Who leads Arabic buying answers in AI? October 2026

We put 1,195 Arabic buying questions to Perplexity and Claude, in 12 sectors and four markets: 714 for Saudi Arabia, 205 for the UAE, 96 for Kuwait and 180 for Egypt, between 95 and 102 per sector. The Tasadr team wrote them the way a customer about to decide writes them, so each one is a buying question, the kind asked just before a purchase, not a general information question.

They do not all ask the same thing. 507 ask for a recommendation, such as "What is the best hotel facing the Haram in Makkah for families?"; 205 ask about price; 170 ask how to choose; 164 ask whom customers trust, such as "Which real-estate offices in Dammam do clients praise for getting things done quickly?"; and the rest compare two options.

Most name a city, as a customer would: Riyadh in 261 questions, Jeddah in 150, Dubai in 127, Kuwait in 95, Cairo in 80, Dammam in 56, Abu Dhabi in 42, Khobar in 39, Makkah in 31, Sharjah in 28, Madinah and Giza in 18 each, Alexandria in 16 and AlUla in 3. The other 231 ask about the country as a whole.

All the Gulf questions are written in Modern Standard Arabic. Half of the Egyptian ones, 90 of 180, are in Egyptian Arabic, such as "Who's the best dentist in the Fifth Settlement?"

Perplexity answered 1,155 of them, on 11 and 12 October 2026, and Claude 1,107, between 12 and 15 October 2026, so at least one engine answered 1,192 questions. The other three, all Saudi, did not complete their measurement and are left out: two about hotels and one about schools.

We split the data into 48 cases, each one a sector in a country. In 38 of them the most-cited site was not a business that provides the service itself. In 35 it was a platform that gathers many businesses, such as a directory of doctors, lawyers, schools or agencies, a review site, a booking portal, a property portal, an online marketplace or a price comparison site. The other three are all in Kuwait: a government portal led one, a local guide site another, and in the third a local directory tied with a newspaper and a blog.

In seven cases the most-cited site was a business's own: marketing agencies in the UAE; B2B software and services in Saudi Arabia, the UAE and Kuwait; car rental in the UAE and Egypt; and gyms in Saudi Arabia. In the remaining three, a business's own site tied with a platform: clinics and e-commerce in Kuwait, and hotels in the UAE.

Saudi Arabia alone shows the same pattern on larger numbers, with between 55 and 62 questions per sector. Platforms led 10 of its 12 sectors; a business's own site led only two, B2B software and services, and gyms.

In some sectors one site recurs in most questions across all four countries. The Arabic edition of one global review site was cited in 83 of 101 restaurant questions and in 81 of 96 hotel questions. One lawyer directory was cited in 61 of 101 questions on law, accounting and consulting, one online marketplace in 62 of 102 e-commerce questions, and one directory of agencies and B2B providers in 59 of 99 agency questions.

The table gives the type of the most-cited site in each case, with the number of that case's questions in which it was cited. We give the type, not the name, because what matters to your decision is where the engine reads in your sector, not one particular site. Where two or more sites tie, we list them all. Kuwait has 8 questions per sector and Egypt 15, so read their figures as counts, not as rates.

It varies a great deal by sector. Hotels are the most concentrated: the answers to 96 hotel questions spread over 457 different sites, and nine of them were cited in a quarter of the questions or more. Restaurants come next with 648 sites across 101 questions, then car rental with 702 across 95, then schools with 780 across 95.

At the other end is e-commerce, whose 102 questions spread their answers over 1,338 sites. B2B software and services follows with 1,180 sites across 102 questions, none of them cited in a quarter of its questions or more, then marketing agencies with 1,055 sites across 99.

The difference matters in practice. In a concentrated sector the engines go back to a few platforms again and again, so your place on them may be part of your place in the answer. A fragmented sector seems to leave room for many more sites. That is our reading of this measurement, not a rule any engine has announced.

Yes, in at least two ways. The first is the share of government sites: a government site was cited in 9.9% of the answers to Saudi questions, 20.5% of the UAE answers, 17.5% of the Kuwaiti answers and 11.9% of the Egyptian answers. These are lower bounds, because we count a site as governmental only when we know it to be one.

The second is the country's own domain, a site whose address ends in the country's code: sa for Saudi Arabia, ae for the UAE, kw for Kuwait and eg for Egypt. A site on the country's domain was cited in 77.1% of the Saudi answers, 84.4% of the UAE answers, 42.1% of the Kuwaiti answers and 56.8% of the Egyptian answers. Kuwait stands out: more than half of its answers cited no site on its own domain.

The leader in schools changes from country to country. In Saudi Arabia a school directory was cited in 42 of 55 questions; in the UAE the school guides of two property portals were cited in 15 and 13 of 17; in Kuwait a government portal was cited in all eight; and in Egypt two childcare and school directories tied at 8 of 15 each.

Restaurants and law did not change leader. The same review site led restaurants in all four countries, and the same lawyer directory led law, accounting and consulting in all four as well. If your business is in one of these two sectors, the platform the engine reads from may be the same wherever your customer is.

For hotels in Makkah and Madinah we measured 16 questions, such as "Which hotels in Madinah do visitors thank for their welcome?" Each of two booking platforms, and the global review site, was cited in 12 of them. What applies to hotels in general is in how AI answers recommend your hotel to guests.

Mostly, yes. Each of the two engines reads the web through a crawler with a published name, and the sites we checked are open to the search crawlers almost without exception. On 14 October 2026 we checked the public files of 150 sites that AI engines cite; Perplexity or Claude cited 71 of them in this note's answers, most of them businesses' own sites.

The check read what any visitor can read: robots.txt, the home page with and without JavaScript, the sitemap and llms.txt. Here is what we found in the 71.

• Home page: all 71 served it
• Search crawlers: in none of their robots.txt files did we find a rule blocking OAI-SearchBot, PerplexityBot or Claude-SearchBot; one site's file could not be read
• Page without JavaScript: two arrived empty, meaning their text is only built in the browser
• Sitemap: we found none on 5 sites
• Training crawlers: 5 sites blocked at least one training crawler: two blocked GPTBot, and two blocked Google-Extended, the name a site owner uses to decide whether the site's content may be used to train Gemini models and in Gemini's answers that draw on Search
• Cloudflare: 30 sites are served through it, and none had turned on its managed robots.txt, a setting with which Cloudflare adds rules to the file that block the known training crawlers, as the owner chooses
• llms.txt: we found one on 43 sites; when we opened these files on 17 October 2026, at least seven had been generated automatically by plugins rather than written by the sites' owners

These figures promise you no result, but they do show where the engine reads when your customer asks it. Four steps follow from them that we think are sensible for any business in these sectors.

• Know the platform that leads your sector in your country: a directory, a review site or a booking portal. Claim your business's listing on it and complete it, with name, city, services, prices and opening hours, because that listing may be the first thing the engine reads about you. The method is in how to claim and complete your listings on directories and platforms
• Make your page answer the question in text: someone who asks "How much is a honeymoon package in Jeddah's hotels?" is waiting for a price. What we see in our measurements is that a page that states the price and the city in its first sentence is likelier to be quoted than one that says "contact us". The method is in the steps to lead AI answers in Arabic
• Make sure AI crawlers can read your site: almost all the cited sites are open to them, and checking robots.txt and the page without JavaScript takes minutes. The steps are in the guide to AI crawlers
• Measure your customers' questions in your market: these figures describe a whole market, and what your customers ask may differ. What you get when you measure their questions is described in how we measure where you stand in AI answers

These data describe what we saw on specific days of measurement, and they have limits you should know before you build on them.

• One measurement: the answers came between 11 and 15 October 2026, and engines' answers can change from week to week, so read the figures as a picture of those days; why ChatGPT's answers change, and when a change is real explains that change
• Two engines only: Perplexity and Claude alone. ChatGPT, Gemini and Google's AI Overviews and AI Mode come in a later note, so do not carry these figures over to them
• Questions we wrote, not real conversations: we wrote the questions the way a customer writes them, but they are not a log of what people ask, and they say nothing about how often each question is asked
• Small counts in Kuwait and Egypt: each sector has 8 questions in Kuwait and 15 in Egypt, so we give their figures as counts and never turn them into a rate for a single sector
• Types sorted for the leaders only: we opened the most-cited site in each case and set its type ourselves; the other sites are not all sorted, which is why the government share is a lower bound and why we publish no share for businesses' own sites
• A citation is not an endorsement: an engine citing a site means it built its answer on that site, not that the site is better than others
• No causes here: the figures describe what happened. They do not say why an engine chose one source over another, nor that a change to your site will change the result

Here is the scope of the measurement as it stands, in figures and dates.

• Questions: 1,195 Arabic buying questions in 12 sectors, written by the Tasadr team the way a customer writes them
• Markets: Saudi Arabia 714 questions, the UAE 205, Kuwait 96 and Egypt 180
• Engines and dates: Perplexity on 11 and 12 October 2026, Claude between 12 and 15 October 2026; at least one of them answered 1,192 questions
• What cited means: a site is cited when it is among the links the answer gives as its sources
• What a site is: we count a site by its address, so the Arabic edition of a global site counts as a site of its own
• Crawler check: on 14 October 2026, on public files only: robots.txt and its rules for the crawlers each engine names; the home page with and without JavaScript; the sitemap; and llms.txt

This note rests on three tables, which we publish in full as CSV files under the Creative Commons CC BY 4.0 licence. You may copy them, republish them and build on them, in research, an article or a client presentation, commercial use included. They name no site: they hold types and counts, not a list of sites.

• The sector and country table: for each sector in each country, the number of questions and answers, the type of the most-cited site and its number of questions, the number of different sites, and the number of answers that cited a government site or a site on the country's domain
• The concentration table: for each sector, its questions across the four countries, the number of different sites, and the number of sites cited in a quarter of the questions or more
• The crawler check table: the number of the 71 sites in each item of the check of 14 October 2026