The eight AI engines we measure in Arabic

Because your customers in Saudi Arabia, the UAE, Egypt and across the Arab world reach them in Arabic, and because we hold text evidence under every verdict: the answer's excerpt, word for word, and the sources it cited. Both conditions hold or we do not measure: no evidence, no number. That is the work of what GEO is.

The two kinds differ in how your customer reaches them. The three search surfaces, Google AI Overviews, Google AI Mode and Copilot, reach them mid-search, inside the results page itself, and that page is usually the same for everyone asking the same question in the same country.

The five chat engines, ChatGPT, Gemini, Perplexity, Claude and Grok, are opened and asked by the customer, and their answers change from one ask to the next, so one lucky answer is never counted as a change. What a measurement gives you is in how we measure your score.

Which engines run on your account follows your plan, and each one contains the one before it:

• Free trial, for 7 days: ChatGPT and Gemini.
• Starter: adds Perplexity and Google AI Overviews.
• Pro and Advanced: add Google AI Mode.
• Business: adds Copilot, Claude and Grok, up to all eight, chosen with you.

They search before they write, then pick from the pages they found whichever answers the question directly. There is no fixed list of the best sites they cite: the engine chooses again on every question, and a page its crawler can reach, answering in its first sentence with the name, the city and a figure, is the closest to being cited.

Each engine has an index it searches and a crawler that reads for it. Google AI Overviews and Google AI Mode answer from Google's own index, Copilot from Bing search, and ChatGPT, Perplexity and Claude each have named crawlers. Their names, and how to allow them in robots.txt, are in can AI engines read your site, and how do you check.

In our measurements in October 2026, on Arabic questions about our own field, the engines kept returning to the same kinds of sites: the documentation of the company that owns the engine, news sites, government bodies, large platforms and the pages of companies that sell the same service. The cited page was rarely an Arabic page written for the Gulf market, which is an opening for whoever writes one. The steps are in the steps to lead AI answers in Arabic.

In October 2026 we measured Arabic buying questions, 1,192 of them in 12 sectors across Saudi Arabia, the UAE, Kuwait and Egypt. In 38 of 48 cases, each a sector in a country, the site Perplexity and Claude cited most was not a business's own site but, mostly, a platform listing many businesses, such as a directory, a review site or a booking portal. The full figures and their data are in who leads Arabic buying answers in AI? October 2026.

Google AI Overviews is a summary Google writes at the top of the results page, above the links, with its sources beside it. Google's Arabic name for it is «النبذات باستخدام الذكاء الاصطناعي», and many search for it as «ملخصات الذكاء الاصطناعي» (AI summaries). Your customer never chooses this surface: they type a question and the summary is already there.

In Arabic, Google does not write one for every question. When it writes none, we record the cell as no answer, and it never counts against your site, because a summary that was never written said nothing about you.

We measure it in Arabic and for your country, so what you read is what your customer sees there rather than a generic version. It unlocks on Starter, at $49 a month, alongside ChatGPT, Gemini and Perplexity, and stays on every plan above it. The practical steps are in the guide to Google AI Overviews in Saudi Arabia.

ChatGPT is OpenAI's chat engine: one continuous answer, with the links it cited under it. What we measure in it is whether your site is cited inside the answer, not where it ranks. ChatGPT reads the web for its search through a crawler called OAI-SearchBot, and OpenAI says sites that block it are not included in ChatGPT search answers.

Its answer moves. The same question morning and evening can name two different shops, so one lucky answer is never counted as a change, and a measurement that spans a range shows it rather than one number pretending to be stable.

We ask it in your customer's words, the way they ask in their market. It is in the 7-day free trial with Gemini, so it is one of the first two engines a new account sees. The detail is in how ChatGPT picks which businesses to recommend.

Gemini is Google's chat app, separate from the results page: you ask, it answers, and it cites the links it used. It is often confused with AI Overviews and AI Mode, but the three are different surfaces giving three answers to one question, so we measure each by name.

Your customer reaches it through the same Google account they use for mail and on their phone, with nothing to install. It draws on Google's search index for its answers, and Google says the Google-Extended token in robots.txt controls whether your pages are used to ground Gemini apps' answers, without touching your site's inclusion in Google Search.

What Gemini reads before it answers, and what to change on your page so it cites it, is in how does Gemini choose sources in Arabic and cite you?

We ask it in Arabic, the way your customer asks, and because its answer moves from one ask to the next, one lucky answer is never counted as a change. It is in the 7-day free trial with ChatGPT, and on every plan from Starter up.

Perplexity is an AI search engine that shows its sources inside the answer itself, numbered beside the sentence each supports. That makes it the clearest of the eight to measure: your link is among the sources or it is not.

People go to it when they want the source before they act. It reads the web for its search through a crawler called PerplexityBot, and we ask it in Arabic, the way your customer asks. It unlocks on Starter, at $49 a month, alongside ChatGPT, Gemini and Google AI Overviews, and stays on every plan above it. What it cites for Arabic questions, and how to let its crawler read your site, is in how Perplexity chooses its sources in Arabic, and how it cites your site.

Google AI Mode is the conversational surface inside Google Search, launched in Arabic on 8 October 2025 as «وضع AI»: a full generated answer with references of its own, not the summary above the results, so we measure it apart from AI Overviews. How it chooses its sources for Arabic questions is in what Google AI Mode is, and how it chooses its sources.

In our measurements on Arabic Gulf questions in September 2026, AI Mode sometimes writes a full answer citing no site at all. Nobody leads that cell: your site is absent and so are your competitors, which is a finding rather than a silence.

We measure it in Arabic and for your country, as your customer reads it inside the search page. It unlocks on Pro, at $129 a month, beside Google AI Overviews, which every paid plan measures.

Copilot is Microsoft's AI assistant, and it reaches your customer while they search in Bing, inside the results page itself. Bing Webmaster Tools' AI Performance report shows a site owner how often their pages were cited in Copilot and in Bing's AI summaries.

How Copilot picks its sources from Bing's index, and how to get your page into that index sooner, is in how does Copilot pick sources in Bing, and cite you?

We ask it in Arabic, as your customer meets it in Bing search. The Business plan unlocks it.

Claude is Anthropic's chat engine: long, ordered answers, citing the links it built on. People reach it through its app and the work tools that embed it, not from a search page, so it is often asked about work. It reads the web for its search through a crawler called Claude-SearchBot.

When the question is a decision question, a citation there reaches a reader preparing a proposal or choosing a supplier. We ask it in Arabic, the way your customer asks, and because its answer moves from one ask to the next, one lucky answer is never counted as a change. It unlocks on Business.

Grok is xAI's chat engine, asked from inside X itself: the platform where people talk about restaurants, clinics and shops before buying. Someone asking there may settle for Grok's answer, without opening another app to ask again.

It joined our list on 22 September 2026, once we had text evidence for every verdict on it: the answer's excerpt, word for word, and the sources it cited. We ask it in Arabic, the way your customer asks, and never count one answer as a change. The Business plan unlocks it, with Copilot and Claude.

Six guides, each explaining how one engine chooses its sources when asked in Arabic, and what to change on your page so it cites you:

• How ChatGPT reads your page and picks the businesses it names: how ChatGPT picks which businesses to recommend.
• When Google writes its summary above the results, and where it takes its references from in Saudi Arabia: how to get cited in Google AI Overviews in Saudi Arabia.
• How AI Mode splits your customer's question into sub-questions, and what it cites: what Google AI Mode is, and how it chooses its sources.
• Why Perplexity numbers its sources, and which Arabic sites it picks: how Perplexity chooses its sources in Arabic, and how it cites your site.
• What Gemini reads before it answers in Arabic, and what the Google-Extended token in robots.txt decides: how does Gemini choose sources in Arabic and cite you?
• How Copilot picks its sources from Bing's index, and what the AI Performance report tells you: how does Copilot pick sources in Bing, and cite you?

We add an engine when we can hold text evidence under every verdict: its answer's excerpt, word for word, and the sources it cited. Until then, any number we sold about it would be a guess, and we do not sell guesses.

The eight above are what we measure as of October 2026. If another engine matters in your market, contact the Tasadr team: what Arabic markets ask for is what we add first.

• Meta AI: we cannot reach its answer text and sources in a way we trust, so there is no evidence to put under a verdict.
• Local and open models on the user's own device: they do not read the web yet, so they cite nothing and there is nothing to measure.
• DeepSeek: a candidate for the day its answers and sources can be measured accurately. Not in yet.