Register
Fix the target register: Saudi-flavoured modern standard Arabic for enterprise channels, with a stated exception policy.
Done when: A sample of outputs is judged in-register by a native reviewer.
Arabic Quality Standard (AQS)
Arabic Quality Standard
A scored rubric for Arabic AI: register, terminology, entities, retrieval and rendering, evaluated, not assumed.
The business problem
Your AI speaks Arabic. Your customers can tell it was translated.
Models trained mostly on English get terms, names and tone wrong, and nobody measures it before launch.
What you hear inside the organisation
The chatbot answers in Arabic nobody speaks.
It mixes up customer names.
Search fails on Arabic spelling variations.
The vendor says Arabic is supported.
Who it serves and what changes
Your teams, before and after the Arabic Quality Standard
Client teams and what they need
Customer experience
Arabic that sounds natural to Saudi users
AI product teams
A test they can run before release
Brand and communications
Consistent terms and tone
Procurement
A fair way to compare vendors
The outcome you get
- Arabic quality scored on your own content
- Right terms, names and register
- Search that handles Arabic variations
- Vendor claims tested, not assumed
A worked example
A telecom operator launching an Arabic customer assistant
The problem we saw
The assistant passed vendor demos, but customers found its Arabic stiff, it misread names, and search missed common spellings.
How the framework solved it
We tested register, terminology, names, retrieval and display on the operator's own content, and compared vendors on the same test.
The business result
The operator picked the model that served Saudi customers best, fixed the gaps before launch, and now retests with every update.
Illustrative example based on common client situations.
In short
What is the Arabic Quality Standard?
Arabic Quality Standard (AQS): A scored rubric for Arabic AI: register, terminology, entities, retrieval and rendering, evaluated, not assumed.
Arabic is usually treated as a translation task at the end of a project, which is why Arabic output reads as translated.
- How it moves
- Each step closes with a check before the next begins
- Owned by
- Run by the Arabic language reviewer with the evaluation lead; signed off by the business owner of the channel.
- You keep
- Arabic evaluation set, Bilingual glossary, Rubric scorecard
How it works
How the Arabic Quality Standard works
The standard is symmetric: an Arabic release blocked for quality is treated exactly like an English one. Parity is the point.
Terminology
Fix one Arabic term per domain concept, with the transliteration policy for product and technical names.
Done when: Terminology is enforced in generation, not corrected afterwards.
Entities & formats
Handle Saudi names, addresses, national identifiers, Hijri and Gregorian dates, currency and numerals correctly.
Done when: Format handling passes a fixed test set drawn from real record shapes.
Retrieval & grounding
Evaluate Arabic retrieval separately: diacritics, morphology, synonymy and mixed-script documents.
Done when: Arabic retrieval quality is measured on its own set, not inferred from English.
Rendering & release
Check right-to-left layout, mixed-direction text, numerals, truncation and screen-reader output.
Done when: Arabic scores at or above the English threshold before release.
Is it right for you?
Is the Arabic Quality Standard right for you?
A strong fit if
- You serve customers or staff in Arabic
- You are choosing between AI vendors
- Users complain about Arabic answers
Probably not yet if
- Your AI use is English only
- You need human translation, not AI evaluation
Free assessment
Discuss the Arabic Quality Standard with our team
A working session with our senior team. We review what fits your situation, what does not, and where to start.
A focused conversation
We listen to your situation and goals.
Fit review
Which parts of this framework apply to you, and which do not.
A written view
Clear next steps, whether or not you work with us.
Go deeper
Applying the Arabic Quality Standard in practice
Open only the part you need: how it starts, what your team keeps, maturity levels, sector rules and the evidence a reviewer will ask for.
Arabic is usually treated as a translation task at the end of a project, which is why Arabic output reads as translated. AQS makes Arabic quality a release condition with a measurable rubric applied to the same evaluation set as English.
Version
v1.0
Last reviewed
August 2026
Family
Saudi Data & AI Standards
How this was developed: Drafted in Arabic first, then mirrored to English. The rubric was built from review findings on Arabic enterprise channels and reconciled with the evaluation obligations in the standards below.
Arabic evaluation set
A golden set built from real Arabic records and questions.
Bilingual glossary
One approved term per concept, enforced during generation.
Rubric scorecard
Register, terminology, entities, retrieval and rendering scored per release.
Banking & finance
Product and contractual terminology must match the approved Arabic of the customer agreement.
- Who reviews it
- SAMA, plus PDPL and NCA controls
- Systems in scope
- Core banking, cards, KYC and AML systems, credit bureau feeds, the general ledger
- Where it lands first
- KYC and AML review, credit decisioning support, complaint and dispute handling
1
Register
- Register guide
- Do-not-use list
2
Terminology
- Bilingual glossary
- Transliteration policy
3
Entities & formats
- Entity test set
- Format rules
4
Retrieval & grounding
- Arabic evaluation set
- Retrieval score sheet
5
Rendering & release
- Rubric scorecard
- Release sign-off
The standard is symmetric: an Arabic release blocked for quality is treated exactly like an English one. Parity is the point.
Stage
Exit gate
Artefacts
1
Register
Fix the target register: Saudi-flavoured modern standard Arabic for enterprise channels, with a stated exception policy.
Exit gate
A sample of outputs is judged in-register by a native reviewer.
Artefacts
- Register guide
- Do-not-use list
2
Terminology
Fix one Arabic term per domain concept, with the transliteration policy for product and technical names.
Exit gate
Terminology is enforced in generation, not corrected afterwards.
Artefacts
- Bilingual glossary
- Transliteration policy
3
Entities & formats
Handle Saudi names, addresses, national identifiers, Hijri and Gregorian dates, currency and numerals correctly.
Exit gate
Format handling passes a fixed test set drawn from real record shapes.
Artefacts
- Entity test set
- Format rules
4
Retrieval & grounding
Evaluate Arabic retrieval separately: diacritics, morphology, synonymy and mixed-script documents.
Exit gate
Arabic retrieval quality is measured on its own set, not inferred from English.
Artefacts
- Arabic evaluation set
- Retrieval score sheet
5
Rendering & release
Check right-to-left layout, mixed-direction text, numerals, truncation and screen-reader output.
Exit gate
Arabic scores at or above the English threshold before release.
Artefacts
- Rubric scorecard
- Release sign-off
Standard
What this framework satisfies
ISO/IEC 42001
Performance evaluation obligations applied per language, not per system.
NIST AI RMF
Measure function, validity and fairness measured on the Arabic population too.
SDAIA
Ethics principles on transparency and inclusiveness evidenced in the language of the user.
EU AI Act
Transparency and instructions-for-use obligations delivered in the user's language.
Translation at the end
Arabic added after launch reads as translated and is rarely re-evaluated. Build the Arabic set first.
Colloquial drift in enterprise channels
Dialect in a banking or government channel reads as unserious. Fix the register and test for it.
English-only evaluation
Scores from an English set say nothing about Arabic retrieval or morphology.
Arabic quality rubric
Score an Arabic AI channel before it ships. Tick only what has been evaluated, not what is intended.
0%Unevidenced0/10
Most statements have no artefact behind them. Start with one domain and produce the evidence before widening scope.
Scoring is deterministic and runs entirely in your browser. Nothing is stored or sent.
Built on
- 1
ISO/IEC 42001: AI management systems
International Organization for Standardization, management system requirements for AI.
- 2
NIST AI Risk Management Framework 1.0 and Generative AI Profile
National Institute of Standards and Technology, Govern, Map, Measure, Manage.
- 3
SDAIA AI Ethics Principles
Saudi Data & AI Authority, national principles for AI.
- 4
EU AI Act
Risk tiering, documentation and transparency obligations.
Citations describe the parts of each standard the framework answers. Bilytica is not an accreditation or certification body.
In Saudi Arabia
Written for Saudi Arabia, not adapted to it
This family exists because the Saudi control set, Arabic quality and the in-country deployment ruling are not add-ons to a global method. They decide whether a system is allowed to run at all.
AuthorityWhat it requiresWhat we buildWhat you keep
SDAIA, PDPL
National AI ethics principles applied as testable controls.
Each principle becomes a check with a pass condition and a named reviewer.
Ethics review record per release, in Arabic.
Arabic quality standard
Saudi Arabic output has to be correct in register, terminology and numerals, in front of a customer.
Arabic evaluation sets built from Saudi content, scored before release.
Arabic quality score with the failing cases attached.
NCA ECC
Deployment location and network boundaries are a documented decision.
A deployment ruling per data class states what may leave Saudi Arabia and what may not.
Signed residency ruling referenced by the architecture record.
Where it runs
The default is in-country, in your environment. Any exception is written down with the data class, the reason and an expiry date.
Arabic, as it is spoken here
Arabic is the governing version of every artefact in this family, with English supplied alongside.
FAQ
Frequently Asked Questions
Which Arabic do you standardise on?
Saudi-flavoured modern standard Arabic for enterprise channels, with dialect permitted only where the channel and the audience justify it.
How is Arabic retrieval measured?
On its own golden set, with morphology and diacritic variants included deliberately rather than filtered out.
Does the standard apply to written content too?
Yes. The same rubric governs interfaces, assistant output and published content so one voice is used everywhere.
Where this connects
Rest of the library
DFFData Foundation FrameworkAIRMAI Readiness ModelVALUEAI Value FrameworkTRUSTAI Assurance FrameworkDELIVERAI Delivery FrameworkAGENTAgent Operations FrameworkSOVEREIGNSovereign AI FrameworkDMMData Management Maturity ModelDPAData Platform Architecture FrameworkDQFData Quality & Observability FrameworkDECIDEAnalytics & Decision FrameworkDGOVData Governance Operating ModelSDCFSaudi Data Compliance FrameworkKSA-READYIn-country Deployment Framework
Topic cluster
Explore the topic cluster
Curated links across our Saudi hubs, sector pages and long-form blog research.