Perplexity AI SEO: how Perplexity, ChatGPT and Gemini pick their sources

Perplexity AI SEO, in short

Perplexity AI SEO is mostly two jobs. Let PerplexityBot read your site, and be the clearest page on the questions your buyers ask.

Perplexity is an answer engine. It searches the web for a question, reads some of the pages it finds, writes an answer and links the pages it used. You cannot buy a place in that list or submit a page to it. You can only make sure Perplexity is allowed to read you, and give it a page worth citing.

This guide puts Perplexity beside ChatGPT and Gemini, because a buyer rarely uses only one. The three run different robots and different search, so the access step differs for each. The content step is much the same.

How the three pick sources, side by side

Each assistant has one robot that decides whether you can be cited, and a separate switch for training. Mixing the two up is the most common way a site blocks itself by accident.

The OpenAI row comes from its crawler page, which says sites opted out of OAI-SearchBot “will not be shown in ChatGPT search answers, though can still appear as navigational links”. The Gemini row comes from Google’s common crawlers page and its page on AI features. The Perplexity row is covered in the next section.

Our Perplexity page quotes its crawler documentation in full, and the engines page lists the same robots for ChatGPT, Gemini and DeepSeek.

What Perplexity’s own pages say

Perplexity documents two robots and recommends allowing the one that builds its search results. It says that robot is not used to train AI models.

Perplexity’s crawler page says PerplexityBot “is designed to surface and link websites in search results on Perplexity”, and “is not used to crawl content for AI foundation models”. It recommends “allowing PerplexityBot in your site’s robots.txt file and permitting requests from our published IP ranges”. It says changes can take up to 24 hours to reach its systems.

The second robot, Perplexity-User, visits a page when a person’s question needs it. Perplexity says it “might visit a web page to help provide an accurate answer and include a link to the page in its response”, and that because a user asked, it “generally ignores robots.txt rules”.

The same page has a long section on firewalls, with steps for Cloudflare and AWS. That is a hint worth taking. A firewall rule that blocks unfamiliar robots can undo an allow in robots.txt, and you will not see it in robots.txt at all. If you use a firewall or a bot filter, check it lets Perplexity’s published addresses through.

Perplexity’s developer Search API page describes its search as returning “real-time, ranked web results”. Perplexity does not publish how it ranks those results or which ones its answer cites. Nobody outside Perplexity knows the weights, and a guide that gives them to you is guessing.

What research says about how answer engines choose

Two studies are worth reading. One found that adding sources, quotes and statistics to a page made it more visible in AI answers, while stuffing keywords did not. The other found that AI search leans hard towards third-party sources over a brand’s own site.

The first is GEO: Generative Engine Optimization by Aggarwal and others, published at KDD 2024. The authors built a test answer engine that read the top five Google results and wrote an answer with a GPT-3.5 model. They rewrote pages in nine ways and measured how much of each answer came from the rewritten page. They also repeated the tests on Perplexity itself.

The second is Generative Engine Optimization: How to Dominate AI Search by Chen and others, a September 2025 preprint, which means it has not been peer reviewed. It compared AI search with Google across many topics, languages and rewordings of the same question.

Read together, the studies say the same thing from two sides. Your own page should be the most checkable answer on the question. And you also need to be named on the pages the assistants trust more than yours, such as reviews, comparisons and lists written by someone else.

Where Perplexity differs from ChatGPT and Gemini

The biggest practical difference is how visible the sources are. Perplexity is built around them, while Gemini says plainly that some answers show no sources at all.

Google’s help page on sources in Gemini Apps says “not all responses include related links or sources”. OpenAI’s crawler page ties the sites shown in ChatGPT search answers to OAI-SearchBot. Perplexity is built around search, and its developer overview describes “web-grounded answers with built-in citations”.

That makes Perplexity a good place to learn from. When a rival is cited and you are not, open the cited page and see what it says that yours does not. With any of the three, you can do this only when the answer shows its sources.

The second difference is the training question. Perplexity says its search robot does not crawl for AI model training, so allowing it is a decision about search only. With OpenAI and Google, search and training are separate switches, and you set each one on purpose.

Allowing an assistant’s search robot is about being cited. Allowing its training robot is a separate decision. Set each one on purpose.

A Perplexity SEO checklist

Work through these in order. The first three are about access and take minutes, and the rest are about the page and take longer.

1. Open your robots.txt and make sure PerplexityBot, OAI-SearchBot and Googlebot are not disallowed, by name or by a rule for all robots.
2. Check your firewall or bot filter. Let Perplexity’s published addresses through, as its crawler page describes.
3. Decide about GPTBot and Google-Extended separately. They are about training and Gemini grounding, not about Perplexity.
4. List the questions a buyer asks before they buy, in their own words, and ask each one in Perplexity.
5. Open every page Perplexity cites for a question you lost. Note what it says that your page does not.
6. Rewrite your page so the answer sits near the top, in plain sentences, with a source or a number behind each claim.
7. Ask to be included on the third-party pages Perplexity cites, where that is fair. A review or a list someone else wrote is earned media.
8. Ask the same questions again after a week or two, and compare with what you saved before.

How to check where you stand

Ask the questions and keep the answers. Perplexity gives you the source list, so the evidence is on the screen.

Do it by hand first. Ten buyer questions in Perplexity, ChatGPT and Gemini is thirty answers to read, and you will learn more from reading them than from any score. Save each answer whole, with its links, and the date. Ask again later, because the same question can get a different answer on a different day. Our method page explains why one answer should never be read as a ranking.

A worked example

The company and the numbers in this example are made up to show the method. They are not a measurement of any real business.

Say you sell scheduling software for dental clinics. You write ten questions a clinic manager might ask, such as which booking tool works with a given practice system. You ask all ten in Perplexity and save the answers. You are named in two of the ten. One rival is named in seven.

You open every page Perplexity cited in the seven answers that named the rival. Four of them are the same two review roundups, and you are on neither. Two are the rival’s own integration pages, which list each practice system by name with a setup time. One is a forum thread. Your own integration page says “works with all major systems” and names none of them.

Now there is a plan, in order. Rewrite the integration page to name each system and say how long setup takes. Write to the two roundup authors with the facts they would need to judge you fairly. Then ask the same ten questions again in two weeks and compare. If the count moves, look at which answers changed and which pages they cited, before deciding it was your change that did it.

Agentcites does this on a schedule. Today it asks Perplexity, ChatGPT, Gemini and DeepSeek up to 50 buyer questions for each website every week. It keeps every answer word for word with the pages it cited, and the citation tracking page shows how it matches those links to your site and your rivals’. It costs $149 a month per website.

What this comparison does not show

This guide compares what each provider documents, not how each one ranks. None of the three publishes its ranking, so the table above is about access and visibility, not about which page wins.

It also does not cover paid placement, the Perplexity publisher program, or Perplexity’s shopping features. Each of those is a separate topic with its own rules, and none of them was read for this guide.