Updated on 7 November 2026, by João Leite, Founder of murmur.

To appear in Perplexity, the site needs to let PerplexityBot read the pages and bring, in the text itself, the answer the question asks for. Perplexity, the company that owns Perplexity, recommends allowing this bot in robots.txt, the file with the access rules for each bot. It also recommends accepting the bot’s IP ranges, the internet addresses it comes from. Its documentation does not list the factors that choose the sources. A 2026 study by Orbit Media shows that Perplexity cites many sources per answer and swaps part of them within two days.

  • The bot that brings the site to search is PerplexityBot. Perplexity recommends allowing it in robots.txt and accepting the IP ranges, and a change in robots.txt can take up to 24 hours.
  • The text needs to come ready in the HTML, the code that leaves the server. In December 2024, Vercel measured that PerplexityBot did not run JavaScript, which assembles parts of the page in the browser.
  • Perplexity cites many sources and swaps part of them. At Orbit Media, in 2026, there were 19.2 sources per answer, and 65% repeated two days later.
  • The sites cited change with the question. LinkedIn led in Orbit Media’s purchase questions, and Reddit and YouTube led in Ahrefs’ list of all subjects.
  • Being read is not being cited. In 2025, Ilan Strauss and colleagues saw Perplexity’s Sonar visit close to 10 relevant pages per question and cite three or four.

How Perplexity answers and shows the sources

Perplexity shows the sources as links in the answer itself, and it can open a page at the moment of the question to take information from it. That is what Perplexity’s documentation on the bots describes. The documentation on the bots and on the API, Perplexity’s service for developers, does not list the factors that decide which page enters the answer and in what order.

The Perplexity bots: PerplexityBot and Perplexity-User

Perplexity declares two AI crawlers, PerplexityBot and Perplexity-User, and each has a function. AI crawlers are programs that go across the web reading pages for the AI company. PerplexityBot brings sites to the search results, and Perplexity’s documentation says what the site needs to do to appear in them:

“To ensure your site appears in search results, we recommend allowing PerplexityBot in your site’s robots.txt file and permitting requests from our published IP ranges listed below.”

Perplexity publishes the IP ranges of PerplexityBot and of Perplexity-User in two files. According to the company, a change in robots.txt can take up to 24 hours to take effect. Also according to the company, PerplexityBot is not used to collect content for the base AI models, the large models trained on text.

Perplexity-User comes in when someone asks the question: it can visit the page to check the information and put the link in the answer. Because the request comes from the person, Perplexity writes that this bot, in general, does not follow robots.txt:

“Since a user requested the fetch, this fetcher generally ignores robots.txt rules.”

murmur, a GEO agency in Brazil that places brands in the answers of ChatGPT, Claude, Google AIO and other AI agents, counted the bots by name in a log. Google AIO is the summary that Google’s AI writes at the top of the search. The log, a record of every request to the server, was from a client subdomain, an address like conteudo.marca.com.br. From 10 to 13 June 2026, 62 of the 1,074 requests recorded came with the name PerplexityBot. The count is by user-agent, the name the program uses to identify itself, without checking the IP, which confirms the bot.

In the site’s own robots.txt, murmur allows PerplexityBot by name, in a rule that blocks no page. To build the file and test it, follow the guide on allowing search bots in robots.txt.

Where Perplexity gets the pages it cites

Perplexity documents two paths to the answer: its own search, fed by PerplexityBot, and Perplexity-User’s visits at the moment of the question. The company keeps its own search index, the collection of pages already read and stored for consultation. In September 2025, it opened this index to developers, in a search API, and described it as continuously updated. According to InfoWorld, which reported the launch, the index has billions of pages and feeds Perplexity’s answers.

Google’s weight in the pages Perplexity cites

Perplexity cites more pages from the top of Google than the other AI models measured, and yet most of them are outside that top. Ahrefs, which sells SEO tools, compared the pages cited by the AI models with the first ten Google results across 15 thousand questions. In Perplexity, 28.6% of the pages cited were in that top, in July 2025, against close to 8% in ChatGPT, Gemini and Copilot. In 2026, Orbit Media also found Perplexity the AI model closest to Google. Even there, the overlap with the first ten Google results was 13% to 23%, measured per page.

Which sites Perplexity cites: it depends on the subject and the market

Perplexity cites LinkedIn, YouTube and Reddit in external measurements, and company sites in the sector in the Brazilian niches that murmur measured. In a study published by Orbit Media, Bill Widmer repeated 72 purchase questions about three brands that sell to companies every week until August 2026. Perplexity cited 19.2 sources per answer, on average, the highest average among the four AI models in the study. The site most cited by it was LinkedIn, with 374 citations, ahead of YouTube, with 248, and Reddit. How much Reddit is worth in each AI model is in the guide on how to use Reddit and Quora to appear in AI models.

Ahrefs’ list on Perplexity, from September 2026, uses more than 3.1 million questions from the United States, on all subjects. In it, YouTube had 20.8% of the citations of the 50 most cited sites, behind only Reddit. English Wikipedia came third, and LinkedIn came 24th, with 0.9%. What Wikipedia is worth in each AI model is in the guide on Wikipedia and AI answers.

In the niche sectors of Brazil that murmur measured, what appeared most was the site of companies in the sector itself. In September 2026, we asked 25 buyer questions to Perplexity in 16 sectors, once each. Of the 372 answers we counted, 330 had a source recorded. In 84% of those 330, the site of at least one company that sells the sector’s product appeared. A blog, portal or independent comparison site appeared in 40% of them, the press in 13% and YouTube in 5%.

In Perplexity, the company sites in the sector were spread across many names: in the median of the 16 sectors, 34 different companies had their site cited. The type of each site came from a reading of what the company that owns it sells, verified by sampling. It was one round, in niche sectors, and the picture holds for them.

who measuredwhensamplewhat Perplexity cited most
Orbit Mediauntil August 202672 purchase questions about three brands that sell to companiesLinkedIn, YouTube and Reddit
AhrefsSeptember 2026more than 3.1 million questions from the United States, on all subjectsReddit, YouTube and English Wikipedia
murmurSeptember 202625 buyer questions in 16 niche sectors of Brazilcompany site in the sector, in 84% of the answers with a source

The measurements by Orbit Media, Ahrefs and murmur ask different things. You read the list for your market in the sources Perplexity shows when your buyer asks. For ChatGPT, the Brazilian overview is in the guide on which sites carry weight in ChatGPT’s answers in Brazil.

Perplexity’s Sonar read more pages than it cited in a 2025 study, and the cited page can support another company’s recommendation. Ilan Strauss, Tim O’Reilly and colleagues analyzed about 14 thousand real conversations with AI searches. Sonar, the search-enabled model Perplexity offered to developers, visited close to 10 relevant pages per question and cited three or four. The study has not yet been peer-reviewed, and Sonar gave way to the Agent API in July 2026.

To move from being read to entering Perplexity’s answer, the page needs the passage that answers the question. Being cited is also not the same as being recommended: the answer can use the company’s page as a source and recommend the competitor. That is why the company name goes in the same paragraph as the fact, the passage the answer pulls out. The guide on what makes AI recommend some companies and not others shows what is missing for the source to become a recommendation.

What to do to appear in Perplexity, in five steps

To appear in Perplexity, the order is: crawler allowed, text in the HTML, one page per question, quotations in the text and the company on LinkedIn.

  1. Allow PerplexityBot in robots.txt and the IP ranges in the firewall. The firewall is the barrier that filters requests before they reach the pages, and it can block the crawler even with robots.txt open. After the change, monitor the server log, as Perplexity asks.
  2. Deliver the text ready in the HTML. In December 2024, Vercel, a hosting company, measured that PerplexityBot did not execute JavaScript, like OpenAI’s crawlers and Anthropic’s ClaudeBot. Perplexity’s page about the crawlers does not address the point, and text that only appears with JavaScript can escape the crawler. That test was not repeated afterward, so the current check is up to you: with the page open in the browser, the Ctrl+U shortcut opens the source code, and the main sentence has to be there. The step-by-step is in the guide on text ready in the HTML.
  3. Answer one question per page, with the answer at the beginning and the company name next to the fact. Each distinct buyer question gets its own page, with the full answer, and the other ways of writing that question do not become separate pages. According to InfoWorld, Perplexity’s search API splits pages into passages and scores each passage against the question. A paragraph that opens with the answer and names the company still stands when it is excerpted. The writing is in the guide on how to write a text to appear in ChatGPT, and the full script is in the ChatGPT step by step.
  4. Include quotations from people who know, in quotation marks and with the name of the person who said it. In the study that gave GEO its name, the work of appearing in AI answers, Pranjal Aggarwal and colleagues also tested Perplexity, on 200 questions. Sentences in quotation marks increased the space of the text in the answer by 22%. The test handed the texts to the AI, and so it measures the effect with the page already in its hands. The guide on the study that created the term GEO explains the calculation.
  5. Take care of the company page on LinkedIn. Semrush sells SEO and AI monitoring tools. In its March 2026 study, 59% of Perplexity’s citations to LinkedIn were to company pages. In ChatGPT Search and AI Mode, Google’s conversational search mode, 59% were to people’s profiles. LinkedIn led in Orbit Media’s buying questions, up to August 2026, and appeared in only 5.3% of Perplexity’s answers in the Semrush study.

How to know if Perplexity already cites your company

To know if Perplexity already cites your company, ask your buyer’s questions, each one in a new conversation, and open the sources of each answer. Note whether the company’s site is among the sources, whether its name appears in the text and who appears in its place. Repeat on another day, because the list changes. In the same 72 questions repeated two days later, Perplexity kept 65% of the sources, the highest stability among the four AI models from Orbit Media.

The server log shows the requests that arrive with the name PerplexityBot. The name is easy to copy, and a request is only confirmed as PerplexityBot’s with the IP in the ranges published by Perplexity. The list of questions and the presence calculation are in the guide on how to measure brand presence in ChatGPT, and they also work for Perplexity.

The 2026 news about Perplexity’s sources: the good and the bad

The good news, from September 2026, is the stable share of the press in Perplexity’s citations, and the bad news is a CNN lawsuit. Otterly, which sells brand monitoring in AI models, tracked questions from 17 sectors in the United States every day in July and August 2026. In Perplexity, news and media were 22.7% of citations before 14 August and 23.4% after. In ChatGPT, the share fell from 21.3% to 10.4%, and Otterly writes that no other of the seven AI models did the same. What the press is worth in each AI model is in the guide on press relations to appear in ChatGPT.

CNN sued Perplexity in May 2026, in a federal court in New York, over copyright and trademark. According to SiliconANGLE, CNN alleges that the reader is confused about CNN’s connection to the articles Perplexity delivers. Perplexity responded, through communications director Jesse Dwyer, that facts are not copyrightable. The lawsuit is about news reports used in the answers, and news and media add up to almost a quarter of Perplexity’s citations in Otterly’s count. At the outset, these were CNN’s allegations, with no ruling from the judge.

At murmur, the work is the page that Perplexity and the other AI models read

At murmur, the work is the page: the guide that answers the questions of the client’s market, on a subdomain of the client’s brand. Before writing, a dedicated person runs the brand’s diagnostic in the AI models and studies the company and the sector. They see where the client is missing from the answers and who occupies that place, and they take care of the content that fills those gaps. The pages go out on the open web and work for any AI that reads the web, Perplexity included. They stay with the client after the contract ends.

The guide for FotoRegistro, a photo printing e-commerce company, had its first part published on 24 June 2026. The guide’s first citation in ChatGPT came on 24 July, 30 days later. That date is from the case study, not a timeframe. On average across murmur’s clients, significant improvement came from 60 days onward, depending on the starting point and the competition.

murmur’s other case studies include numbers from ChatGPT, Claude, Google AIO and Perplexity. At Mindo, a motion design studio in São Paulo, ChatGPT’s recommendation rate went from 12% to 23% across the 134 questions. The measurements are from 25 and 26 June and from 13 and 14 July 2026. In Perplexity, Fly Vet appeared in 79% of the 258 questions without its name on 2 November 2026, against 27% on 22 July.

At Visio, operations software for retail, murmur measured the same 135 questions in June and in July 2026. The brand went from 0% to 45% of Google AIO answers, from 2% to 22% in ChatGPT and from 1% to 7% in Claude.

murmur monitors ChatGPT, Claude, Perplexity and Google’s AI Overview, which account for most of the questions. It builds and maintains the client’s brand subdomain, without touching the main site. Allowing PerplexityBot and the IP ranges on the main site goes to the client’s team as a suggestion.

Frequently asked questions

Can hosting or the firewall be blocking Perplexity without me knowing?

They can, and without warning: hosting bot protection and firewall rules block PerplexityBot before it reaches the pages. Perplexity writes that a site with a web application firewall, the WAF, may need to explicitly allow its bots. The documentation includes a walkthrough for Cloudflare and AWS WAF, with a rule that checks the user-agent and the IP together. On 15 September 2026, Cloudflare, which sells this control, retired “Block AI Bots” and separated the search bot from the training bot: fewer than 1% of its sites block search, and 17% block training. Ask the person who manages the hosting whether the search bot is allowed, and check the server log to see whether PerplexityBot receives the page. How to find out about a block is covered in the guide about hosting that blocks AI bots.

Does Perplexity respect robots.txt?

According to Perplexity’s documentation, robots.txt controls PerplexityBot, and Perplexity-User generally does not follow the file, because it acts at the request of whoever asked. In August 2025, Cloudflare, which sells network and security for sites, said it had seen Perplexity use an unidentified bot. According to Cloudflare, this bot presented itself as Chrome on a Mac and made 3 to 6 million requests per day. In a test, Cloudflare created new sites with a robots.txt that prohibited everything and saw Perplexity answer with their content.

How do I see visits coming from Perplexity in Google Analytics?

In Google Analytics, visits coming from Perplexity appear with the source perplexity.ai, and come from people who clicked a source in the answer. What the click does not show appears in the brand’s presence in the answers, which murmur tracks. Presence is how many answers the brand appears in, divided by the total number of answers. The step-by-step instructions for Analytics are in the guide about traffic coming from AI in Google Analytics.

Is it worth appearing in Perplexity?

It is worth appearing in Perplexity when your buyer uses it, and the work is almost the same as for the other AI models. In Brazil, Vivo closed a partnership with Perplexity in November 2024 to give Perplexity Pro, the paid version, to Vivo Valoriza customers. In October 2025, the offer reached prepaid, with six months for anyone who tops up R$ 30 or more.

Does appearing in Perplexity help you appear in ChatGPT and Claude?

Appearing in Perplexity helps with the other AI models, because the same page serves every model that reads the web. The foundation is the same: text in the HTML, the answer at the beginning and one page for each genuinely different buyer question. Each AI model builds its own list of sources. In two of the three question sets from Orbit Media, one per brand, ChatGPT and Perplexity shared only 7.6% and 5.4% of the sites cited. That is why presence is checked in each AI model, with the same buyer questions. How ChatGPT and Claude reach sources is in the guide about how to get your company cited by AI models.

See how Perplexity answers about your market

To see how Perplexity answers about your market, use murmur’s free diagnostic. You enter the site and see, without signing up and in less than two minutes, how Perplexity and other AI models answer your buyer. murmur’s sector studies show which companies and sources appeared in one round of questions for each market, and each study covers only that round.

Read also

References