Which sources do ChatGPT and Perplexity cite most for B2B questions?

For B2B questions, ChatGPT and Perplexity mostly cite a small, recurring group of sources. These are reference sites like Wikipedia, community platforms like Reddit, review and comparison sites, trade media and official documentation. For software comparisons, recent “best X” lists dominate. Your own website counts, but rarely as the only voice.

You ask ChatGPT which software is best in your category. Your competitor comes first, and the sources underneath are lists and review sites you are not on. That source landscape is exactly what a GEO agency maps. It is a core part of generative engine optimization.

Which domain types keep coming back?

AI citations cluster around a handful of recognisable domain types, not around a long random list. When you put several citation analyses side by side, you see the same categories float to the top.

• Encyclopedic and reference sources. Wikipedia and comparable sites summarise facts in a structured, neutral way. AI likes to use them as an anchor for definitions.
• Community and discussion platforms. Reddit and forums carry a lot of weight for questions about experiences, pros and cons or “what actually works”.
• Independent review and comparison sites. For software choices, AI leans on platforms that place products side by side, such as G2 and Capterra.
• Trade media and industry publications. Titles that follow a market provide context and figures that AI treats as authoritative.
• Official and primary sources. Documentation, product pages and research from the source itself serve for hard specifications.

The exact share per domain varies strongly by study, by question and by moment. Semrush analysed more than 100 million AI citations over 13 weeks in 2025. Reddit and LinkedIn were in the top five of ChatGPT, Google AI Mode and Perplexity. So the stable finding is the pattern: a limited set of source types captures the lion’s share of citations. That is why publishing content at random yields little.

Why does AI trust exactly those sources?

AI cites sources it can recognise as reliable, findable and unambiguous. Three mechanisms explain the choice.

First, authority through consensus. If several independent places say the same thing about you or your category, the model treats that as a reliable signal. External mentions therefore often carry more weight than a nice sentence on your own site. We work this out in why brand mentions carry more weight than backlinks.

Second, extractability. Sources that answer a question directly, factually and in a clear structure are easier to cite than slabs of marketing text.

Third, verifiability. When a model searches the web live, it picks pages it can retrieve and check at that moment. How AI decides what it dares to cite, you can read in grounding: how AI determines what is true.

Do ChatGPT and Perplexity differ in their source choice?

Yes, and the difference lies mainly in how fresh the sources are.

Perplexity is built around live web search. For almost every question it retrieves current pages and places numbered sources underneath. Recent, crawlable pages carry a lot of weight here. How to get into that list, we cover in getting found in Perplexity.

ChatGPT mixes knowledge from the trained model with live-retrieved sources. When it searches the web, its answer includes inline citations by default, according to OpenAI’s documentation. For general questions it draws more on what the model already “knows”. Established, often-cited brands have an advantage there. For the practical side, see getting found in ChatGPT.

On both platforms, user discussion and reference works rank strikingly high, often above the classic SEO winners. Why Reddit weighs so heavily, you can read in how Reddit makes your brand visible in ChatGPT.

Which pages does AI cite most often for software comparisons?

Mostly recent “best X” lists. Ahrefs studied ChatGPT answers to 750 broad buying questions about software, products and agencies. “Best X” lists made up 43.8% of all source pages, the largest page type. Of those lists, 26% had been updated in the two months before.

For a question like “best accounting software for a Belgian SME”, you usually see these page types come back:

How to build such a page yourself is covered in comparison pages for AI citations. For review profiles, see G2 and Capterra reviews for B2B.

Your own articles can also become a source, especially on specific questions. Of 41 new articles by Hedgehog Company (younger than two months), Google showed an AI Overview on their own question 31 times. Hedgehog was a source 23 times. On the broader head term it was only 1 in 18. How that knowledge base came about, you can read in the Hedgehog case.

How do you use this without chasing loose citations?

Steer on presence in the source types that recur for your questions, not on individual mentions. How to align this with a business buying journey, you can read in GEO for B2B.

• Map your source landscape. Ask ChatGPT and Perplexity your buyers’ questions and note which sites keep surfacing.
• Make your own pages extractable. Answer one clear question per page, at the top, in plain language.
• Build external confirmation. Make sure your name, claims and figures are consistent on review, comparison and trade-media sources.
• Measure the same questions every month. Track whether you are mentioned more often and more correctly. How to set that up, you can read in measuring AI visibility.

We measure the same AI buyer prompts every month for our clients. That shows us which sources AI cites for their market and where the gaps are.

When we are not the right choice

If you sell in a niche where buyers do not yet compare through AI, this work is too early. Check that first with your last ten deals. If you mainly need customer reviews, that is work for your customer success team, not for an agency.

Frequently asked questions