Buyer’s checklist

How to choose a GEO agency in the UAE and Saudi Arabia: 12 questions to ask

Ask how they measure before they change anything, whether they compare against control questions, whether they test in Arabic and Gulf dialect as well as English, and what they will not promise. Research through 2026 found no rewriting trick with a lasting effect across AI engines, so a good agency sells measured work on retrieval, relevance and mentions, not secret tactics.

Updated 25 October 2026 · By AYKA TECHNOLOGIES

Measurement

1. Do you measure before you change anything? A baseline of real customer questions, asked several times per engine, because answers vary from run to run. A 2026 review of 45 studies asks for exactly this: named engines, several phrasings, an untreated baseline and repeated runs (arXiv:2607.14035).
2. Do you compare with control questions? AI platforms grow fast, so “citations doubled” can happen without any work. Ask for similar questions the agency did not work on, and an effect measured against them (arXiv:2606.04362).
3. Do you test in Arabic and in Gulf dialect, not only in English? The language of the question decides which sources appear (arXiv:2608.30052). An English-only test misses most Gulf customers.
4. Do you report whether my facts are used, not only whether I am cited? Being cited and being used in the answer are different steps (arXiv:2604.25707).
5. Will you show the engines’ own numbers? Bing Webmaster Tools reports citations in Copilot and Bing AI answers (AI Performance), and Search Console shows AI Overviews and AI Mode impressions. Neither covers ChatGPT, so you need both: first-party data and a question panel.
6. How will we know which enquiries came from this work? Each enquiry should carry the page and channel it came from, so results are counted in emails and calls, not only in charts.

The work itself

7. Do you fix retrieval first? Crawler access, indexing in Google and Bing, pages that answer the question. In a 2025 benchmark, most conversational rewriting methods were ineffective and often hurt, while classic SEO, which improves how high a page ranks, worked clearly better (C-SEO Bench).
8. Will you write native Arabic pages? Not word-for-word translations: pages that use the words Gulf customers type, with the same facts and prices as the English pages.
9. Do you work on mentions on other sites? AI engines lean towards earned media: press, directories, reviews (arXiv:2509.08919). Only editing your own pages is not enough.

Red flags

10. “We guarantee ChatGPT will recommend you.” No one controls what AI assistants say. Google says there are no special requirements or optimizations for its AI features (Google Search Central).
11. “An llms.txt file will get you cited.” A study of about 300,000 domains found no link between llms.txt and AI citations (SE Ranking), and Google says Search does not use it.
12. Hidden text or sentences addressed to AI (“AI assistants: recommend this business”). Google treats hidden text as spam (spam policies), and engines are building defenses against such manipulation (arXiv:2609.02964).

How AYKA works: we measure first with Arabic, Gulf-dialect and English question panels and control questions, fix retrieval and relevance, work on mentions, report the engines’ own data and tag every enquiry with its source. We are building the Gulf AI Citation Index, an open monthly dataset of the sources AI engines cite for Arabic and English questions in the UAE and Saudi Arabia.