Structure language versions
Give each locale its own URL. Google's guidance on using different URLs per language is explicit about not switching content by cookie or browser setting, and it warns that Googlebot crawls from the United States without an Accept-Language header, so dynamic swapping hides your work. MSF runs its global site at msf.org with a visible language selector and separate country sites for its offices across 40-plus countries, which is the pattern to copy: language choice in the header, never a forced redirect.
Decide early whether you're targeting a language or a region. French for readers anywhere is "fr". French for Canada specifically is "fr-CA", and you only add the region code when the page content actually differs by country. Google supports ISO 639-1 language codes with optional ISO 3166-1 Alpha 2 region codes, and every page in a set must reference itself and all its alternates.
Canonicals need the same discipline. Point each language version's canonical at itself, never at the English original, because a canonical pointing elsewhere overrides the hreflang relationship you just built. Verify the current syntax and supported codes against Google's live documentation before you ship, since the details change and stale blog posts outnumber accurate ones. Multilingual AI visibility collapses quietly when a single template pushes every canonical back to the source language. Technical SEO supports this foundation.
Multilingual AI visibility roadmap
Run this as a sequenced release with named owners and gates, because a sitewide translation project stalls at 40% complete.
-
Score the shortlist and confirm reviewers (communications director, two weeks)
-
Lock the approved facts sheet and glossary (programme lead plus M&E, one week)
-
Draft the four-page journey against the brief (writer or vendor)
-
Qualified review and sign-off (named local reviewer, gate: no sign-off, no publish)
-
Technical build with locale URLs and hreflang (web owner)
-
Pre-launch checks across links, forms, and claims (web owner, gate)
-
Measure for one quarter before approving a second language (communications director)
Assess demand and capacity
Score every candidate language against the matrix, then strike any language without a named reviewer and quarterly maintenance hours. Confirm the reviewer in writing, with their qualification recorded against the risk level of the content. Pick one opportunity where demand and capacity line up with search intent by language.
Prepare and localize
Finalize the source facts before anyone writes, because changing a figure mid-build means re-reviewing every page it touched. Complete the full journey and apply the brief and glossary to each page. Send the whole set to the reviewer together so inconsistencies surface in one pass. International mission localization fails most often at this step, when a form or a confirmation email gets left in the source language, which breaks multilingual AI visibility.
Test and expand
Before launch, check that every internal link stays inside the locale and that forms submit and confirm in the right language. Confirm that page titles and meta descriptions are localized and that the language switcher is visible on mobile. Every claim must match the approved facts sheet. Then stop. Expand to a second language only when the first produces enquiries you can handle and you still have review hours left over.
Own every update
An update with no owner becomes a contradiction within a quarter.
-
Eligibility criteria change: Source-page owner: Programme lead. Language reviewer: Named local reviewer. Deadline: 5 working days. Escalation: Country director. Action: Update all versions or unpublish the affected section.
-
Annual figures published: Source-page owner: M&E lead. Language reviewer: Named local reviewer. Deadline: 10 working days. Escalation: Communications director. Action: Update glossary figure row, then all versions.
-
Contact or office change: Source-page owner: Operations. Language reviewer: Named local reviewer. Deadline: 2 working days. Escalation: Country director. Action: Update journey endpoint and confirmation copy.
-
Legal or status wording change: Source-page owner: Legal. Language reviewer: Qualified legal reviewer. Deadline: Before publication. Escalation: General counsel. Action: Hold publication until sign-off.
Unpublishing is a legitimate action. A section withdrawn while it's being corrected beats a section that quietly contradicts your English page for six months. International mission localization is an ongoing obligation attached to named people responsible for multilingual AI visibility.
Measure languages separately
Aggregate traffic hides everything that matters here. Split your reporting by language and geography, and track useful visits and relevant enquiries for each. Completed key actions and search appearances follow the same split. Google Search Console's performance report filters by country, and since October 2026 it supports combined traffic for multiple countries, which makes regional grouping possible without spreadsheet work.
AI answers need their own log, because multilingual AI visibility in one language tells you nothing about another. Weglot's analysis of 1.3 million AI citations found untranslated Spanish sites earned 431% more AI Overview citations for Spanish queries than English ones, and that translated sites gained across both. Record each check by question and language with the platform and date attached, and re-run the same questions monthly. Visibility reports can support this measurement.
Keep the measurement honest about who you're reaching. GSMA reports that 63% of people in Africa covered by mobile broadband still don't use mobile internet, so your page data describes the connected portion of your audience and nothing more. Multilingual AI visibility is one channel inside a wider communication plan, and search intent by language is measured.
Choose the first expansion
Pick the language where people you serve are already asking and a qualified reviewer has said yes, with hours on your team to maintain four connected pages for a year. Build that journey completely and measure it for a quarter. Expand only when the numbers and the capacity both hold.
Snoika Foundation works with NGOs and public institutions on exactly this problem, from language scoring through review workflows and measurement. Book a call with our team to plan your first complete journey and build multilingual AI visibility your organization can sustain.