is as sound as Austria’s: four of five score 100. The 30-point gap between Public Health and the next-best site comes entirely from schema (65 vs 10), freshness (92 vs 0) and entity clarity (75 vs 40).

Where Belgium needs to improve

Ranked by how many institutions are affected and how much one fix would unlock, not by the size of the gap alone.

1. Structured identity (all five)

Four sites have no schema at all. Public Health has an Article block, but no Organization or Person entity behind it, so its author and publisher are named without being anchored. No site has sameAs links, so an AI system must guess what “FPS Justice” is rather than read it. The highest-leverage fix is to copy what Public Health already outputs and add an Organization block with sameAs to Wikidata. In a state with three or four official languages, alternateName can join the variants of each institution’s name. FOD Justitie appears on the Justice page, which the engine parses only as “Home”.

2. Dates and accountability (four of five)

Freshness scores run 0, 0, 13, 0, and 92. Justice is the only D-grade site where a date was found, and the Inspector reads it as 2,853 days old, about 7.8 years, with no update signal. Home Affairs still surfaces a newsletter edition from October 2015 among its candidate queries. Residence documents, elections, asylum procedures and social security rules all change. Undated pages read as stale to engines that weigh recency, and Public Health shows the CMS can already emit dateModified.

3. Named responsibility (all five)

E-E-A-T density is 0 on four sites and 17 on Home Affairs, the only page with any signals (one experience, two trust). Public Health’s Article names an author and publisher, but the density score is still 0. Across engines, Claude scores highest on all five sites (average 69.6), then ChatGPT (63.4), Copilot (57.2), Perplexity (42.2) and Gemini (38.6). Gemini and Perplexity, the engines that lean hardest on schema, dates and authorship, are the weakest, which ties directly to fixes 1 and 2.

4. Language and interface text in the content layer (four of five)

Belgium’s multilingualism shows up as noise. Justice carries French labels (“Thèmes”, “Nouvelles”) and a Dutch institution name on its English URL. Home Affairs’ English URL returned German-language content, with English template strings (“Share Footer EN”, “Footermenu 3 EN”) among the candidate queries. Social Security lists “website in Dutch” and “website in French” as queries. Cookie-banner phrases (configure, accept, refuse) appear as queries on the portal, Justice and Social Security. Language switchers and consent text that leak into the content layer dilute what each site is about.

5. Metadata hygiene, alt text and depth (all five)

Meta descriptions are missing on four of five, and the Inspector flags canonical and meta together, so canonical status needs a source check. Alt text is incomplete on Public Health (7 images, 46% coverage), Justice (2) and Social Security (1). Every page is under the Inspector’s 600-word target (272 to 448 words), and Social Security is below the 300-word minimum.

Belgium has already built the machine-readable layer once, on one site. What is missing is the decision to use it everywhere.

Institution profiles

belgium.be, the federal portal – Grade D – 45/100

The front door that introduces itself as “Home”

The gateway to federal information is parsed with a generic label. Council of Ministers decisions, federal government jobs, Your Europe, and services for family, residence documents and births all sit on this page, but none is tied to a named institution.

Strengths: Perfect structure (100), the highest semantic richness in the set (46/100) and the widest civic range of the five.
Improve first: Organization and WebSite JSON-LD with sameAs, dateModified on Council of Ministers decisions, then a meta description.
What AI sees: Entities only as generic “Product” and “Organization” labels, with no named institution. Cookie-banner phrases sit among the 16 candidate queries, and co-occurrence scores 0/15.
Why it matters: The page describes the government meeting every Friday, which is exactly the kind of dated, recurring fact that needs a machine-readable date.

ibz.be, FPS Home Affairs – Grade D – 45/100

The English page that answers in German

The audited /en URL returned German-language content and is parsed as “Home DE”. The only person entity in the set appears here, a first name with a communications role. It also carries the only E-E-A-T signals in the five (17/100).

Strengths: Perfect structure and the best extractability in the set (57).
Improve first: Serve English content on the English URL, remove the “Footer EN” template strings, then Organization schema with the full institution name and dates.
What AI sees: German headings (Unser Auftrag, Sicherheit, Asyl und Migration, Wahlen), an October 2015 newsletter item, and Civil Security, Crisis Centre and Immigration Office as queries but not as detected entities. Query alignment scores 0/10.
Why it matters: Identity, elections, asylum and crisis management are high-stakes and time-sensitive, and the page is the least legible about which institution is speaking.
Check: Confirm whether the German content on /en is a real language fallback or an artifact of the audit.

justice.belgium.be, FPS Justice – Grade D – 47/100

The oldest signal in the set

The only D-grade site with a detected date, and the date is nearly eight years old. It also has the lowest entity graph stability in the set (15/100, “unstable”) and no primary entities detected at all.

Strengths: Perfect structure and the highest score of the four D sites.
Improve first: Verify which date the Inspector is reading, then add Organization schema with alternateName in Dutch and French, dateModified, and a meta description. Two images need alt text (86% coverage).
What AI sees: No entities and no topic clusters. Candidate queries are dominated by Court of Cassation items (language use in the Brussels periphery, oath-taking and installation), plus French and Dutch labels and cookie text.
Why it matters: A citizen asking an AI assistant about the justice system gets an institution that cannot be named or dated.
Check: The date came from text or meta tags only, so it may be a template or footer date rather than a content date.

socialsecurity.belgium.be, FPS Social Security – Grade D – 35/100

Highest entity stability, lowest overall score

The lowest overall index and the lowest score on all five engines (Perplexity 28, ChatGPT 51, Claude 55, Gemini 25, Copilot 42). It is the only site with a structural warning, two H1 tags, and its 272 words fall below the 300-word minimum.

Strengths: The best entity graph stability in the set (60/100) and the only non-zero co-occurrence score (4/15). “Social Security” is detected as a concept.
Improve first: One H1, a body of 300+ words saying what the site is for and whom it serves, then schema, dates and one alt text.
What AI sees: Social Security, Federal Public Service and Artificial Intelligence as concepts, plus “Conversion Rate” as a topic, which is a tool-taxonomy artifact. Candidate queries include BELMOD (a microsimulation tool), Brexit guidance for Belgian rights in the UK, and the Dutch and French language links.
Why it matters: This is the national overview of the social security system, and it is explained on the thinnest page in the set.

www.health.belgium.be, FPS Public Health – Grade B – 77/100

The proof inside the state

The only site that names itself (“FPS Public Health”) and the only Grade B. It has a valid Article block with 7 of 8 fields (headline, author, publisher, datePublished, dateModified, description, image), a freshness score of 92, and the best score on every engine (Perplexity 72, ChatGPT 73, Claude 80, Gemini 78, Copilot 78). Its estimated Perplexity citation share is about 64%, against 24 to 27% on the other four.

Strengths: Working date pipeline, schema 65, entity clarity 75.
Improve first: Organization and Person JSON-LD with sameAs to Wikidata (knowledge-graph anchoring is 0/100), alt text for seven images (46% coverage), then depth above 600 words (currently 310).
What AI sees: The same name detected as a Product, a Person and an Organization. Headlines such as “One World, One Health” and a plain-language cleaning-safety item sit alongside public consultations that the tool tags as “digital analytics”.
Why it matters: It shows the fix works on a real Belgian federal platform without touching structure. The remaining gap is entity identity, not capability.

One national fix, five local ones

The same gaps repeat closely enough that most of the uplift is a template-level job.

Scope and method

Each institution’s English entry page (/en) was audited once on 4 October 2026 with the AI Visibility Inspector (v1.9.3) and the Ivica Srncevic Framework. This is a homepage snapshot, not a full-site audit. Belgium has three official languages, and the Dutch, French and German versions were not audited, so results may differ. The sample covers five federal sites, not the whole federal administration, and the regional and community governments are outside it. The shared-template reading is an inference from identical scores and repeated patterns, and should be confirmed in the page source. Engine citation percentages are the tool’s estimates and are best read relatively. The Inspector’s “content gaps” list and several commercial labels (such as “Conversion Rate” on Social Security) use vocabulary that does not suit government sites and were left out of the analysis.

Profile network. For the other National AI profiles country, see Austria.
Also see how Belgium TOP 10 enterprises perform in AI.

Calculate your potential exposure below