How does Perplexity choose sources in Arabic and cite you?

Perplexity is an AI search engine. It searches the web when you ask, condenses what it finds into one answer, and puts numbers beside its sentences, each pointing to the page the sentence came from.

Perplexity's help centre describes three steps: it works out what you are asking, searches the internet for articles, websites and other authoritative sources, then compiles the most relevant points into a short answer. It says every answer carries numbered citations linking to the original sources, so the reader can check them.

Perplexity runs two crawlers under different names, both documented on its crawlers page. PerplexityBot reads the web to surface and link websites in Perplexity's results. Perplexity-User visits a specific page when a user's question calls for it.

The company says Perplexity-User generally ignores robots.txt, because a user asked for the page, and that neither crawler collects content to train foundation models. Both exist for search and answers, not for training.

People who search Google for «perplexity بالعربية» ("Perplexity in Arabic") want to know whether it understands them. It does, and it answers in Arabic. Its sources are another story, told in the next section.

This guide is part of AI engines in Arabic, which covers all eight engines and how each one picks its sources. What applies to ChatGPT alone is in how ChatGPT recommends your business and cites your site.

From many pages, most of them on non-Arab sites. On Arabic questions about our own field, Perplexity cited more different sites than any engine we asked, and only a small share were Arab or Gulf sites.

From our measurements in October 2026: at Tasadr (تصدَّر) we asked five engines Arabic questions, in Modern Standard Arabic and Gulf dialect, about optimising for AI answers, in three measurements on 5, 6 and 8 October that together covered 91 questions. Perplexity cited roughly three times as many different sites as ChatGPT or Gemini did on the same questions.

Only about 6% of its citations went to Arab or Gulf domains. What it cited most were Arabic translations published by foreign companies that sell answer-optimisation services, global tech sites, and Google's documentation and OpenAI's own pages.

Yet it was the engine that cited Arab and Gulf news sites most: Al Jazeera, Asharq Al-Awsat, the UAE's Al Bayan and the Saudi site Twasul. Three of those four were cited by no other engine on these questions.

The limits, plainly: these are questions about our field, not about clinics, property or online stores, and three measurements in one week do not make a rule. They are enough to say that the road into Perplexity in Arabic also runs through what others write about you, not through your own site alone.

Yes, when a Saudi page answers the question more clearly than the rest and its crawler can read it. But Arab sites took a small share in our measurements; most of its citations came from outside the region.

Our figures do not show that Perplexity prefers foreign sites. The likeliest reading is that few Arabic pages answer these questions clearly, so the engine takes what it finds, and translated pages fill the gap where pages written for Arabic readers are missing.

A study by Chen and colleagues, published in September 2025, found that GPT and Perplexity lean towards sources in the question's language more than Google does. It tested Chinese, Japanese, German, French and Spanish, not Arabic, so treat it as a pointer rather than proof of how Perplexity treats Gulf questions.

So the opening is practical. A Saudi page that answers a Saudi question in Arabic, such as «وش أفضل برنامج رواتب للشركات في الرياض؟» ("what's the best payroll software for companies in Riyadh?"), with the city, the price and the date, competes in a thin field. Section five has an illustrative example of a Riyadh payroll company.

Keep robots.txt open to it and make sure your firewall or CDN does not turn it away. Perplexity itself recommends allowing PerplexityBot in the file and accepting its requests from the IP addresses it publishes.

Perplexity says a change to the file can take up to 24 hours to apply, and that sites behind a web application firewall may need to allow its crawlers explicitly, matching both the user-agent name and the IP address, because a name alone can be spoofed.

Blocking PerplexityBot asks Perplexity not to surface your pages in its results. Perplexity-User generally ignores the file, as the company says, so do not rely on robots.txt alone to keep a page from being read.

The full detail, with the crawler names for ChatGPT, Claude and Google and a ten-minute check of your own site, is in can AI engines read your site?

A page that answers the question in its first sentence, with dated facts a reader can check, plus mentions of your business on independent sites. That makes a sentence on your page worth a number beside it.

• Answer first. Make your customer's question a heading and answer it underneath in two lines, with the name, the city and the number. Perplexity puts a source number beside a particular sentence, and a sentence that answers on its own is easier to attribute than a paragraph that circles the topic.
• Facts that survive a check. Perplexity shows every citation for the reader to open, so state the price, the timeline and the date of the last update in plain text, and the reader finds on your page the same sentence they read in the answer.
• Coverage in Gulf media. A genuine news story about your business, an interview with your manager, or a release with your figures that a business site publishes. News sites were among what Perplexity cited for our Arabic questions, and Chen and colleagues found that AI engines lean strongly towards sources independent of the brand.
• One page per real question, not per phrasing. A page that answers the question and its follow-ups in its sections beats ten near-identical pages that repeat themselves in different words.
• Nothing important behind a login or inside an image. A price inside an image, or a table that opens only after you sign in, is not text the crawler can read.

Ask it your customers' questions in their words and open the numbered sources under each answer. If your link is among them, Perplexity cited you; your name without a link is a mention, not a citation.

Repeat the same questions on different days before you judge, because one answer is a snapshot, not a trend. The step-by-step manual test is in how to check whether your site leads ChatGPT and Gemini answers, and the same method works for Perplexity.

In Google Analytics 4 you will find visits referred by perplexity.ai among your traffic sources. They show that a reader clicked a source number and reached you, not every time you were cited: many people read the answer and click nothing.

Tasadr, Arabic-first AI search competitive intelligence, measures Perplexity on every paid plan, from Starter to Business. It asks your customers' questions in Arabic and in their dialect, and records for each answer whether your site was cited, mentioned without a link or absent, and who was cited instead. The plans are on pricing.

Nobody can guarantee that Perplexity will cite you. It chooses its sources afresh for every question, sells no place among its source numbers, and does not document how it ranks the pages it finds.

Our figures come from questions about our own field in a single week, so do not read them as a rate for every sector. When we measure another sector, we publish what we find with its date. What we can do is measure regularly and show you who was cited instead of you, and why their page is clearer.