Where LLMs pull their B2B citations from in 2026

TL;DR

We ran 2,400 buyer-style B2B queries, 800 each on ChatGPT, Claude, and Perplexity, between January 8 and February 21, 2026. We classified every citation in every response by source type. Three findings worth your attention:

• 65% of B2B citations come from just five source types. Vendor product pages, review aggregators (G2, Capterra, TrustRadius), Reddit, independent analyst-style blogs, and the platform’s own documentation.
• The cite-graph diverges sharply by platform. Perplexity is Reddit-heavy. Claude leans on primary documentation. ChatGPT spreads more evenly but underweights Reddit. If you optimize for one without checking the others, you’re invisible on two-thirds of buyer questions.
• The long tail is shorter than expected. The 50th-percentile source appears in only 0.3% of responses. Most “citations” are concentrated on a small, identifiable source set, which means your optimization target is more tractable than it looks.

Methodology

Query set. 80 query patterns covering 14 B2B categories (DevOps, HR tech, sales tools, security, observability, FinTech back-office, marketing automation, customer success, analytics, dev productivity, data infrastructure, RevOps, compliance, AI infrastructure). Each pattern run as 30 variants, comparison queries, alternative queries, workflow queries, skeptic queries, and category fit queries. Total: 2,400 queries.

Sample period. January 8 – February 21, 2026.

Platforms. ChatGPT (GPT-4-class model, web-enabled), Claude (Sonnet-class model, web-enabled), Perplexity (default settings, public mode).

Classification. Every citation in every response was logged with its domain, then classified by source type using an 11-category taxonomy (vendor self-page, review aggregator, Reddit, analyst blog, newsletter, primary research, documentation, news, partner content, academic, government). Disagreements (n=87) were resolved by a second pass.

What we excluded. Hallucinated citations (citations the LLM described but couldn’t actually link to). These accounted for 4.1% of all citations and were dropped from the corpus.

Source-type distribution (aggregate across all three platforms)

Source type	Share of all citations
Review aggregators	19.4%
Vendor self-pages	17.1%
Reddit	14.6%
Analyst-style blogs	8.9%
Documentation	4.8%
Newsletters	3.7%
Primary research	3.1%
News	2.4%
Partner content	1.9%
Academic	0.4%
Government	0.1%
Other (long-tail)	23.6%

The top 5 categories account for 64.8% of all citations. The long-tail “other” bucket contains 1,847 unique domains, none of which appeared in more than 0.4% of responses.

Where the platforms diverge

The aggregate masks the per-platform behavior. Three patterns stand out:

Perplexity is Reddit-heavy.

Reddit accounts for 22.1% of Perplexity citations, vs. 14.6% in the aggregate. The skew is even sharper on workflow and skeptic queries, Reddit’s share rises to 31% for “is X worth it” patterns.

If your category has an active subreddit and you’re not present there, Perplexity is treating your competitors as the only side of the conversation.

Claude leans on documentation and primary sources.

Documentation sites (your own docs, your competitors’ docs, neutral how-to resources) account for 11.2% of Claude citations vs. 4.8% in the aggregate. Primary research (whitepapers, original studies) hits 6.8% on Claude vs. 3.1% aggregate.

Claude is the platform where original research most reliably converts to citation. If you have proprietary data, Claude is where publishing it pays back fastest.

ChatGPT spreads more evenly.

ChatGPT’s distribution most closely matches the aggregate, with no single source type above 19%. This makes ChatGPT the “median” target for B2B AEO programs, optimize for the aggregate distribution and ChatGPT performance follows.

Implications for your AEO program

1. The optimization target is smaller than people think.

You don’t need to be present on 1,000 sources. You need to be present, credibly, on 50-80 of the right ones. That’s a tractable program for a 2-3 person team.

2. Reddit isn’t optional.

If you’re not on Reddit, you’re missing roughly 15% of all B2B citations and roughly 22% of Perplexity-specific ones. The cost of serious Reddit presence is one senior team member, two days a week, for 6 months. There’s no cheaper way to recover that share.

3. Primary research pays back disproportionately on Claude.

If you have or can commission original research, publish it. Even a small (n=80) study, framed properly, will show up as a Claude citation across multiple adjacent queries.

4. Review aggregator presence is foundational.

G2, Capterra, TrustRadius, whichever your category lives on. If you’re absent there, you’re invisible to 19% of all citations across all three platforms. This is the single highest-leverage source-type investment for a vendor starting from zero.

What we’ll measure next

Three things we’re tracking for a follow-up study in summer 2026:

• Whether the Reddit weight holds. Reddit citations have grown 3.2× year-over-year. We’re skeptical the curve continues; we’re watching for a plateau or correction.
• The impact of citation-link presence on click-through. Our initial pass suggests cited buyers convert at 2-4× the rate of organic-search buyers. We’re instrumenting a deeper attribution study with three engagement clients.
• Cross-platform citation overlap. If you appear on ChatGPT, how likely are you to appear on Claude and Perplexity for the same query? Initial pass: about 38% overlap. That’s lower than expected and suggests platform-specific work is required, not just a single optimization that flows across all three.

If you’d like the per-category breakdown or to discuss the implications for your specific vendor profile, book a call.