How does Perplexity work and what does it do differently from ChatGPT

How does Perplexity work when someone asks it a question: it searches its own index of web pages and the live web, picks the passages that best answer the question and writes an answer with numbered citations linking to the sources. The first difference from ChatGPT is the index, since Perplexity runs its own while ChatGPT Search pairs its own index with third-party indexes, Perplexity wrote on 13 November 2026.

TL;DR

Perplexity answers a question by searching its own web index and the live web, scoring individual passages and writing an answer with numbered citations to its sources.
Perplexity’s Decider model judges pages while Perplexity crawls them and filters anti-bot walls, CAPTCHAs, error pages, spam, SEO filler and navigation-only pages out of the index.
Perplexity says ChatGPT Search pairs its own index with third-party indexes, while Perplexity runs its own index and names no crawler that collects content for training AI models.

How Perplexity answers a question

Perplexity answers in five steps: it decides whether to search, splits the question into searches, retrieves pages, scores passages and writes a cited answer. A multi-head classification system processes each user query, and one head decides whether the query should trigger web search, according to Perplexity’s Decider post of 12 November 2026.

Perplexity’s 13 November 2026 post on search engine types says a large language model then reads the question and may split it into several searches, a process called query fan-out. The searches run against Perplexity’s own index, and Perplexity can also open a page live while answering.

Perplexity scores parts of pages. Its Search API announcement of 25 September 2025 says the index divides documents into fine-grained units that are each scored against the query. Perplexity’s Photon post of 24 September 2026 adds that lexical retrieval begins with the query’s terms and finds documents that contain them, so a passage written in the buyer’s words is easier for Perplexity to find. The path from question to answer runs like this:

Figure 1. Perplexity decides whether to search, fans out the question, retrieves pages, scores passages and writes a cited answer.

A brand’s page can be cited in the last step only if one of its passages was retrieved and scored in the two steps before it. How Photon ranks those candidates, and where a CDN blocks Perplexity’s bots, is covered in Perplexity ranks pages in its own index, and the CDN decides access.

What Perplexity filters from its index

Perplexity removes low-quality pages from its index while it crawls them, using Decider, a classification model it released on 12 November 2026. The Decider post names two kinds: pages that did not load properly for the crawler, such as anti-bot walls, CAPTCHAs or error pages, and pages that loaded but carry no helpful information, such as spam, SEO filler and navigation-only pages. Perplexity’s Decider filter for SEO filler pages covers the release. The filter sorts pages this way:

Figure 2. Decider removes anti-bot walls, CAPTCHAs, error pages, spam, SEO filler and navigation-only pages from Perplexity’s index.

Four of the five rows end outside the index, and the fifth shows Decider also classifying each page’s domain, intended audience and any harmful, illegal or personally identifiable information in the same pass. Perplexity says Decider does this considerably faster and cheaper than an LLM judge. The decision is made at crawl time, before any question is asked, so a filtered page is missing from Perplexity’s index for every question.

A site whose CDN answers PerplexityBot with a challenge page hands Perplexity’s crawler an anti-bot wall, one of the page types the filter removes. A page that only lists links to other pages fits the navigation-only type, and a page padded with text written for search engines fits the SEO filler type.

Perplexity sources and citations

Perplexity shows its sources as numbered citations in each answer, linking to the original pages so a reader can check each claim, Perplexity’s help center says in a page last modified on 12 November 2026. In an example in Perplexity’s 13 November 2026 post, a question about moving a teenager’s learner’s permit from Illinois to Arizona drew 15 sources.

Some Perplexity citations carry a shield icon with one of three labels, Government, Academic or Trusted, according to Perplexity’s help page on source labels, last modified on 6 October 2026. Perplexity rates the whole website a source comes from, checking questions such as whether the site corrects its mistakes, says who wrote each piece and keeps news separate from advertising and opinion.

Perplexity gives the Trusted label to sites that appear often enough in results and publish within their own area of expertise, with reuters.com and mayoclinic.org as its examples. Most domains carry no label, and Perplexity’s help page says a missing label means the domain has not been rated yet and is not a negative judgment.

Perplexity vs ChatGPT

Perplexity vs ChatGPT comes down to the index, the crawlers and the citations. Perplexity runs its own index, while ChatGPT Search uses its own index paired with third-party indexes, according to Perplexity’s 13 November 2026 post on search engine types. OpenAI’s side is set out in where ChatGPT gets its information when it searches the web.

Perplexity’s crawler documentation names PerplexityBot, which surfaces and links websites in Perplexity’s search results and is not used to crawl content for AI foundation models, and Perplexity-User, which may visit a page when someone asks a question and generally ignores robots.txt rules. OpenAI’s crawler documentation lists OAI-SearchBot for search, ChatGPT-User for user actions and GPTBot for training. A robots.txt change takes up to 24 hours to apply at Perplexity and about 24 hours at OpenAI.

OpenAI’s web search documentation for developers says inline citations show only the most relevant references, and the number of sources the model consulted is often greater than the number of citations. Perplexity’s help center says each Perplexity answer includes numbered citations. The two engines compare this way:

Figure 3. Perplexity runs its own index and no training crawler, while ChatGPT pairs indexes and OpenAI runs GPTBot for training.

The index and the training crawler are where Perplexity and ChatGPT differ most, and OpenAI’s documentation says a site can allow OAI-SearchBot to appear in search results while disallowing GPTBot.

Perplexity and ChatGPT also cite different websites. Reddit is the single most cited domain in Agenzy’s 2026 AI source study, taking 2.1% of 2,283,923 citations and appearing in 92.8% of 138 business projects, with ChatGPT citing it 5.0x more heavily than Perplexity. A brand’s position on ChatGPT says little about its position on Perplexity, so each engine is checked on its own.

How to rank in Perplexity

A page has to be reachable before Perplexity can cite it, and then it has to answer the buyer’s question in the buyer’s words. PerplexityBot has to reach the page as content, since Perplexity’s crawl-time filter removes anti-bot walls and CAPTCHAs. Perplexity’s crawler documentation asks sites to allow PerplexityBot in robots.txt and to permit requests from the IP addresses it publishes, and the firewall settings for each bot are in the AI crawlers list.

Perplexity’s 13 November 2026 post says pages need the same SEO foundation to get cited in AI answers, and it recommends sections that each answer one specific question clearly, even when pulled out of context. Perplexity scores passages and may split one question into several searches, so each section of a service or comparison page should open with one buyer question in the buyer’s words and answer it in the first sentence.

Mentions on other websites come next, because AI engines judge a brand by how many trusted sources mention it and how widely, and give little weight to what the brand says about itself. For Perplexity, that points to news, trade and reference sites that publish within their own field, the kind of site Perplexity describes when it explains its Trusted label.

What it means

The head of marketing should have the website team and the content lead work through Perplexity in this order:

Allow PerplexityBot and Perplexity-User in robots.txt and in the CDN or firewall, then confirm in the server logs that both bots receive pages instead of challenge screens.
Add useful content to pages that only list links, and fix pages that fail to load, since Perplexity’s crawl filter drops both kinds.
Rewrite service and comparison pages so each section opens with a buyer question in the buyer’s words and answers it in one or two sentences.
Pitch the news and trade sites in your field with something they can report.
Run the buyer questions through Perplexity and ChatGPT separately each month, and record which pages each engine cites.

FAQ

Which AI models does Perplexity use?

Perplexity uses several large language models, and its help center names GPT-6.1 Sol and Claude Sonnet 5.5 among them as of 12 November 2026. Perplexity Pro subscribers can choose a model from the model menu, or leave the setting on Best, which picks a suitable model for each question.

How fresh is Perplexity’s index?

Perplexity’s systems process tens of thousands of index update requests each second, the company said in its Search API announcement of 25 September 2025. The same infrastructure powers Perplexity’s public answer engine, and Perplexity calls information staleness one of the biggest failure modes for AI agents.

What is Pro Search in Perplexity?

Pro Search is Perplexity’s deeper search mode, which researches across multiple sources and returns longer, organized answers. Perplexity’s help center says Pro Search applies automatically when a user’s plan includes it, and free users get a limited number of Pro Searches per day.

How long does a Perplexity Deep Research report take?

A Deep Research report takes 2 to 4 minutes, according to Perplexity’s help center, which says the feature performs dozens of searches and reads hundreds of sources. Perplexity’s 13 November 2026 post on search engine types adds that the finished report can be exported as a PDF or a document.

Does paying Perplexity earn a site a Trusted label?

No. Perplexity’s help page on source labels says partnerships, payments and other business arrangements do not affect a site’s label, which the source review process sets alone. Labels can change as sites are reassessed, and a website owner can ask about a label through Help, then Contact Support.

Does Perplexity cite the same websites as ChatGPT?

Often not. Each AI engine answers from its own set of sources, so a brand cited by Perplexity can be missing from ChatGPT’s answers to the same question. The websites ChatGPT, Google AI Overviews and Perplexity cite most differ by engine, with news and media taking a larger share on Perplexity.

How do I check whether Perplexity cites my site?

Ask Perplexity the questions your buyers ask, without your brand name, and open the numbered citations under each answer to see which pages it used. Repeat the same set every week, as a prompt set for AI visibility tracking describes, and read the server logs for PerplexityBot and Perplexity-User visits.

Is Perplexity the same kind of product as Google AI Mode?

Yes, by Perplexity’s own grouping. Its 13 November 2026 post lists Perplexity and Google AI Mode as AI answer engines that write an answer with citations from their own index or a licensed one. Google’s version answers inside Google Search, as how Google AI Mode differs from the Gemini app and AI Overviews explains.

Can Perplexity get facts about a brand wrong?

Yes. Perplexity’s 13 November 2026 post warns that large language models may hallucinate when they write the answer, and tells readers to open the cited pages and check the claims. A wrong price or an old name in an answer costs deals, so a brand that finds one should open the cited pages and correct the page that carries it.

How do I know whether Perplexity brings in customers?

Ask every new customer how they found the company, since analytics undercounts buyers who met a brand in an AI answer and arrived later through Google or a direct visit. How to measure AI search revenue when analytics credits Google sets out the method.

How do I get cited in ChatGPT as well?

Begin with Bing and OpenAI’s search crawler, since ChatGPT search finds candidate pages through Bing and OpenAI’s crawler documentation recommends allowing OAI-SearchBot in robots.txt. Pages written in the buyer’s words help on both engines, and the rest of the method is in how to rank in ChatGPT answers and get your brand cited.