SEO & Marketing

How ChatGPT and Perplexity Choose Sources and How to Become One

AI answers cite only a few sources. How ChatGPT and Perplexity find and choose them, and what your website needs for it.

of Josef KujawskiPublished: 21 October 2026Updated: 13 October 20266 min read1.072 Words

How ChatGPT and Perplexity Choose Sources and How to Become One

Short answer

ChatGPT and Perplexity answer current questions by searching the web live, retrieving fitting pages and building an answer with source links. They choose pages that are accessible and crawlable, answer the question directly and clearly, and appear trustworthy. So it matters not to block the AI providers' crawlers, to write content in clear sections with direct answers, and to be mentioned as an expert on other sites too. There is no guarantee of being cited.

Few spots, big impact

A Google results page shows ten organic results. An answer from ChatGPT or Perplexity often cites only a handful of sources. Those who appear there are perceived as a recommendation. Those who are missing do not exist in that moment.

In our foundational article on GEO optimisation we explained what Generative Engine Optimization is. Here we go one step deeper: how do these systems get their sources in the first place?

How an AI answer with sources is created

Process: question, web search, reading pages, answer with sources — Simplified flow of an AI answer with web search.

A question is asked. The system decides whether it needs current information from the web.
It searches the web. ChatGPT search uses search partners and its own crawlers, Perplexity searches the web for every query. ^[3] ^[2]
It reads the pages found and looks for passages that answer the question.
It writes an answer and links the pages it relies on.

At Google, AI Overviews run on the same foundation as regular search: a page must be indexed and eligible for snippets. ^[4]

Know the crawlers

No access, no source. The most important bots and their purpose:

Crawler	Provider	Purpose
OAI-SearchBot	OpenAI	Capture pages for search in ChatGPT
ChatGPT-User	OpenAI	Fetch pages when users ask for them
GPTBot	OpenAI	Collect content for model training
PerplexityBot	Perplexity	Index pages for search results
Googlebot	Google	Search and AI Overviews

OpenAI documents that search and training crawlers can be controlled separately. ^[1] So you can appear in ChatGPT answers without releasing content for training. Check your robots.txt and firewall rules, as some hosting and security services block AI bots by default.

Kernaussage

Before thinking about content, check access. A page that OAI-SearchBot or PerplexityBot may not read cannot be cited there.

What makes a page citable

Three layers: content, technology, mentions — Citability comes from content, technology and mentions.

The providers do not disclose exactly how they weight sources. But their documentation and practice show clear patterns:

Direct answers: the key statement comes at the start of a section, not after three paragraphs of introduction.
Clear structure: one question per section, meaningful subheadings, lists and tables.
Concrete facts: figures, definitions, steps and cited sources instead of generic statements.
Visible expertise: author, company and experience are visible.
Freshness: maintained pages with dates are more likely to be used for time-sensitive questions.
Mentions elsewhere: those named in trade articles, directories and forums are more likely to appear.
Technical readability: content is present in the HTML, not only visible after heavy JavaScript. More in Technical SEO.

AI search rewards the same virtue as a good conversation: answer the question instead of talking around it.

– Josef Kujawski, Managing Director Maluure GmbH

What does not help

Keyword repetition: language models understand meaning, not frequency.
Hidden text for bots: risky and against search engine guidelines.
Special "AI files" as a cure-all: they do not replace accessible, good content.

Measure whether it works

Check referrals from chatgpt.com and perplexity.ai in your web analytics. Add spot checks: regularly ask the questions your customers would ask and note which sources are cited. Our own experience from the launch phase of our website is described in the article SEO for Lovable.

Which sources does ChatGPT use?

A distinction that often gets mixed up helps here:

Mode	Where the knowledge comes from	Sources visible?
Answer from training knowledge	Data the model was trained on	No, no links
Answer with web search	Pages retrieved live from the web	Yes, as links in the answer
Answer from uploaded files	Documents users provide themselves	Yes, reference to the file

For your visibility, the second mode is what matters most. Only there can ChatGPT link your page as a source. You have little direct influence on training knowledge, except that good, frequently cited content is more likely to flow into future models if you allow training crawlers.

How do I get cited as a source by ChatGPT?

There is no guarantee. But these steps increase the likelihood significantly:

Check access: robots.txt and firewall allow OAI-SearchBot and PerplexityBot.
Ensure indexing: the page is indexed by Google and Bing, as many AI systems rely on search indexes.
Identify questions: which concrete questions does your audience ask? Each deserves its own clear section.
Answer directly: the first sentence under each heading fully answers the question.
Provide evidence: figures, sources, your own experience, an author with expertise.
Build mentions: trade articles, interviews, directories and partnerships make sure your name appears in connection with your topic on other sites too.
Keep it current: review content regularly and show the update date.

Kernaussage

You cannot control whether an AI cites you. But you can control whether it finds you, understands you and trusts you.

Checking your website step by step

Open robots.txt at yourdomain.com/robots.txt and look for entries for GPTBot, OAI-SearchBot, ChatGPT-User and PerplexityBot.
Check security services: some bot protection settings at your host or CDN block AI crawlers without it showing in robots.txt.
Ask test questions: write five to ten questions your customers would ask, for example "Which agency in Cologne builds websites with Lovable?", and regularly check which sources are cited.
Analyse web analytics: watch referrals from chatgpt.com, perplexity.ai and similar services as a separate source.
Use Search Console: visibility in Google, including AI Overviews, is reported there together with regular search.

Example: how content becomes citable

A typical paragraph as often written: "In today's digital world it is more important than ever to deal with load time, because many factors play a role here." This sentence contains no answer.

Written to be citable: "A website should load its largest visible element in no more than 2.5 seconds. That is Google's threshold for a good Largest Contentful Paint." Concrete, verifiable, direct. AI systems like to take exactly such sentences into their answers. We explain this threshold in Core Web Vitals.

If a paragraph is understandable and correct even without the rest of the article, it has a good chance of ending up in an AI answer.

– Josef Kujawski, Managing Director Maluure GmbH

Conclusion

ChatGPT and Perplexity choose sources they can reach, understand and consider trustworthy. Give search crawlers access, answer questions directly and make sure others talk about you too.

Frequently asked questions

How does ChatGPT choose its sources?+

For questions that need current information, ChatGPT searches the web, retrieves fitting pages and links the sources used. OpenAI does not disclose exactly how selection is weighted.

How does Perplexity find sources?+

Perplexity searches the web for every query, evaluates the pages found and shows the sources numbered next to the answer. It uses its own crawlers for this.

Do I need to allow AI crawlers on my website?+

If you want to appear in AI answers, do not block search crawlers like OAI-SearchBot and PerplexityBot in robots.txt. You can allow or block training crawlers separately.

What is the difference between GPTBot and OAI-SearchBot?+

GPTBot collects content for model training, OAI-SearchBot is used to show pages in ChatGPT search results. Both can be controlled separately in robots.txt.

Which content do AI search engines prefer to cite?+

Content that answers a question directly, is clearly structured, contains concrete facts or figures and comes from a recognisably expert source.

Does classic SEO help for ChatGPT and Perplexity too?+

Yes. Crawlability, clean technology, good content and mentions on other sites are the foundation for AI search as well. GEO builds on SEO but does not replace it.

Can I guarantee being mentioned in AI answers?+

No. Selection is not transparent and changes constantly. You can only increase the likelihood by getting the fundamentals right.

How do I measure whether I appear in AI answers?+

Through referrals from chatgpt.com or perplexity.ai in your web analytics and through regular spot checks with typical questions from your audience.

Which sources does ChatGPT use?+

Without web search, ChatGPT answers from its training knowledge, without links. With web search it retrieves current pages and links them as sources. Only in this mode can your website appear as a source.

How do I get cited as a source by ChatGPT?+

Allow search crawlers, ensure indexing, answer your audience's questions directly and with evidence, build mentions on other sites and keep content current. There is no guarantee.

Sources

Overview of OpenAI Crawlers, OpenAI (2025)
Perplexity Crawlers, Perplexity (2025)
Introducing ChatGPT search, OpenAI (2024)
KI-Funktionen und deine Website, Google Search Central (2025)

#GEO#AI Search#ChatGPT#Perplexity#SEO

Author

Josef Kujawski

Managing Director & Creative Director at Maluure

Josef leads the strategic and creative development of brands at Maluure. For over a decade, he has guided medium-sized businesses from brand positioning to digital delivery.

Managing Director of Maluure GmbH, Cologne. Specialising in corporate design, brand strategy and digital brand experiences.

LinkedIn Website