~0.5%
Share of web data that is in Arabic, against more than 400 million native speakers
Source: Qatar Computing Research Institute, HBKU, stated as an engineering constraint in the Fanar 2.0 publication
+40.9%
Largest single citation lift measured in the peer-reviewed generative engine optimisation study, from named expert attribution
Source: Aggarwal et al., GEO: Generative Engine Optimization, ACM SIGKDD 2024
27B
Parameters in Fanar 2.0, Qatar's sovereign Arabic model, up from 9 billion in the first release
Source: Qatar Computing Research Institute at Hamad Bin Khalifa University, announced December 2025
38
Languages Google added to AI Mode in October 2025, including Modern Standard Arabic
Source: Google MENA product announcement, October 2025
About GEO & AEO in Qatar
Qatari buyers now ask an AI assistant to shortlist suppliers before anyone picks up a phone. Whether your brand appears in that answer depends on something different from where you rank, and in Arabic it depends on something different again.
A Thin Arabic Source Pool Cuts Both Ways
Qatar Computing Research Institute, the team that built Qatar's own Arabic model, states the problem plainly in its own technical publication: Arabic accounts for roughly half a per cent of web data, against more than 400 million native speakers. Every retrieval system working in Arabic draws from that pool. The first consequence is the obvious one, that Arabic answers have less to work with. The second consequence is the one worth acting on. A field with very little high-quality Arabic material is a field with very little competition for the position of being the source that gets cited. That is not a promise of results. It is a description of why the same work costs more effort and returns more room in Arabic than it does in English.
Qatar Built Its Own Arabic Model, Which Adds a Surface
Most markets have one set of answer engines to think about. Qatar has that set plus a sovereign one. The national AI strategy, published in 2019 under what is now the Ministry of Communications and Information Technology and developed with Qatar Computing Research Institute, names Arabic language processing as an explicit national priority, and the country followed through by building its own model family: Fanar, released at 9 billion parameters in December 2024 and at 27 billion parameters in December 2025, trained inside Qatar on Qatari infrastructure, with development continuing. It runs as a closed environment specifically so that data does not leave, it handles Gulf dialect alongside Modern Standard Arabic, and it includes components built for Islamic content. For a Qatari organisation, especially one that is state-linked or operating in Islamic finance, that is a retrieval surface which simply does not exist in other markets and which no global agency playbook accounts for.
The Same Question, Asked in Two Languages
One buyer, one intent, two languages. The engine does not simply translate its answer. It reaches for a different set of sources, and a brand can be present in one version and absent from the other.
Who are the leading facility management providers for commercial buildings in Doha?
Sources skew toward English-language material:
English corporate sites and English service pages
English-language regional business coverage
English site content
Regional English media
English directories
The same question, in Modern Standard Arabic, using the phrasing a Qatari tender document would use
A narrower pool, and a different set of names surfaces:
Organisations with genuine Arabic-language pages
Brands named in Arabic-language coverage
Arabic site content
Arabic media
English fallback where Arabic is thin
4 paths, not 1 number
An Arabic question answered from an Arabic source, an Arabic question answered from an English source, an English question answered from an English source, and an English question answered from an Arabic source are four separate results. Reporting them as a single Arabic visibility percentage hides where the loss actually happens.
Illustrative of the language-localisation pattern documented in the peer-reviewed and vendor research cross-checked for this page. Not a captured screenshot of any specific query, and the vendor names shown are categories rather than real organisations.
What The Research Actually Measured, and What It Did Not
Two findings shape how we work in Qatar. One is about where retrieval loses accuracy. The other is about which content choices measurably raise the chance of being cited. Both are peer-reviewed, and both are narrower than the marketing claims circulating in this market.
Measured loss when query language and document language do not match
Multilingual-E5Travel set
Amiraz et al., The Cross-Lingual Cost: Retrieval Biases in RAG over Arabic-English Corpora, ACL ArabicNLP 2025. BGE-M3 figures are retrieval accuracy loss, Multilingual-E5 figures are Hit@20 loss. Two different measures, shown separately rather than averaged. The loss follows the language mismatch, not Arabic itself.
+40.9%
The Largest Measured Citation Lift Came From Named Expert Attribution
The peer-reviewed generative engine optimisation study presented at ACM SIGKDD in 2024 tested content changes against citation outcomes. Quoting a named expert produced the biggest single lift. Statistics paired with a named source and inline citations both produced substantial lifts. Keyword stuffing reduced citation rates. That is the evidence base this page is built on, and it points at editorial credibility rather than at technical tricks.
Not measured
No Independent Arabic Citation Benchmark Exists For Qatar
Across every source reviewed for this page, no independent study measures how often answer engines cite Arabic sources for Arabic questions in Qatar specifically, and none measures Arabic against English answer quality inside the Qatari market. That absence does not stop the work. It changes what an honest report looks like, and it means any Arabic visibility figure you are shown, including one from us, has to arrive with its query set, its dates and its calculation method attached.
The retrieval figures come from general-domain benchmarks, not from measurements taken inside Qatar. We cite them for the mechanism they demonstrate and do not restate them as Qatari market statistics.
Our GEO & AEO Services in Qatar
Six workstreams aimed at one outcome: being the source an answer engine names when a Qatari buyer asks in the language they actually ask in.
Bilingual Prompt Panel Design
A fixed, written panel of buyer questions in three variants: Modern Standard Arabic in the register your category actually uses, Gulf dialect where buyers speak conversationally, and English for the large non-Arabic-speaking share of the market.
The panel is agreed with you before any baseline is taken, and it does not change quietly between reports. A visibility number is only comparable over time if the questions behind it stayed the same, and we would rather argue about the panel at the start than explain a moved goalpost later.
Four-Path Citation Testing
Every prompt is tested along four paths: Arabic question with Arabic sources available, Arabic question where only English sources exist, English question with English sources, and English question where the answer lives in Arabic.
Peer-reviewed work at ACL ArabicNLP in 2025 measured losses from thirteen to forty-two per cent when the query language and the document language disagreed, depending on the embedding model and the domain. Collapsing those four paths into one percentage hides the single most actionable finding about a bilingual site.
Arabic Entity and Schema Infrastructure
One entity, two languages, one graph. Stable identifiers, an Arabic alternate name that is actually the name your market uses rather than a transliteration, inLanguage declared with proper language tags, and sameAs links to the public knowledge bases that answer engines lean on for verification.
Stated honestly: schema is language-neutral and helps a machine understand what it is reading, and Google has said publicly that structured data does not guarantee any particular result. We implement it as hygiene, not as a lever we pretend to control.
Citable Content Units in Arabic and English
Sections built so that a single extracted passage still makes sense on its own: the question in the heading, the answer in the opening sentences, one claim per unit, a named source attached to every figure.
Named expert attribution produced the largest measured citation lift in the peer-reviewed study, so subject-matter quotes are treated as structural rather than decorative. Tables are real HTML tables containing real text, because Arabic content locked inside an image or a scanned document is not reliably readable by anything.
Earned Coverage in Arabic-Language Outlets
Placement and mentions in Arabic-language media and directories, because a brand that appears only on its own site gives an answer engine one source to weigh and no corroboration.
This is the slowest and most manual workstream, and the one where a thin Arabic source pool works in your favour: there is far less high-quality Arabic material competing for the same citation slot than there would be in English.
PDPPL-Aware Monitoring and Reporting
Citation tracking runs on prompts we write and brand mentions we observe, which is non-personal data. That keeps the monitoring layer clear of the obligations that attach to processing personal data under Law No. 13 of 2016.
Where reporting does pull in personal data, from lead forms or self-reported attribution, a processor agreement and defined purposes come first. Cross-border processing is permitted in principle under the law's cross-border data flow provisions, so an overseas team is workable, and clients registered in the Qatar Financial Centre fall under that centre's separate regime instead.
ranking first and being cited are two different outcomes, in two different languages each
Qatar's Sovereign Arabic Stack, and Where It Fits In Your Plan
Qatar is one of very few markets where the national government commissioned its own Arabic model rather than waiting for global ones to improve. That changes the surface an agency has to think about, and it changes it differently depending on who your buyer is.
| Surface | What it is | Who it matters most to |
|---|
| Global answer engines |
ChatGPT, Gemini, Google AI Overviews and AI Mode, Perplexity, Copilot, Claude. Google brought AI Overviews to the region and to Arabic in May 2025, then added Modern Standard Arabic to AI Mode in October 2025 as part of a 38-language expansion. |
Every commercial buyer. This is where the majority of discovery happens and where a bilingual prompt panel does most of its work. Note that the Arabic support announced for AI Mode was Modern Standard Arabic, not Gulf dialect. |
| Fanar, the sovereign model family |
Built by Qatar Computing Research Institute at Hamad Bin Khalifa University with the Ministry of Communications and Information Technology. Released at 9 billion parameters in December 2024 and 27 billion in December 2025, trained inside Qatar on Qatari infrastructure, with development continuing. Handles Modern Standard Arabic alongside Gulf, Levantine and Egyptian dialects, and is available through an interface and an API. |
State-linked organisations, regulated sectors, and anyone whose buyers or internal teams use a national tool precisely because it runs as a closed environment and data does not leave. Also relevant to Islamic finance and halal commerce, where the platform includes components built specifically for Islamic content. |
| The national policy layer |
Qatar's national AI strategy, published in 2019 and developed with Qatar Computing Research Institute, names Arabic language processing as an explicit national priority. The Qatar Central Bank has issued AI guidance for licensed financial institutions. |
Anyone preparing a tender response or a board paper. Alignment with stated national priorities is a procurement signal in Qatar in a way it is not in most private-sector-led markets. |
Model versions move. We track the sovereign stack as a developing family rather than a fixed product, and we would treat any agency that quotes you a citation rate for a specific model version without a date and a method as guessing.
What We Report, and What We Refuse To Report
This market is full of visibility percentages with no method behind them. The fastest way to tell a measurement apart from a sales number is to look at what the vendor is willing to define.
What we report
Cited-answer rate. The share of panel prompts where an answer engine names your brand and attributes a source, reported separately per engine and per language path.
Source-language split. For every citation earned on an Arabic prompt, whether the cited document was Arabic or English. This is the number that tells you where to publish next.
Entity accuracy. Whether the engine describes your business correctly, in both languages, including name, services and location. A confidently wrong description is worse than an absence.
Framing. Whether you are named neutrally, positively, or with a hedge attached, since a cautious mention is a different commercial outcome from a recommendation.
Reproducibility. Prompt, date, engine, language and the raw answer text, exported so your own team can re-run it without us.
What we refuse to report
A single Arabic visibility percentage. It averages four different language paths into one figure and destroys the only diagnostic that matters.
Mentions counted as citations. An engine can name a brand without attributing a source. Only one of those two things sends anything back to your site, and merging them inflates a report.
Ranking data offered as citation evidence. A page can rank first and never be cited, and a page from deeper in the results can be cited repeatedly. Using one to evidence the other skips a step.
A guaranteed citation. Nobody controls what a model outputs. Any guarantee here is a claim about something the agency does not own.
Only the prompts that went well. The failed prompts are the working document. A report that omits them is a brochure.
Why Choose Us as Your GEO Agency for Qatar?
Documented GEO Practice Since 2023, and a Method We Will Hand You
Most agencies now selling GEO in the Gulf added it to an existing search service. In a category with no independent Arabic benchmark to appeal to, the only honest differentiator is a method a client can inspect, re-run and disagree with.
3
Offices: Jakarta, Bandung, Bali
We Separate the Four Language Paths
Arabic question with Arabic sources, Arabic question with only English sources, and the two English equivalents. Reported apart, because peer-reviewed work shows the loss sits in the mismatch and a single Arabic percentage would hide exactly that.
Anchored to Peer Review, Not Vendor Multiples
Our method rests on the generative engine optimisation study presented at ACM SIGKDD in 2024 and on the ACL ArabicNLP 2025 cross-lingual work. Several widely quoted vendor statistics about this market were checked during research and deliberately left off this page.
Monitoring Built To Stay Off Personal Data
Citation tracking runs on prompts we author and mentions we observe, which keeps the monitoring layer clear of Law No. 13 of 2016 obligations. Where personal data does enter reporting, the processor agreement comes before the dashboard.
We Account For the Sovereign Stack
Qatar commissioned its own Arabic model family and named Arabic language processing a national priority in 2019. For state-linked and regulated buyers that is a real retrieval surface, and a global agency playbook does not mention it at all.
Explore Related Services
GEO in Qatar works hardest alongside the rest of the SEOv2 stack, because the technical and content foundation is shared even when the outcomes are not.
Want To See Your Brand Tested in Arabic and English, Side by Side?
We will build a small bilingual prompt panel for your category, run it across the engines your buyers use, and hand you the raw answers with prompts, dates and languages attached, including the prompts where you did not appear. Contact our team to get started.
Request Your Bilingual Citation Audit
Frequently Asked Questions About GEO & AEO in Qatar
What is the difference between GEO, AEO and SEO, and do we need all three?
SEO is about ranking positions in a list of links. AEO is about winning the single extracted answer. GEO is about being named and attributed inside a synthesised answer. They share a technical and content foundation, which is why they are usually scoped together, but they are not the same outcome. The sharpest way to see the difference: a page can rank first and never be cited, and a page from deeper in the results can be cited repeatedly. That is why rank position cannot be used as evidence of citation performance.
Is a GEO agency just an SEO agency that added AI to its service list?
In this market, often yes, and there is a reliable test for it. Look at whether the offer promises featured snippets and position zero alongside AI citation claims. Those are ranking constructs, and their presence signals a repositioned SEO service rather than a citation practice. Then ask for a prompt set, per-engine results, dates, and the calculation method behind any percentage. A vendor doing citation work can produce all of those on request. A vendor doing SEO with a new label usually cannot.
Do AI engines cite Arabic sources when the question is asked in Arabic?
Sometimes, and the honest answer is more useful than a yes or a no. Research indicates that citation pools shift toward the language of the query, which means English-only content is largely ineligible when a buyer asks in Arabic. Separately, peer-reviewed work published at ACL ArabicNLP in 2025 measured losses of thirteen to forty-two per cent when the language of the question and the language of the supporting document did not match, depending on the embedding model and the domain. The mismatch is what causes the loss, not Arabic itself. In practice that means the answer needs to exist in the language the question will be asked in, which is why we test four language paths rather than reporting one Arabic figure.
Can you prove citation results in Arabic, and what should we ask to see?
Ask for four things and accept nothing less. The full prompt panel, in Arabic and in English, including how many prompts and who chose them. The engines tested and the date range, with more than one run, because answer engines vary between runs. The raw answer text exported so your team can re-run it independently. And the prompts where your brand did not appear. A vendor who will show you the failures is measuring something. A vendor who only shows wins is selling something. We hand over all four as standard, and we would encourage you to demand the same from anyone else you shortlist.
Which engines matter most for a Qatari audience?
The global set does most of the work: ChatGPT, Gemini and Google AI Overviews, plus Perplexity, Copilot and Claude. Google brought AI Overviews to the region and to Arabic in May 2025, then added Modern Standard Arabic to AI Mode in October 2025 as part of a 38-language expansion. Alongside that, Qatar has a sovereign surface. Fanar is worth including where your buyers are state-linked, in regulated sectors, or in Islamic finance and halal commerce, because the platform runs as a closed environment and includes components built for Islamic content. Engine mix should be set per client rather than assumed.
What is Fanar, and does it matter commercially?
Fanar is Qatar's sovereign Arabic model family, built by Qatar Computing Research Institute at Hamad Bin Khalifa University with the Ministry of Communications and Information Technology. It was released at 9 billion parameters in December 2024 and at 27 billion in December 2025, trained inside Qatar on Qatari infrastructure, and development is continuing. It supports Modern Standard Arabic alongside Gulf, Levantine and Egyptian dialects and is reachable through both an interface and an API. Whether it matters commercially depends entirely on who your buyer is. For a state-linked organisation or a regulated institution choosing a national tool because data stays in-country, it matters a great deal. For a purely private consumer brand, it is a secondary surface.
Should Arabic content be in Modern Standard Arabic or Gulf dialect?
Both, assigned by intent. Modern Standard Arabic carries institutional, legal, procurement and business-to-business language, and it is the register Qatari organisations write in. Gulf dialect belongs where buyers speak conversationally: frequently asked questions, consumer pages, local intent. Two things are worth knowing on the answer-engine side. The Arabic support Google announced for AI Mode was Modern Standard Arabic specifically. And the main published Arabic dialect benchmark does not include Qatari dialect at all, so dialect handling is genuinely under-measured rather than solved. We treat dialect as something to test in your category rather than something to assume.
How do you handle our data under Qatar's privacy law?
The design goal is to keep the monitoring layer off personal data entirely. Citation tracking uses prompts we author and brand mentions we observe, which is non-personal. Where reporting genuinely needs personal data, from lead forms or self-reported attribution, Law No. 13 of 2016 applies and a processor agreement with defined purposes comes first. The law is supervised by the National Cyber Security Agency, through its National Cyber Governance and Assurance Affairs division, with the National Data Privacy Office as the enforcing body, and enforcement decisions have already been issued against operators. Clients registered in the Qatar Financial Centre fall under that centre's separate regime.
Does our data have to stay inside Qatar for you to work on it?
No. The law expressly recognises cross-border data flow, and as a general principle a controller is not to restrict transfer of personal data outside the state, with intervention reserved for cases where processing would breach the law or risk serious harm to the data or the individual's privacy. There is no blanket data-localisation mandate covering this kind of work, so an overseas delivery team is workable. What remains is accountability rather than prohibition: a processor agreement, purposes defined in writing, and special categories of data handled with the extra permissions they require.
Can a foreign agency do this work for a Qatari organisation?
For private-sector clients, yes, and it is common. For government and state-linked contracts the position is different. Public tenders run under Law No. 24 of 2015 and its amendments, bids are normally submitted in Arabic unless the tender document says otherwise, and local establishment requirements can apply before a contract is signed rather than at the bidding stage. No rule singles out marketing or search services, so the tender document controls rather than a general assumption. If public-sector work is the target, that should be resolved at the start of an engagement.
What KPIs will you report, and how often?
Cited-answer rate per engine and per language path, the source-language split for citations earned on Arabic prompts, entity accuracy in both languages, how your brand is framed when it is named, and full reproducibility data. Monthly, bilingually, against the same prompt panel each time so the numbers are comparable. What we will not report is a single blended Arabic visibility percentage, mentions counted as citations, ranking data presented as citation evidence, or a selective view that omits the prompts where nothing appeared.
Can you guarantee we get cited in ChatGPT or Google AI Overviews?
No, and treat any agency that says otherwise with suspicion. Nobody outside those companies controls what a model outputs, and outputs vary between runs of the same prompt. What can be committed to is the work and the evidence: the prompt panel, the baseline, the content and entity changes, the re-testing schedule, and a report that shows movement in both directions. The peer-reviewed study at ACM SIGKDD in 2024 found that named expert attribution, statistics with named sources and inline citations all measurably raised citation rates. That is a probability we can work on, not an outcome anyone can promise.
How long does citation building take in a bilingual market?
Longer than a single-language market, for a specific reason rather than a vague one. Two content and entity foundations have to be built rather than one, technical remediation on the Arabic side often has to land before content work produces anything measurable, and earned Arabic-language coverage is slow and manual by nature. The offsetting factor is real: with Arabic accounting for roughly half a per cent of web data according to Qatar Computing Research Institute, there is far less high-quality Arabic material competing for the same citation slot. We set a baseline first and report movement against it rather than quoting a timeline we cannot control.
How does GEO interact with the SEO work we already have?
They share a foundation and diverge on outcomes. Crawlability, site structure, page speed, clean markup and genuine content quality serve both. What GEO adds is a different unit of work: passages that stand alone when extracted, one claim per unit, a named source on every figure, question-shaped headings, entity consistency across both languages, and corroboration from sources you do not own. If
our SEO service for Qatar is already running, GEO layers onto it rather than replacing it, and the prompt panel becomes the shared measurement instrument for both.
Question asked in English
Same question asked in Arabic
