Building a Europe AI Can See: Switzerland’s National AI Profile

Switzerland’s five audited pages average 52.6 on the AI Retrieval Index, fifth of the eight countries compared in this series so far. Sweden scored highest at 68.8, followed by the United Kingdom at 64.4, Norway at 58.0 and Hungary at 55.8. Switzerland is ahead of Ireland and Germany, both at 51.8, and Greece at 49.4. All five Swiss pages are graded D.

This profile covers the English-language entry points of the Swiss federal government: the federal portal at admin.ch and four of the seven federal departments, Home Affairs (FDHA), Justice and Police (FDJP), Finance (FDF), and Environment, Transport, Energy and Communications (DETEC). When someone asks an AI assistant what the Federal Council decided this week, who leads a department or what a new federal consultation covers, these are the pages the answer should come from. All five sit on the admin.ch domain and read alike, so the result describes how the Swiss federal web presence reads to AI as much as it describes five separate institutions. The spread between the best and the worst page is 5 points, narrower than in the Sweden (6), United Kingdom (17), Ireland (19) and Norway (27) profiles.

The audits measure three things about each page: whether an AI system can read it, whether it can tell who published it, and whether it can judge that the content is current. All five audits were generated on 19 October 2026.

The main finding: the Swiss federal pages are easy for AI to read and give it very little to identify them with. None of the five carries JSON-LD, none shows a date signal an engine can read, and all five score 0 on freshness. Structure (99.0) and extractability (77.0) are second only to Sweden among the eight countries. Entity clarity (44.0) is the lowest of the eight, schema and metadata (23.0) is level with Hungary at the bottom, and freshness (0.0) is the lowest outright. All five pages raise the Structural Decay alert, which appeared on a single page in the UK profile. Entity connectivity and KG anchoring are 0 on all five, so no page declares an organisation, a person or a modification date in structured data.

This research is part of the Building A Europe AI Can See series, which also includes National AI profiles of Austria, Belgium, Croatia, Cyprus, Czechia, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Norway, Sweden, Switzerland, United Kingdom.

Read also the Ireland National AI Profile.

The Profile in One View

The spread between the best and the worst page is 5 points. All five pages share the same three sub-grades: A for Structure, B for Depth and F for Schema. DETEC is the lowest at 49, for two reasons the other four do not share: it is the only page without Open Graph tags, and the only one that fails the balanced heading density check.

Switzerland sits near the top on how its pages are built and at or near the bottom on what they declare. Structure (99.0) is second to Sweden, with Hungary third at 98.0. Extractability (77.0) is also second to Sweden, 1.0 ahead of Hungary. Entity clarity (44.0) is the lowest of the eight, 2.0 below Ireland. Schema and metadata (23.0) is level with Hungary at the bottom. Freshness (0.0) is the lowest, 9.0 below Hungary, and the only dimension where every page scores the same.

Against the UK, Switzerland has 3.0 more on structure and 3.6 more on extractability, and 13.0 less on entity clarity, 18.0 less on schema and 43.2 less on freshness, which adds up to 11.8 points on the index. Against Sweden it trails on every dimension, by 1.0 on structure up to 65.0 on freshness, and 16.2 points on the index. Against Ireland and Germany it is 0.8 points ahead on the index. It has 13.0 and 25.0 more structure and 3.8 and 8.6 more extractability, which more than offsets lower entity clarity (2.0 and 11.0 less), schema (3.0 and 8.0 less) and freshness (14.2 and 21.4 less).

Other indicators across the five pages:

• Entity graph stability averages 73.8. DETEC (80) reads “stable”. The other four read “moderate”, at 76 (portal), 73 (FDF) and 70 (FDHA and FDJP). The UK averaged 61.0, Sweden 72.2 and Norway 44.0. Primary entity, secondary entities, co-occurrence and clarity are at full marks on every page, schema is 0 of 20 on every page, and query alignment (0 to 10) accounts for the difference.
• Entity connectivity is 0 of 100 on all five pages.
• KG anchoring is 0 of 100 on all five pages. None carries a sameAs link to Wikipedia or Wikidata.
• E-E-A-T density is close to zero: 3, 3, 1, 1 and 4 (average 2.4, the same as the UK). The portal and FDHA score 3, FDJP and FDF 1, and DETEC 4. Trustworthiness records no signal on any page.
• None of the five pages carries JSON-LD. In the UK profile, three of five did. The audits report “No schema blocks found” on every page.
• No page has a machine-readable date. All five show the Structural Decay alert and score 0 on both freshness blocks.
• No page sets a canonical URL. Meta descriptions pass on all five and Open Graph tags on four. DETEC is the exception.
• 31 images across the five pages lack alt text. The portal has 9 (coverage 18%), FDHA 16 (16%), FDJP 1 (80%), FDF 2 (78%) and DETEC 3 (40%). No page has full coverage.
• Semantic richness averages 57.2. The portal and DETEC score 68, FDF 60, and FDHA and FDJP 45.

How the Five Engines Read the Swiss Pages

Per page, as score (estimated citation):

Claude (77.2) and ChatGPT (74.2) read the Swiss pages best, Copilot (52.0) is third, and Perplexity (46.4) and Gemini (34.8) read them worst. Copilot passes Perplexity on score, as it did in the UK profile. Against the UK, the Swiss pages score 1.0 lower on ChatGPT and 1.4 lower on Claude, but 13.4 lower on Perplexity, 15.6 lower on Copilot and 19.2 lower on Gemini. By the Inspector’s own description, ChatGPT and Claude reward topical depth and hierarchy, which the Swiss pages have. Perplexity, Copilot and Gemini lean on structured data and authorship signals, which they lack.

The citation estimates follow the same split. ChatGPT has the highest estimate at ~49.8%, followed by Claude (~45.8%), Perplexity (~32.2%), Copilot (~27.0%) and Gemini (~11.8%). Against the UK, ChatGPT is level (0.2 points higher), Claude is 3.2 points lower, Gemini 7.0 lower, Copilot 14.8 lower and Perplexity 19.6 lower. As in the UK profile, Perplexity has a higher estimate than Copilot despite a lower score.

The five pages score almost alike on every engine. Copilot returns 52 and ~27% on all five. ChatGPT ranges from 71 to 76, Claude from 74 to 79, Perplexity from 42 to 48 and Gemini from 30 to 36. FDJP and FDF score two points higher than the portal and FDHA on ChatGPT and Claude, and DETEC is the lowest of the five on four of the five engines.

Gemini finds the same gaps on every page. All five lack Article or Person JSON-LD and the Person and Organization pairing, and all five are flagged for a missing canonical. Alt-text coverage is flagged as a problem on the portal (18%), FDHA (16%) and DETEC (40%), and as partial on FDJP (80%) and FDF (78%). Gemini’s estimate is ~15% on the two pages with partial coverage, ~10% on the portal and FDHA, and ~9% on DETEC.

The findings for the other engines are identical on all five pages. Perplexity notes that there is no JSON-LD, so it cannot anchor citations, and that the missing canonical puts citation deduplication at risk. ChatGPT notes that entity signals are weak and credits the content depth and structure. Claude notes the absence of author or Person schema as an expertise signal and low contextual disambiguation. Copilot notes that there is no Article or Person schema, no valid JSON-LD block and no canonical URL, and advises adding both Person and Organization schema.

What the Five Audits Reveal

Structure is strong (99.0, range 95–100). Every page has a single H1, H2 sub-sections, logical nesting and passes the fragmented intent check. Four pages score 100. DETEC scores 95 because the balanced heading density check fails.

Extractability is solid but untidy (77.0, range 75–80). All five pass the word density, list structure and internal link density checks, and all five fail both the table and structured data check and the alt-text check. FDJP and FDF score 80 and the other three 75. The two higher-scoring pages are the two with the best alt-text coverage (80% and 78%, against 16% to 40%), although the audit shows the alt-text check as failed on all five. Of the 31 images missing alt text, 25 are on the portal and FDHA.

Entity clarity is low and uniform (44.0, range 40–45). All five pass the primary entity check and fail the Article, Person and Organization schema checks. Four pages pass the Open Graph title check and score 45. DETEC fails it and scores 40.

Schema and metadata is the second-lowest in the comparison (23.0, range 15–25). Four pages score 25: no JSON-LD and no canonical, but a meta description, Open Graph tags and no robots block. DETEC scores 15 because it also has no Open Graph tags.

Freshness is zero on every page (0.0). Each audit shows the “Structural Decay Detected” alert (“No date signals found – content age is unverifiable”) and fails all ten checks across its two freshness blocks. The audits count machine-readable date signals only. Several headlines quoted in the same audits carry dates in the text, among them “on Monday 5 October” and “on 30 September” on the FDJP page, “on 14 and 15 September” and “on 7 Oct” on the DETEC page, and a “popular vote on 29 November 2026” on the portal. None of these counts. The audits recommend adding datePublished and dateModified to JSON-LD on every page.

The trust layer is nearly empty. The portal records two authoritativeness signals, FDHA one experience and one authoritativeness signal, FDJP and FDF one authoritativeness signal each, and DETEC one expertise and four authoritativeness signals. Trustworthiness records none on any page.

The entity graph looks the same on every page. Primary entity (20 of 20), secondary entities (20 of 20), co-occurrence (15 of 15) and clarity (15 of 15) are full marks on all five, and schema is 0 of 20. Query alignment is 6 of 10 on the portal, 0 on FDHA and FDJP, 3 on FDF and 10 on DETEC, which gives 76, 70, 70, 73 and 80.

What the detector finds differs by page, and none of it is backed by schema. “Federal Council” is detected at 80% on the portal, FDJP, FDF and DETEC. FDHA yields “Home Affairs” (80%) instead. “Organization / Brand” is detected on all five, at 73% on the portal and 55% on the others. “Author / Expert” is read as a Person with no schema on FDJP and FDF (55%) and on DETEC (73%). “Product / Service” is read as a commercial entity on the portal (73%), FDF (81%) and DETEC (73%). “Artificial Intelligence” appears as a primary concept (73%) only on DETEC.

Query eligibility sits in a narrow band around 51%. The audits generated 155 queries across the five pages, against 222 for the UK and 119 for Sweden. Average eligibility is 51.5% on FDJP, 50.9% on the portal and DETEC, 50.7% on FDF and 50.1% on FDHA, and 50.8% overall. Scores run from 45% to 59%, and 79 of the 155 queries (51%) score exactly 50%, including 27 of the 31 on FDHA. The highest score in the set is 59%, for “Third national report on the risks of money laundering and terrorist financing?”, tagged FAQ on both the portal and FDF. The lowest is 45%. 117 of the 155 queries are tagged INFORMATIONAL.

Part of the query set comes from shared page elements and shared headlines. “What is further information?” appears on all five pages, “What is social Media?” on the portal, FDJP and FDF, “What is about us?” on FDHA, FDF and DETEC, and “What is media releases?” on FDJP and DETEC. The portal and FDF share the IMF, World Bank and G20 meetings in Bangkok query (54%) and the money-laundering report query (59%). The portal and DETEC share a headline on youth protection on communication platforms (49%). The same Federal Council news appears on the portal and on department pages, and none of the pages states in structured data who published it or when.

Automated detection mislabels some of what it finds. An announcement about the Swiss President stepping down is turned into two “How to” queries (HOW-TO, 53% and 57%). Seven queries on four pages end with the word “comparison”, tagged COMPARATIVE, among them the section heading “Current topics comparison”, and none of the pages carries comparison content. Of the 19 queries filed under the topic AI, only five are about AI, all on DETEC. The others include “Food safety and veterinary affairs”, “Too Big To Fail”, “Financial affairs” and “Greater legal certainty in plant breeding”, the last tagged COMMERCIAL. A cyber attack on a software supplier and a payment transactions headline are tagged TOOL-QUERY, and “Conversion Rate” appears as a topic on FDF and DETEC. These are artefacts of how the pages are read, and the Boundaries section says how far to trust them.

Page Profiles

Federal portal (admin.ch)

The audit reads the page as being about “Welcome to the official website of the Swiss government”. It scores 53, grade D – Poor.

Strongest findings. Structure is 100 and semantic richness 68, the joint highest in the set. Query alignment (6 of 10) is the second highest of the five pages, which puts the entity graph at 76. The 29 queries follow the Federal Council’s agenda: Swiss–EU relations (54%), the IMF, World Bank and G20 meetings in Bangkok (54%), international monetary assistance, a new cybersecurity law, the popular vote on 29 November 2026, neutrality, good offices for peace and Swiss engagement in international organisations (all 50%). The best scores are the money-laundering report (59%, FAQ) and the Parmelin announcement (57%, HOW-TO).

Weaknesses. There is no JSON-LD and no canonical URL, and freshness is 0 with the decay alert. Nine of the 11 images lack alt text (coverage 18%), and Gemini’s estimate is ~10%. The detector finds “Federal Council” (80%) and the German interface labels “Die Suche” (80%) and “Nach Datum” (64%) as named entities, reads President Guy Parmelin only as an “emerging concept” (56%) and finds “Product / Service” as a commercial entity (73%). Several headlines are tagged against their content, including “Ukraine: Federal Council approves additional Swiss private-sector projects” (TOOL-QUERY, topic AI), “New agreement on digital trade” (DIGITAL ANALYTICS) and “Subscribe to news” (COMMERCIAL).

What it means for AI systems. This is the front door of the Swiss federal government, and a query on the page says the English site carries only a selection of information about the Federal Council. An engine finds a clean hierarchy and current headlines on foreign relations, finance, security and votes. It does not find a structured statement of who publishes the page, a canonical URL, or a date it can read.

Home Affairs (FDHA, edi.admin.ch)

The audit reads the page as being about “The FDHA – at the heart of everyday life”. It scores 53, grade D – Poor.

Strongest findings. Structure is 100. The 31 queries cover the department’s areas: social insurance, health, culture, equality, combating racism, data and statistics, meteorology and climatology, archiving and information management, food safety and veterinary affairs, and diagnostics, virology and immunology. The page names Federal Councillor Elisabeth Baume-Schneider in five queries (profile, speeches, interviews and a signed greeting card among them). The flattest query set in the group: 27 of the 31 queries score exactly 50%, and the average is 50.1%.

Weaknesses. Sixteen of 19 images lack alt text (coverage 16%), the weakest record in the set, and Gemini’s estimate is ~10%. There is no JSON-LD and no canonical URL. Query alignment is 0 of 10, semantic richness 45 and E-E-A-T density 3 (one experience and one authoritativeness signal). The detector reads the head of the department only as “Elisabeth Baume”, cut at the hyphen, and picks up the German labels “Die Suche” and “Nach Datum” as named entities. A telephone number and an email address are read as queries, and one query is the German-language “diagnostik, Virologie und Immunologie”.

What it means for AI systems. The page lists what the department does, in a form an engine can parse. It does not declare the department as an organisation, link the minister to a record or carry a date.

Justice and Police (FDJP, ejpd.admin.ch)

The audit reads the page as being about “Federal Department of Justice and Police FDJP”. It scores 54, grade D – Poor, joint highest in the set.

Strongest findings. Extractability is 80, and only one of five images lacks alt text (coverage 80%), the best alt-text record of the five. The 25 queries are mostly headlines: Switzerland and India strengthening ties during a presidential visit, the dismantling of a ransomware group’s network in an international operation, the UN High Commissioner for Human Rights visiting Bern, protection status S continuing and the consultation on its future after March 2027, new anti-money-laundering rules, lifting visa restrictions on Ethiopia, and the adoption of the Swiss–EU stabilisation package. Most score 54%.

Weaknesses. There is no JSON-LD and no canonical URL, freshness is 0 and E-E-A-T density is 1. Query alignment is 0 of 10 and semantic richness 45. Some headlines are tagged against their content, among them “Greater legal certainty in plant breeding” (COMMERCIAL, topic AI) and a notice about fraudulent emails sent in the department’s name (DIGITAL ANALYTICS, topic AI). The detector reads “Beat Jans” as an emerging concept (56%) and “Author / Expert” as a Person (55%), both without schema.

What it means for AI systems. The page is the best-kept of the five at image level and reads as well as the others in structure. It also carries a warning about fraudulent emails sent in the department’s name, while giving an engine no structured statement of who the genuine publisher is.

Finance (FDF, efd.admin.ch)

The audit reads the page as being about “Federal Department of Finance FDF”. It scores 54, grade D – Poor, joint highest in the set.

Strongest findings. Extractability is 80, and two of nine images lack alt text (coverage 78%). Semantic richness is 60, the highest outside the portal and DETEC. It has two of the three FAQ-tagged queries in the set: “Third national report on the risks of money laundering and terrorist financing?” (59%, the set’s highest) and “Issue result for Confederation bonds?” (55%). Other headlines cover the automatic exchange of information with the EU, the IMF and World Bank meetings in Bangkok, international monetary assistance, new Confederation bonds, fiscal policy and the federal budget outlook.

Weaknesses. There is no JSON-LD and no canonical URL, freshness is 0 and E-E-A-T density is 1. “Product / Service” is the highest-clarity entity the detector finds on the page (81%), while the department itself is not listed. The head of the department appears as “Federal Councillor Karin Keller”, cut at the hyphen. “Too Big To Fail”, “Financial affairs” and a headline on systemically important banks are tagged DIGITAL ANALYTICS or TOOL-QUERY under the topic AI, and the audit also lists a cyber attack on a software supplier (“data leak confirmed”) as a TOOL-QUERY.

What it means for AI systems. The page covers bonds, banking and the budget, subjects where engines look for authoritative, dated figures. It gives them text to read and no structured date for a bond issue result or a risk report.

Environment, Transport, Energy and Communications (DETEC, uvek.admin.ch)

The audit reads the page as being about “Federal Department of the Environment, Transport, Energy and Communications DETEC”. It scores 49, grade D – Poor, the lowest in the set.

Strongest findings. It has the most queries (44), the joint highest semantic richness (68), the only “stable” entity graph (80, with query alignment 10 of 10) and the highest E-E-A-T density (4, from one expertise and four authoritativeness signals). It is also the only Swiss page with a clear AI theme. “Artificial Intelligence” is detected as a primary concept (73%), and five queries are about AI: the strategic direction for the Geneva AI Summit 2027 and the appointment of Fabiola Gianotti as Delegate, the AI for Good Global Summit in July, President Parmelin’s meeting with the Indian Prime Minister and the 2027 summit announced for Geneva, the AI summit in India, and AI and digitalisation at the Swiss IGF. Other headlines cover the IAEA General Conference, high-speed internet, FM radio licences, the Swiss Post universal service for payment transactions, the first SWEETER call for proposals, the timber harvest, the 2030 Agenda and critical infrastructure.

Weaknesses. It is the only page without Open Graph tags, which also fails its Open Graph title check, and the only one that fails the balanced heading density check, so structure is 95, entity clarity 40 and schema 15. There is no JSON-LD and no canonical URL, and freshness is 0. It has the lowest engine scores of the five on Perplexity (42), ChatGPT (71), Claude (74) and Gemini (30), and Gemini’s estimate is ~9%. Three of five images lack alt text (coverage 40%). Of the 10 queries filed under AI, half are about radio frequencies, domain names or railway rules.

What it means for AI systems. The page that carries Switzerland’s AI summit announcements is the lowest-scoring page in the set. An assistant asked about the Geneva AI Summit 2027 finds the announcements in text, with nothing structured to say who announced them or when.

What This Means

The Swiss profile is one profile. Five pages land within 5 points of each other, all graded D, with the same sub-grades and the same structural findings. The identical scores suggest a shared publishing approach across the federal sites, which the audit cannot confirm. Where the UK profile split on whether a page carried structured data, the Swiss profile has no split: none does. DETEC’s lower score comes from two missing basics, Open Graph tags and balanced headings, and not from a different architecture.

The cost shows in the engines that read structured data. ChatGPT and Claude score the Swiss pages within 1.4 points of the UK pages, because both countries build clean hierarchies. Perplexity, Copilot and Gemini score them 13.4 to 19.2 points lower, and their citation estimates are 14.8 to 19.6 points lower for Perplexity and Copilot, and 7.0 lower for Gemini. The structure that earns Switzerland second place on two dimensions does not make up for the schema, entity and freshness layers that are missing.

What the UK departments show. The three UK department pages that carried a valid GovernmentOrganization and BreadcrumbList block averaged 69.3, which is 16.7 points above the Swiss average. On structure (95) and extractability (75) they match or trail the Swiss pages, which score 99.0 and 77.0. The difference sits in schema (50 against 23.0), entity clarity (65 against 44.0) and freshness (53.0 against 0.0). That is a comparison of two sets of pages, not a test of one change, but it shows where the gap is.

The remaining work is small and shared. The audits point to the same short list on every page: set a canonical URL (all five), add an Organization block with sameAs links to Wikipedia or Wikidata, move datePublished and dateModified into JSON-LD, add alt text to the 31 images (25 of them on the portal and FDHA), and give DETEC its Open Graph tags and heading balance. Because the pages read so alike, changes made once at template level could reach all five.

Scope and Method

What was audited. Five Swiss federal pages, one page each, assessed with the AI Visibility Inspector and Srna Frameworks. All five audits were generated on 19 October 2026.

• admin.ch/en, the federal portal
• edi.admin.ch/en, the Federal Department of Home Affairs (FDHA)
• ejpd.admin.ch/en, the Federal Department of Justice and Police (FDJP)
• efd.admin.ch/en, the Federal Department of Finance (FDF)
• uvek.admin.ch/en, the Federal Department of the Environment, Transport, Energy and Communications (DETEC)

These five were chosen as the public door of the Swiss federal government: the portal and four of the seven departments, covering home affairs, justice and police, finance, and environment, transport, energy and communications.

What the research measures. Each report gives an AI Retrieval Index and grade, Structure, Depth and Schema sub-grades, and five scored dimensions: Structural Integrity, Data Extractability, Entity Clarity, Schema & Metadata, and Freshness & Decay. It adds scores for five engines (Perplexity, OpenAI / ChatGPT, Claude, Google Gemini, Microsoft Copilot) with estimated citation percentages, a Schema Intelligence section (validation, entity connectivity, KG anchoring, E-E-A-T density, semantic richness), query eligibility and an entity graph stability score.

Boundaries

• Five pages on one domain family, not five institutions. All five are subdomains of admin.ch, and they return identical scores on several dimensions (structure 100, entity clarity 45 and schema 25 on four pages) and identical engine findings.
• One page per entry, English version only. The portal’s own text says the English site is a selection of information, and the German, French and Italian versions were not audited and may read differently. Subpages, PDFs and the three departments not covered (Foreign Affairs, Defence, Civil Protection and Sport, and Economic Affairs, Education and Research) are outside the scope.
• Findings describe each page on the audit date. The headlines quoted are those on the pages on 19 October 2026, and several are truncated by the audit.
• Averages are simple means of the five page scores. The comparison figures for the UK, Sweden, Norway, Ireland, Hungary, Greece and Germany are those published in the United Kingdom profile and the earlier profiles.
• Each report produces two freshness blocks per page. Both score 0 on every page.
• The audit’s scoring is not fully reconciled. FDJP and FDF score 80 on extractability and the other three 75, although the same checks pass and fail on all five. Four pages score 25 on schema and metadata with three of five checks passed. The Inspector does not explain either.
• The Inspector reads pages without running scripts. If any of these pages adds JSON-LD or dates with JavaScript after load, the audits cannot show it. They report on JSON-LD and do not say whether a page uses microdata or RDFa instead.
• The “no date signals” result counts machine-readable signals only. Dates written in headlines are not counted, as noted above.
• Entity and query detection is automated and produces artefacts. Hyphenated surnames are cut (“Elisabeth Baume”, “Karin Keller”), German interface labels are read as named entities, a telephone number and an email address are read as queries, announcements are turned into “How to” queries, the word “comparison” is attached to headings and headlines, “Product / Service” is read as a commercial entity, and “Conversion Rate” is read as a topic. Queries and entity labels are quoted as the audit displays them.
• Every report lists a similar set of generic secondary content gaps, such as Large Language Model and Dark Funnel. The list is not interpreted here.
• Citation figures are the Inspector’s estimates, not observed citations in live engine answers.
• Headlines, names and figures in the page profiles are those detected on the audited pages. They are reported as page content and are not independently verified.

Calculate Your Potential Exposure

Reading poorly to AI has a cost. The AI Visibility Cost of Inaction Calculator lets you enter your own figures and see the exposure for your organization. <!– Paste the calculator embed/shortcode block here, as in the United Kingdom article.