How does Gemini choose sources in Arabic and cite you?
The Tasadr team · ·13 min read
Gemini is Google's chat app, and Google says it may ground its answers in content from the Google Search index at the moment of the question. It can cite your site when your page is in that index, is not opted out through Google-Extended, and answers your customer's Arabic question in its first line, with name, city and figures.
On this page (10)
What is Gemini, and why measure it on its own?
Gemini is Google's chat app. Your customer opens it at gemini.google.com or on their phone, types a question and reads a written answer, sometimes with a Sources button at the end of it or inside it that opens the links it shows as the answer's sources. Those links are what we call a citation.
Gemini is not AI Overviews, and it is not AI Mode. They are three Google products that may write three different answers to the same question; how they differ is in is Gemini the same as AI Overviews? So do not judge where you stand in Gemini by what you read on Google's results page.
Many people who search for Gemini in Arabic ask which is better, Gemini or ChatGPT. That is a user choosing for themselves. A business owner's question is a different one: which of the two does my customer open when looking for my service? There is no way to know that in advance, so you measure both, each by name, as in what is the difference between ChatGPT and Gemini for my business?
This guide belongs to the AI engines we measure in Arabic, and each engine has its own guide: ChatGPT in how ChatGPT picks which businesses to recommend, Perplexity in how Perplexity chooses its sources in Arabic, and Microsoft's answer inside Bing search in how Copilot chooses its sources in Bing search.
How does Gemini build its answer on Google Search?
When Gemini needs current information or a source, content from the Google Search index is put in front of the model at the moment of the question. Google calls this grounding, and its page on common crawlers says the aim is an answer that is more factual and more relevant.
Google does not spell out every step of this for the app's users, but it describes grounding with its search for developers, in a document called Grounding with Google Search, updated on 13 October 2026. The model reads the question and decides whether a search would improve the answer; if it would, it writes one or more search queries and runs them, reads the results, and writes its answer with its sources. Google says this works with all available languages.
Two consequences matter to a site owner. First, a page Google's index does not know is a page Gemini cannot find when it grounds on search, so getting your page indexed comes before everything else. Second, Gemini does not put links in every answer: Google says some responses carry no sources, and that a response without a Sources button is one Gemini gave no links for. Such an answer may mention your name, but it sends nobody to your site.
What the model writes without a source can be wrong. Google's Gemini help says it can hallucinate and present inaccurate information as factual, which is why you want the index to hold a page that states the facts about your business in plain text. What to do when an answer gets your details wrong is in how to make AI answers about your business accurate.
And how does this relate to retrieval-augmented generation?
Grounding in Google Search is one form of a broader principle called retrieval-augmented generation: the system retrieves pages related to the question, then the model writes its answer from them. The difference between what a model knows from its training and what it reads at the moment of the question is set out in what is a large language model, and how does it answer?
What does Google-Extended decide, and should you block it?
Google-Extended is not a crawler that visits your site. It is a name you write in robots.txt: Google says it sends no requests under a separate name, that crawling is done by its usual crawlers, and that the token is used for control only.
With it you decide whether what Google collects from your site may be used for two things: training future generations of Gemini models, and grounding the answers of the Gemini apps. On the same page, updated on 15 October 2026, Google says Google-Extended does not affect your site's inclusion in Google Search and is not used as a ranking signal.
So blocking it is one decision with two effects. Disallow Google-Extended and you keep your pages out of Gemini's training, and also out of what the Gemini app grounds its answers on, while they stay in Google Search as they were. AI Overviews and AI Mode are features inside Google Search, with controls of their own, covered in does blocking Google-Extended take me out of AI Overviews?
How many sites actually block it?
On 14 October 2026 we checked the robots.txt file of 150 sites the engines had cited in our measurements. Only 4 blocked Google-Extended, and each of them also blocked at least one other training crawler: for them it reads as a training decision, and its effect on Gemini's answers came with it.
Of the 71 among them that Perplexity or Claude cited in the buying questions we measure in Saudi Arabia, the UAE, Kuwait and Egypt, 2 blocked Google-Extended. Blocking it is rare among the sites engines cite, and those two remained a source for another engine.
Should you block it?
If you want the Gemini app to cite you, do not block it. If your team has a commercial or legal reason to keep your content out of model training, know before you write the rule that Google's page puts training and grounding under the same name.
At Tasadr (تصدَّر) we do not block it on our own site, because what we write is meant to be read and quoted. How to write the rules for each crawler is in how to write a robots.txt that lets them in.
What does Gemini take from Google Maps when asked about a place?
Google's Gemini help says Gemini uses public information from Google Maps to give the asker a place's address, description, website, ratings and opening hours, and directions to it. Much of that information comes from the business's Google Business Profile.
The same page says these features are being released gradually, and are available to people aged 18 or over, signed in with a personal Google Account, who ask about a place in English. It does not mention Arabic, so we cannot say that your customer asking in Arabic about a restaurant in Jeddah goes down the same route.
The work is the same either way: write on your Google Business Profile exactly what your site says, the name, the address, the hours and the link to the branch's page, as in what to write on your profile so it matches your site. If Gemini reads Maps, it finds what your site says; if it searches the web, it finds the same facts.
What should you change on your page so Gemini cites it?
Five things. The first two are Gemini's own, because it draws on Google's index; the other three help you on every engine.
- A page in Google's index: check its URL in Google Search Console. If it is not indexed, Gemini cannot find it when it grounds on search. The usual reasons are a new page Googlebot has not crawled yet, a noindex tag left over from the design stage, or a robots.txt rule that blocks it from Googlebot.
- Google-Extended not blocked, if you want the Gemini app to ground on you. Open your robots.txt yourself: some hosting and protection services offer to add rules for AI crawlers on your behalf.
- A sentence that answers first: put your customer's question as a heading, and under it one or two sentences with the name, city, figure and date. How to write it is in how do you write a paragraph AI engines will quote?
- An Arabic page, not a translation: your customer asks in Arabic, and a page written in Arabic from the start answers in their words. Write your business name in Arabic and in Latin letters, the way people write it.
- Dated facts, as text: price, hours, branches and the date of the last update, written on the page rather than inside an image. What cannot be read as text may never be grounded on.
What about structured data and llms.txt?
Google names neither as a requirement for the Gemini app, and says of its AI search features that they need no special files and no special markup. If you add structured data, make it match what your page says, and do not expect it to do what it does not, as in which structured data does your site need, and what can it not do?
What does Gemini cite for Arabic questions?
We do not publish a figure here on Gemini's sources for Arabic questions until we have read its latest measurements line by line: a figure that describes one engine needs a measurement of that engine alone, with its date and its number of questions.
Until we publish it, the quickest way to learn what Gemini cites in your field is to ask it your customers' questions and open its sources, as in the next section, then read the pages it cited instead of you and compare their order and how plainly they answer with your own page. If the difference lies there rather than in the size of the site, it is something you can change.
How can you tell whether Gemini cites you?
Ask it your customers' questions in their words, and open the Sources button on every answer. A link to your page among them is a citation; your name alone is a mention without a link. The method step by step, and what Google Search Console counts and what it does not, is in how can I tell if Gemini cites my site?
At Tasadr we measure Gemini alongside ChatGPT in the free trial, which lasts 7 days with no credit card.
On every paid plan we measure it apart from Google AI Overviews and Google AI Mode, record one of the four outcomes for every answer, and keep the excerpt of the answer that names you or your competitor, word for word as the engine wrote it, with its links and its date. So for every question you know who was cited instead of you. Plans and their limits are in Tasadr plans and prices.
What can nobody promise you?
Nobody can guarantee that Gemini cites you. On every question it decides whether to search and what to pick from what it found, Google does not document how it weighs the pages it reads, and nobody sells a place among the sources it cites.
Our Google-Extended figures cover 150 sites the engines cited in our measurements, checked on 14 October 2026, not the whole web. What we can do is measure Gemini regularly on your own questions, and show you on each one who was cited instead of you.
Sources
- Google's common crawlers · Google Search Central ·
- Grounding with Google Search · Google AI for Developers ·
- View related sources from Gemini Apps · Gemini Apps Help
- Learn about responses from Gemini Apps · Gemini Apps Help
- Find places & get directions in Gemini Apps · Gemini Apps Help
- AI features and your website · Google Search Central ·
Frequently asked questions
Does Gemini read my site?
Does blocking Google-Extended take me out of Gemini's answers?
Why does Gemini put no links in some of its answers?
Does Gemini read my Google Business Profile?
Does Gemini write about my business from what it learnt before?
Does Tasadr measure Gemini?
Related guides
- How to know if your site shows up in ChatGPT and GeminiA check by hand that tells you today whether your site leads ChatGPT and Gemini answers, how to read it, when it is not enough on its own, and an action plan for what comes next.Read the guide
- How to get cited in Google AI Overviews in Saudi ArabiaWhat Google AI Overviews is, how it differs from AI Mode, when it writes an answer to an Arabic query in Saudi Arabia, and how we measure it.Read the guide
- How does Copilot pick sources in Bing, and cite you?How Copilot chooses its sources from Bing's index, which tags keep your page out of its answers, how IndexNow gets a page into the index faster, and what the AI Performance report tells you about your citationsRead the guide
Try Tasadr free for 7 days
Add your site, write up to 10 questions your customers ask, and see who leads the answer today. No credit card needed.
Start your free trial