
Right-to-Left Knowledge Base Design: A Complete Practical Guide
Right-to-Left Knowledge Base Design makes a help center usable when its main content follows a right-to-left writing direction. It connects navigation, article layout, mixed-language text, search, and support forms so readers can find an answer and complete a task without losing context or misreading technical details.
For a U.S. documentation or product team adding Arabic or Hebrew support, the difficult work often appears between components: an English error code inside a translated sentence, a search result in the wrong language, or a contact widget that follows different rules from the article. Translating the article body addresses only part of that experience.
This guide explains the design decisions, implementation patterns, and release checks for an RTL knowledge base. The examples use English labels and technical identifiers. Actual language quality, text shaping, and mixed-script behavior must be evaluated with representative localized content.
Last verified: November 19, 2026.
Start with a complete support journey
Choose one representative task before redesigning the entire help center. For example, a reader searches for an error, opens a troubleshooting article, copies a case identifier, and contacts support if the instructions do not resolve the issue. Map the screens, overlays, and confirmation states involved in that task.
Define the supported language, product version, and audience for each route. An internal employee knowledge base may have different authentication and search boundaries from a public customer help center. A locale, a language, and a writing direction are separate settings; none should be used as a shortcut for the others.
Writing direction follows the script being used. W3C notes that some languages can be written in more than one script, so a broad country or language assumption is not a reliable implementation rule. Establish direction from the actual content and supported locale configuration. See W3C guidance on languages using RTL scripts.
Give each part of the experience an owner. This prevents an article editor from being asked to fix a search configuration problem or a translator from having to compensate for a broken layout.
| Layer | Decision to document | Typical owner |
|---|---|---|
| Locale and routing | Which language and product-version article opens from a direct link or language switch? | Documentation operations and engineering |
| Shared interface | How do navigation, drawers, alignment, spacing, and focus behave? | Design system and frontend teams |
| Article content | Which passages, identifiers, screenshots, and instructions need localized treatment? | Content and localization teams |
| Search | Which content is indexed, filtered, and returned for each language? | Search owner and support operations |
| Forms and widgets | Which component controls labels, input direction, validation, and confirmation? | Support platform owner |
| Publishing | Which language and direction settings survive editing, translation, and publication? | Content tooling and quality assurance |
Use the same illustrative task throughout implementation review. In the examples below, Atlas Sync (v2.4), KB-1042, and E-104 are fictional test values. They provide consistent punctuation and identifier patterns without suggesting a real customer incident.
Separate language, direction, alignment, and isolation
These mechanisms solve different problems. A page can identify its language correctly while using the wrong base direction. A field can be right-aligned while its punctuation still behaves incorrectly. Choose the mechanism that matches the boundary you are configuring.
| Mechanism | Purpose | Knowledge-base use |
|---|---|---|
lang | Identifies the language of content. | The page language and passages written in another language. |
dir="rtl" or dir="ltr" | Establishes a known base direction. | The localized page or an opposite-direction content region. |
dir="auto" | Derives direction from the text using a directional heuristic. | User-entered text whose direction is not known in advance. |
text-align: start | Aligns content at the start edge of its writing direction. | Article paragraphs, labels, and callouts. |
bdi | Isolates inline text from surrounding bidirectional text. | Names, product strings, identifiers, and other inserted values. |
For an Arabic-language page, the root declaration can be <html lang="ar" dir="rtl">. For a Hebrew-language page, use the appropriate language tag, such as he. These declarations describe actual localized content; changing the attributes does not translate an English article. W3C recommends declaring the page language on html and marking language changes at the relevant element. See declaring language in HTML.
Set known content direction in HTML. Right alignment alone does not establish the same text semantics. When an RTL help-center shell displays a full English fallback passage, give that passage its own English language and LTR direction rather than relying on the surrounding page settings.
Use automatic direction for unknown text
As explained in MDN’s dir reference, auto uses directional characters to determine direction. It does not recognize the language or infer what the reader intended. A query that starts with a Latin product name can choose LTR even when the remaining words use an RTL script.
Evaluate this behavior with the actual values users enter. Use explicit direction for a known LTR identifier or command; consider automatic direction for an unknown name or free-text subject. Do not apply one setting to every field simply because the page is RTL.
The HTML specification defines bdi as an isolation boundary. Its default direction is automatic rather than ordinary inheritance from its parent. Isolation does not identify the language, so add language metadata separately when the content calls for it.

Keep mixed text and technical values intact
Bidirectional text combines runs with different writing directions. In a knowledge base, the important cases include product names, version numbers, email addresses, URLs, file paths, and identifiers embedded in a translated sentence. Problems often become visible at punctuation boundaries rather than in the letters themselves.
Wrap the complete opposite-direction phrase, including punctuation that belongs to that phrase. For example, keep the parentheses in Atlas Sync (v2.4) with the product string. Leave sentence punctuation outside when it belongs to the surrounding sentence. W3C’s inline bidirectional markup guidance explains why tight boundaries matter.
<bdi dir="ltr" lang="en">Atlas Sync (v2.4)</bdi>
<bdi dir="ltr">KB-1042</bdi>
<bdi dir="ltr">https://example.com/help/KB-1042?code=E-104</bdi>
These are separate inline fragments for insertion into appropriate article content, not a complete localized page. Test them inside real translated sentences, including a trailing period, paired parentheses, an adjacent number, and two technical values separated by a slash.
Keep text in its logical typing order. W3C distinguishes the stored character sequence from the visual order produced during rendering. Reversing a string or its digits in the source to make one screenshot look correct can damage copying and downstream use. See logical versus visual text ordering.
Treat display and copying as separate checks
Copy the complete value into a plain-text destination and compare it with the intended source. Check punctuation, capitalization, spaces, and line breaks. A URL that looks intact can still be unsuitable for pasting if an editor or copy button adds characters. Test the actual copy mechanism, not just mouse selection in an authoring preview.
For HTML content, prefer explicit markup boundaries to scattered invisible directional controls. W3C documents Unicode controls for situations where markup is unavailable, such as some plain-text contexts. Use those controls deliberately and verify what the receiving system gets. See Unicode controls for bidirectional text.
Do not use bdo as a general repair for mixed text. It deliberately overrides normal bidirectional behavior and requires an explicit direction. A routine product-name or case-identifier problem usually calls for correct boundaries and isolation, not an override.
Build navigation and layout that work in both directions
Preserve the meaning of navigation
A breadcrumb describes ancestry. Keep its underlying data in the intended ancestor-to-current-page sequence, then render that hierarchy appropriately for the locale. Verify the separators and linked destinations together; reversing a data array merely to change the visual starting edge can change the meaning of the trail.
A category grid expresses information priority. Decide which category should be encountered first, then check the desktop grid and mobile stack. Related articles are recommendations; previous and next controls represent a sequence. Label those controls according to their purpose and keep their destinations in the relevant product version and language.
For a table of contents, verify that each link reaches the correct heading and that a sticky header does not conceal the destination. A mobile navigation drawer needs its own opening, dismissal, focus, and return behavior. Moving a desktop sidebar to another edge does not settle those interaction decisions.
Use logical CSS for direction-dependent spacing
CSS logical properties express layout relative to writing mode and direction. In a horizontal RTL layout, inline-start is the right edge; in horizontal LTR, it is the left. This makes a leading callout border or start-side padding easier to share across localized templates.
.kb-article {
max-inline-size: 48rem;
margin-inline: auto;
padding-inline: 1rem;
text-align: start;
}
.kb-callout {
border-inline-start: 0.25rem solid #1254a1;
padding-inline-start: 1rem;
}
This CSS illustrates spacing and decoration. It assumes that language and direction are already set correctly in the document. Review each existing left/right rule by purpose: a reading-order-dependent border is different from an annotation pointing to a fixed location in a product diagram.
Avoid adding row-reverse automatically. A horizontal RTL flex container already lays out row from the right. Reversing it changes the visual progression again, without rewriting the source sequence. MDN’s flex ordering guidance explains the implications for visual order and keyboard navigation.

Mirror directional controls according to meaning
Review back and forward navigation, disclosure indicators, and other directional controls against the design system and the task they represent. Do not flip every asset. Microsoft’s mirroring guidance identifies familiar media controls that retain their orientation across LTR and RTL interfaces.
Keep logos, screenshots, maps, and product diagrams faithful to what they represent. Capture the localized product interface when available. If readers must use an English product screen, explain its labels in the article and preserve the screenshot as an accurate view of that screen.
Make article content readable at every width
Review typography with representative content
Evaluate the chosen fonts with real headings, emphasized text, numbered procedures, links, and any diacritics used in the supported content. Review line height and component height together. A fixed-height title card may accommodate a short English label yet fail with a longer translated heading.
Keep procedural steps in their intended source order. Check list markers, continuation lines, nested lists, and callouts at narrow widths. Do not use manual spaces or line breaks to force alignment; the result should remain readable when the font, viewport, or text size changes.
Use locale-aware formatting policies for dates, times, measurements, and amounts where those values occur. Confirm the actual product requirements with the localization owner. RTL direction alone does not determine the reader’s region, preferred numerals, calendar, or terminology.
Give tables and code their own direction decisions
Table direction and cell-content direction are separate choices. As W3C’s table examples demonstrate, direction affects column progression and inherited text direction. Decide which column begins the comparison, then verify that headers remain associated with the intended values.

A localized table can contain an LTR SKU, version, or command in individual cells. Keep those values intact, and avoid reversing the dataset just to move the first column. At narrow widths, retain access to essential columns and make any table scrolling understandable.
Give a command or code sample an explicit direction appropriate to its contents. This small fragment keeps the command LTR without changing the surrounding article:
<pre dir="ltr"><code>curl --version</code></pre>
Let long URLs and filenames wrap when needed without changing their stored values. MDN documents that overflow-wrap: anywhere allows an otherwise unbreakable token to wrap without inserting a hyphen. Code that needs preserved line structure may be better served by its own scrollable region. Check copying in either presentation.
Use images when the visual state contributes information: a localized menu, a direction-sensitive table, or a form error. Keep captions and alternative text in the article’s language. A screenshot should illustrate the instruction beside it rather than repeat a paragraph or hide the relevant control inside a full desktop capture.
Design search for both input and retrieval
Search has two separate requirements: readers must be able to enter and inspect a query, and the system must return useful content. Input direction addresses the first. Publication status, language filters, indexing, and query processing affect the second. A neatly aligned search box does not establish that the right article is discoverable.
Define a small, purposeful query set
- A localized task phrase: the wording a reader would use to describe the problem.
- An exact product label: an English label that remains visible in the software.
- A mixed-language query: localized wording combined with a product name or version.
- A technical identifier: an exact error code or case-related string containing punctuation.
For each query, record the intended article and language before testing. Repeat the query under the relevant filter settings. Review suggestions, result titles, snippets, highlighted matches, and breadcrumbs as separate text boundaries; a correct input field does not guarantee a correct result template.

Ask the search owner how punctuation, spelling variants, and language-specific analysis affect retrieval. Test technical identifiers both as readers copy them and as they type them. Add approved synonyms from actual support terminology rather than assuming that translation alone covers the phrases people search for.
Check platform behavior and empty results
Document360 provides a concrete example of why input behavior needs its own check. Its search documentation says an RTL-only language selection uses RTL input and alignment, while a mixed RTL/LTR selection follows the default language. This describes the search field; it does not guarantee relevance or coverage.
Zendesk’s help-center search documentation warns that the article translation must match the help-center locale for search to behave as documented. Confirm both the visible language and the stored locale when diagnosing missing results.
Design an empty-results state that retains the query and offers a useful next step, such as adjusting a filter, browsing a relevant category, or contacting support. Keep any query echoed in the message properly isolated. If cross-language results are offered, label their language instead of silently implying that they match the selected locale.
Make forms and support widgets usable
A reader moving from an article to support should retain the useful context of the task. Where the integration supports it, carry the article reference, relevant product version, selected language, and error identifier into the request flow. Let readers inspect that context before submitting it.
Choose direction per field. A free-text subject may need automatic direction, while a case identifier with a defined Latin format can use LTR. The following English-labeled fragment illustrates that distinction:
<label for="support-subject">Subject</label>
<input id="support-subject" name="subject" dir="auto">
<label for="case-id">Case ID</label>
<input id="case-id" name="case_id" dir="ltr">
In the deployed form, use the interface language for labels, help text, validation messages, and confirmation. Keep visible labels available after the placeholder disappears. Test cursor movement, selection, pasted values, filenames in attachments, and the submitted value. Review multiline fields with content beginning in different directions.
Follow the entire error-recovery path. Identify the field with a problem, explain how to correct it, preserve valid entered data, and keep the submit action reachable. WAI’s form notification guidance includes field-level messages and error summaries that help users locate corrections.

Treat embedded widgets as separate components. For example, Zendesk’s multilingual configuration guidance distinguishes standard request-form text from untranslated custom fields, which can fall back to the Support account’s default language. Verify your actual form rather than assuming every label follows the article locale.
After submission, check that the confirmation identifies what happened and presents any request number intact. A successful-looking button state is not enough if the reader cannot recognize the identifier or understand the next step.
Preserve context when readers change languages
When an equivalent translation exists, keep the reader on the corresponding article and product version. A language switch that sends every reader to the homepage discards useful task context. Provide a visible language choice and make the current content language understandable, including when the interface and article body differ.
Distinguish missing, unavailable, and outdated content
Define the reader-facing response for each state: the language is not enabled, the article has no published translation, the translation exists but is outdated, or a translated parent page is missing. These conditions need different fixes; treating all of them as a generic localization failure makes diagnosis slower.
Platform rules matter. Zendesk’s localization documentation states that translated articles need their parent section and category translated into the same language. A published article is therefore not the only item to verify when a reader cannot reach the localized content.
Document360’s multilingual documentation distinguishes an unpublished secondary-language article, which can display default-language content, from a published translation that remains visible even when outdated. Do not assume a fallback mechanism automatically replaces stale translations.
For any fallback, identify the available language and provide a useful next action. Give a full English fallback passage its own lang="en" and dir="ltr" boundary within an RTL shell. Keep the language selector available so readers can recover from an unwanted choice.
Make public language versions discoverable
For a public knowledge base, plan language-specific URLs and links alongside the visible switcher. Google Search Central recommends separate URLs for language versions and advises against automatic language redirects that prevent access to other versions. Test direct article links as well as navigation from the homepage.
Where equivalent localized pages exist, implement the appropriate hreflang relationships. Google’s localized-page guidance requires each participating version to list itself and its alternates, with reciprocal links. Map actual corresponding pages; a language homepage is not a substitute for a missing article translation.
Keep these signals distinct: lang identifies content language, dir establishes direction, and hreflang describes alternate language or regional pages for search. Correct RTL presentation alone does not establish indexability or a ranking improvement. An access-controlled internal knowledge base also needs an access policy separate from public search requirements.
Verify accessibility with realistic tasks
Check the complete reading and interaction sequence with representative localized content. Headings should describe their sections, controls should have understandable names, and procedures should remain meaningful when read or navigated without relying on the visual arrangement alone.
WCAG’s Focus Order criterion requires sequential navigation to preserve meaning and operation when order matters. It does not prescribe one universal left-to-right or right-to-left tab sequence. Test whether the reader can predict the next control and complete the task.
Keep source order meaningful and avoid positive tabindex values as a patch for a confusing layout. Test the search overlay, navigation drawer, language menu, feedback controls, and support launcher. Confirm that opening and closing each component leaves focus somewhere useful, usually back at its initiating control.
For vertically scrolling content, WCAG Reflow guidance uses a width equivalent to 320 CSS pixels, subject to exceptions for content requiring a two-dimensional layout. A table exception does not justify making the whole article require horizontal scrolling. Review enlarged text, narrow layouts, and the contents of cells or code regions separately.
Keep focus indicators visible. WCAG 2.2’s Focus Not Obscured minimum criterion requires that author-created content not entirely hide a focused component. Keeping the entire control unobscured is a useful stronger design goal, particularly around sticky headers and support widgets.
Mark genuine language changes and test representative passages with the assistive technologies your audience uses. A visual screenshot cannot establish pronunciation, keyboard behavior, or accessibility conformance. Combine automated checks with manual task testing and review by people who read the supported language.
Evaluate the platform and publishing workflow
An RTL option in a settings menu is a starting point. Before committing to a template or accepting a customization, request evidence for the reader interface, editor, search overlay, forms, embedded tools, and any exports you plan to use. A capability demonstrated in one surface does not establish it in the others.
Use a test article containing a long title, an opposite-direction product phrase, a punctuation-heavy identifier, a table, and a code block. Save it, translate it through the normal workflow, preview it, and inspect the published page. Check whether language attributes, direction settings, and isolation boundaries survive each step.
For WordPress or another CMS, distinguish article markup from theme behavior. Inspect the published output rather than assuming that a correctly configured editor preview proves the theme, plugin, or embedded widget behaves correctly. Keep essential RTL rules in maintained templates and components where practical instead of repeating fragile fixes in every article.
Document the limits of any platform workaround. If a search widget cannot handle the required language configuration, or an editor removes necessary markup, identify the supported implementation path before scaling content production. Confirm that the chosen fix can be maintained by the team responsible for future updates.
Launch with an acceptance matrix and clear ownership
Use observable outcomes to decide whether the localized experience is ready. The following matrix is a proposed starting point, not a report of completed tests. Run it against the published implementation with actual translated content and the browser and device combinations your team supports.
| Reader task | Expected outcome | Evidence to record | First layer to investigate |
|---|---|---|---|
| Open a direct localized article URL. | The intended article, language, and product version appear. | URL, locale, version, and displayed title. | Routing and publication settings. |
| Search with a localized phrase and a mixed-language query. | The query remains readable and the intended article is discoverable. | Query, filters, returned article, and result language. | Input template, locale, indexing, and filters. |
| Read a product string beside an identifier. | The complete phrase and its punctuation remain understandable. | The actual localized sentence and its rendered state. | Direction boundaries and inline isolation. |
| Copy a URL, identifier, and command. | The pasted values match the intended source. | Source and pasted character sequences. | Source text, editor transformations, and copy controls. |
| Change languages across available and unavailable translations. | The article context or a clearly explained fallback remains usable. | Starting URL, selected language, destination, and publication state. | Article mapping and fallback policy. |
| Navigate by keyboard and at a narrow width. | Essential content and actions remain available in a meaningful sequence. | Viewport or zoom setting, focus sequence, and any obstruction. | Layout, source order, overlays, and responsive components. |
| Correct and submit a support request. | Errors are understandable, data is retained, and confirmation is usable. | Validation state, corrected values, and confirmation behavior. | Form configuration and integration. |
| Edit and republish the article. | Required content boundaries and tested behavior remain intact. | Relevant published markup and repeated task results. | Editor, translation pipeline, theme, and cache. |
Release a verified path before expanding coverage
Begin with the shared template and a representative set of difficult content. Fix defects that prevent the selected support task from being completed, make essential controls unavailable, corrupt copied values, or send readers to an unexplained language state. Then expand the same patterns to additional categories and article types.
Assign a localization reviewer to wording and native reading behavior, an engineering owner to templates and interactions, a search owner to retrieval, and a content owner to article mapping and freshness. A release owner should record unresolved defects and decide whether the intended task remains usable.
Monitor problems by locale and keep regression fixtures
After launch, review zero-result queries, repeated reformulations, article feedback, failed form interactions, and support escalations by locale where your tools provide that information. Compare like tasks and product versions. A lower click rate or higher contact rate is a signal to investigate, not proof that direction alone caused the difference.
Retain the query set, translated sentences, long filenames, and expected destinations as regression fixtures. Rerun the affected checks after changes to fonts, themes, navigation, editors, search settings, or embedded support tools. Include the published page in that review, not only the component preview.
The release decision should come from the reader’s complete task: reaching the right answer, understanding the procedure, preserving its technical details, and continuing to support when necessary. Keep that task as the reference point whenever an RTL design decision becomes a debate about appearance alone.



